Linguistic Linked Data for Content Analytics: a Roadmap

Executive Summary

As data is being created at an ever-increasing pace, more and more organisations will exploit insights generated from the data to optimize existing processes, improve decision making and to modify existing or even generate radically new business models.

Data has been in fact regarded as the oil of the new economy. Consequently, data linking and content analytics are the key technologies to refine this oil so that it can actually drive the motor of many applications. Refinement consists of: i) homogeneization, ii) linking, iii) semantic analysis, and iv) repurposing.

Homogeneization consists in making sure that the data is described using agreed terminologies and standardized vocabularies or ontologies in machine readable formats. Thus, data becomes interoperable and easily exploitable due to the semantic normalization that avoids conceptual mismatches. Refinement also includes linking data across datasets and sites (resulting in linked data), and is crucial to make data exploitable as a whole rather than as isolated, unrelated datasets. This also includes linking unstructured datasets that originate from different natural languages. In order to exploit data meaningfully, it needs to semantically analyzed. This holds in particular for unstructured data, e.g. textual data from which the key messages and facts need to be extracted and expressed with respect to standardized vocabularies. Structuring and linking unstructured data such as text is referred to as linguistic linked data (LLD). Finally, repurposing consists of transforming data so that it can be used for a different purpose than it was originally created for. Repurposing can include merging and mashing up datasets, format transformations, modifications to the data to fit different audiences (experts vs. novices), speakers of different languages, etc.

This document presents a roadmap to create the infrastructure that makes all the above possible and it focuses on three overarching application fields and needs: i) Global Customer Engagement Use Cases, ii) Public Sector and Civil Society Use Cases and iii) Linguistic Linked Data Life Cycle and Linguistic Linked Data Value Chain

With respect to Global Customer Engagement Use Cases, the challenge for the future will be to create ecosystems in which data from different sources and modalities come together, and builds the basis for developing omnichannel experiences for customers. This requires linking of data across modalities and techniques for repurposing and composing information items into stories and narrations that are more amenable and accessible to users. Consistency of message across channels, languages and audiences will be important and crucially supported by (linguistic) linked data technology. As we observe a shift from marketing activities characterized by a push and active recommendations to a new paradigm of customer engagement that is transparent as it represents a commodity that recognizes and fulfills customer needs in real time, richer linked semantic descriptions of users, products, contexts and intentions will be needed to support matchmaking.

In the area of Public Sector and Civil Society Use Cases, linked data can make an important contribution to the creation of a single digital market in which national barriers are overcome. A crucial ingredient for the creation of a single digital market is the development of ontologies and terminologies that harmonize the concepts used in different countries and jurisdictions, as a basis to reach interoperability and develop a new generation of (public) services that is implemented across countries. This is in spirit to the vision behind the Connecting Europe Facility (CEF). New robust methodologies for alignment of different conceptualizations originating from different cultural, national and linguistic contexts, as well as techniques for collaborative, cross-border ontology engineering will be needed. We also foresee that in the near future we will need a better understanding of the domains in which cross-lingual and cross-border communication is urgently needed as a basis to develop shared vocabularies that support language-independent communication in key domains.

As regards Linguistic Linked Data Life Cycle and Value Network Requirements, the future will have to bring an ecosystem in which linguistic resources are easily exploitable by content analytic providers and workflows in a way that provenance, licensing and metadata are clearly exposed to support trustful exploitation of resources. Future efforts will need to create principles for a market in which linguistic linked data, both open and closed, can be traded, developing new business models that include also non-monetary transactions. The goal will be to create an ecosystem in which both linguistic resources and services building upon these are: i) easily discoverable, ii) trustful and certified, iii) comparable and benchmarkable, iv) easily composable and exchangable, v) multilingual, vi) scalable. This involves two challenges: i) bringing the stakeholders together to establish principles for such a market, and ii) develop a technical infrastructure including standardization of APIs and vocabularies that supports plug&play principles.

The first draft of this document has been created by LIDER project partners (FP7 CSA, refererence number 610782 in the topic ICT-2013.4.1: Content analytics and language technologies). The analysis builds on the findings and forecasts of a number of existing public reports on the topic, by aggregating and analyzing different needs and forecasts expressed in these documents as a basis to identify application areas in content and big data analytics where linked data could represent a key enabling technology. Building on the insights about application areas where the potential of linked data technology, in particular linguistic linked data technology, is regarded as very high, a roadmap is defined that extrapolates the above mentioned findings to define a R&D roadmap that can support research organizations, enterprises and funding agencies in decision making and to prioritize R&D investments. A further goal for this roadmap is to define an R&D agenda at the intersection of the language resource, natural language processing and Big Data communities.

Background and Context

This roadmap is a product of four roadmapping events that the LIDER project has organized (see also [here]):

• 1st LIDER Roadmapping Workshop in Athens, 21st of March, 2014, collocated with the European Data Forum (EDF)
• 2nd LIDER Roadmapping Workshop in Madrid, 8th-9th of May, 2014, collocated with the Multilingual Web Workshop (MLW)
• 3rd LIDER Roadmapping Workshop in Dublin, 14th of June, 2014, collocated with Localization World Dublin
• 4th LIDER Roadmapping Workshop in Leipzig, 2nd of September, 2014, collocated with SEMANTICS

Further, the LIDER project has interacted with relevant stakeholders in the context of a number of community groups including:

• W3C Community Group on Linked Data & Language Technologies (LD4LT), with 79 participants
• W3C Community Group on Best Practices for Multilingual Linked Open Data (BMLOD), with 79 participants
• W3C Community Group on Ontology Lexica (ontolex), with 90 participants

Thus, this roadmap is based on the needs and predictions of close to 100 relevant stakeholders from both academia and industry, which have been gathered in the above mentioned events and community groups.

Introduction

Content is growing at an impressive, exponential rate. Exabytes of new data are created every single day (Pepper et al. 2014). In fact, data has been recently referred to as the oil (Palmer et al. 2006) of the new economy, where the new economy is understood as a new way of organizing and managing economic activity based on the new opportunities that the Internet provided for businesses (Alexander 1983).

There are several indicators that clearly corroborate that the exponential growth of data will continue:

• Volume: The data streams already generated today are huge. Only one hour of customer transaction data at Wal-Mart -- corresponding to 2.5 petabytes -- provides 167 times the amount of data housed for example by the Library of Congress.
• Growth Rate: 90% of the data available today has been generated in the last two years only (SINTEF, 2014). The International Data Corporation (IDC) estimates assume that all digital data created, replicated or consumed is growing by a factor of 30 between 2005 and 2020, doubling every two years. By 2020, it is assumed that there will be over 40 trillion gigabytes of digital data, corresponding to 5.200 gigabytes per person on earth (Gantz and Reinsel 2012).
• Internet of Everything: Cisco estimates that currently less than 1 percent of physical objects are connected to IP networks. However, this is estimated to change radically to up to 50 billion devices connected to the Internet by 2020, corresponding to between 6 and 7 devices per person on the planet. These 50 billion devices will constantly generate data at a scale without precedent.

Content analytics, i.e. the ability to process and generate insights from existing content, plays and will continue to play a crucial role for enterprises and organizations that seek to generate value from data, e.g. in order to inform decision and policy making.

A basic distinction can be made between structured and unstructured data. Structured data is essentially data that follows a given pre-defined schema or data model, such as data in standard relational or non-relational (including NoSQL) databases, or data expressed in Web languages such as the Resource Description Framework (RDF). Unstructured data does not follow a predefined schema and comprises texts, blogs, pictures, and sensor data.

Current estimates suggest that only half a percent of all data is being analyzed to generate insights (Gantz and Reinsel 2012). Furthermore, the vast majority of existing data are unstructured and machine-generated (Canalys 2012), with data automatically generated by mobile devices and sensors constituting the majority.

As corroborated by many analysts, substantial investments in technology, partnerships and research are required to reach an ecosystem comprising of many players and technological solutions that provide the necessary infrastructure, expertise and human resources necessary to make sure that organizations can effectively deploy content analytics solutions at large scale to generate relevant insights that support policy and decision making, or even to define completely new business models in a data-driven economy.

Assuming that such investments need to be and will be made, this report explores the role that Linked Data and Semantic Technologies can and will play in the field of content analytics and will generate a set of recommendations for organizations, funders and researchers on which technologies to invest as a basis to prioritize their investment in R&D as well as on optimizing their mid- and long-term strategies and roadmaps.

The main sources this report draws upon are the following:

• Forrester Research Report: TechRadar^TM for AD&D Pros: Digital Customer Experience Technologies, QS 2013 (Yakkundi et al. 2013)
• iVIEW Report by the International Data Corporation (IDC): The Digital Universe in 2020: Big Data, Bigger Digital Shadows, and Biggest Growth in the Far East (Gantz and Reinsel 2012)
• The Global Information Technology Report 2014, published by the World Economic Forum
• Gartner Hype Cycle Special Report 2014
• Text Analytics 2014: User Perspectives on Solutions and Providers by Alta Plana Corporation (Grimes 2014)
• Report Rethinking Personal Data: A New lens for Strengthening Trust, published by the World Economic Forum (2014)
• The LT-Innovate Innovation Manifesto, published by LT Innovate (2014)
• Strategic Research Agenda for Multilingual Europe 2020, published by the META Technology Council (2012)
• Results of the Roadmapping Workshops organized by the LIDER project
• IEEE CS 2022 Report
• Call for Papers from the Linked Data and Semantic Web communities

The report is structured as follows: in the next Section General IT Trends we discuss general IT trends as identified by Gartner in order to position this roadmap with respect to major trends in the field of IT. Section Needs in Content Analytics summarizes the current needs in content analytics, focusing on the survey carried out by Alta Plana (see Section Survey on Text Analytics Needs). Further, we analyse the needs of the application development and delivery industry as identified by Forrester's TechRadar^TM report (Yakkundi et al. 2013) in Section Application Development & Delivery (AD&D). Finally, we summarize the main outcomes of the LIDER roadmapping workshops organized by the LIDER consortium as part of the MLODE workshop and collocated with the SEMANTICS conference (see Section 4th LIDER Roadmapping Workshop).

Section The Funders' Perspective summarizes the main topics funded by the European Commission in the context of Horizon 2020. We discuss in particular the vision and objectives behind the EC's Connecting Europe Facility (CEF) program and how linked data technology can contribute to this vision.

The actual roadmap is presented in Section Roadmap, and mentions the most promising application areas and future directions for the role of linked data, and in particular linguistic linked data, in content analytics. Section Conclusion concludes and discusses ways forward, providing some recommendations for funders and researchers.

General IT Trends

As a basis to identify general IT trends, we consider Gartner's well-known Hype Cycle from 2014. A visualization of the cycle can be found in Figure Gartner Hype Cycle 2014 The hype cycle focuses on newly emerging technologies as they move into mainstream adoption.

The topics mentioned in the hype cycle of 2014 are the following seven technologies

• Mobile: According to the analysis of Gartner, mobile technology is becoming the main vehicle for business applications, allowing organizations to reach more users than via other conventional channels. Due to the pervasiveness of mobile devices among customers, mobile technology is creating disruptive opportunities for business. Thus, Gartner is placing mobile technology as a technology moving rapidly to the peak.
• Internet of Things (IoT): The Internet of Things (IoT) is also considered as a technology moving rapidly to the peak. Given that many companies are defining a strategic agenda for digital business, it is only logical that technologies operating on the physical world will go digital and become part of the network. This is expected to have a large impact and even be transformational with respect to the digital business models and production process, e.g. in the manufacturing area.
• Analytics: As more and more data is generated, e.g. through devices connected to the network as part of the Internet of Things (see above), analytics over these data streams will be an essential ingredient. Analytics technologies are thus regarded as moving into the peak. Gartner foresees that the delivery of analytics as a service (business analytics PaaS) will be a major and important trend. A further trend can be observed in the convergence of information technology (IT) and operational technology (OT), corresponding to the growing use of IT solutions in OT vendors' products. The primary driver in bringing the two worlds together is the need to use analytics from diverse data to improve decision making across the supply chain.
• Big Data: Gartner predicts that Big Data is moving over the peak. Big Data is related to cost-effective information processing of high-volume, high-velocity and highly varied information assets to generate enhanced insights and thus support decision making. While the interest and demand for Big Data solutions is still undiminished, according to Gartner it is moving over the peak due to the convergence of solutions to a set of promising solutions and approaches. The movement of Big Data topics over the peak and the movement of analytics towards the peak clearly suggest that the hype and interest in information and data processing is fostering adoption of these technologies in the value chain.
• Cloud: According to Gartner, cloud technology is generally moving towards the trough. Key topics in the area of cloud technology which are still climbing the peak are cloud access security, cloud application development services, and cloud integration platform services. Topics that are moving over the peak towards the trough include cloud computing, private cloud computing, and hybrid cloud (referring to the coordinated use of cloud services across isolation and provider boundaries among public, private and community service providers, or between internal and external cloud services). Mobile cloud, referring to the use of cloud services for information sharing across devices, has moved quickly over the peak to the trough, indicating maturity for mainstream adoption. Personal cloud services, defined as the individual's collection of digital content, are moving off from the peak, due to their main usage for storage and synchronization, without providing any further significant added value.
• Social: Social technologies are regarded as having shown a significant shift from peak position to the trough. This is mainly due to the fact that users and vendors have realized that generating added and business value on the basis of social technologies is more challenging than expected. The move to the trough, on the other hand, clearly indicates that some social technologies are becoming mainstream, in particular in the area of digital business. Nevertheless, it is foreseen that social technologies, services and disciplines will soon become mainstream technologies within digital business. A technology that is expected to move towards the plateau are peer-to-peer communities, i.e. virtual collaboration environments fostering collaboration among people and organizations outside of the enterprise. Technology for analyzing social networks has moved into the trough due to the difficulty of collecting relevant and reliable data and because turning knowledge into action has been found to be difficult. Some technologies have completely fallen down toward the trough, including social gaming, social TV and social profiles.
• Payment and Digital Business: Technologies for integrating payment methods with loyality programs have moved off the trigger position to trigger midpoint position. Digital Business profiles are placed by Gartner right below the peak of inflated expectations, due to the high number of organizations claiming to have a digital strategy aligned with their overall business strategy.

Summary

The fact that the topics social and cloud are identified by Gartner as moving over the trough to the plateau of maturity clearly shows that these technologies are maturing and becoming part of the IT value chain. Gartner expects that by 2016 cloud and social will be so pervasive that 60% of organizations will have adopted them. The impact that cloud and social have on enabling organizations to run their operations more efficiently, drive new capabilities, serve customers and partners effectively, and respond to disruptive threats and opportunities in the market more rapidly is expected to continue to be key for organizations. The focus in the next years will be more on understanding how social and cloud-based solutions can be applied to generate added value. In any case, any roadmap such as this one in the area of IT will have to consider the prominent role that these two technologies will continue to play in the future. The third relevant technology, Big Data, is regarded by Gartner as moving beyond the peak of inflated expectations, leading to early signs of technological convergence and adoption. Big data can surely be regarded as a key trend in IT that will soon lead to mature solutions that generate valuable insights for enterprises that understand how to effectively make use of these techniques. In this line, IBM has predicted that most companies will have a dedicated role of a data scientist that oversees the organization and understands how to exploit data to optimize the business (Teerlink et al. 2014).

Needs in Content Analytics

This section analyses and summarizes the current main needs in content analytics industries and applications. It summarizes the main results from a survey carried out in 2014 by Seth Grimes from Alta Plana corporation to understand current needs in content analytics. We further reflect on the main results from the LIDER roadmapping workshop that was organized by the LIDER consortium in the context of the MLODE workshop in Leipzig. We further summarize the main challenges identified by Forrester in their TechRadar^TM report ().

Survey on Text Analytics Needs

This section analyzes the top needs in the field of content analytics, focusing in particular on text analytics. The summary of needs here is based on the report Text Analytics 2014: User Perspectives on Solutions and Providers by Seth Grimes from Alta Plana (Grimes 2014). The survey described in the above report gathers responses from 200 participants from the content analytics sector. The survey targeted users or prospect users of text analytics technologies, integrators or consultants or managers as well as executive staff in those roles. Researchers and developers applying text analytics technologies were also invited to participate.

The four technology-related growth drivers for text analytic solutions identified by the survey are the following:

• Open Source: Open source is expected to lower barriers both to technology adoption for researchers and more sophisticated users
• API Economy: The availability of hosted, on-demand API-based web services is expected to lower entry barriers and provides enormous flexibility for adopters
• Data availability: availability of data is crucial and increasing
• Synthesis: Providing support for summarizing and synthesizing datasets, and providing answers to specific information needs is crucial.

The five key market drivers, i.e. application fields, that have the potential to generate business value identified are the following:

• Customer Interactions: including support for customer services and customer experience by deploying text analytics to transitional channels such as contact centers but increasingly also to social media, with the goal of optimizing service and fostering engagement.
• Omnichannel solutions: comprising the analysis and aggregation of data from different channels, e.g. survey, social media, news warranty, chat, contact center data, etc.
• Consumer and market insights: consisting in the deployment of text analytics solutions for market research and the generation of insights about market/customer needs, trends, and opinions. An observable trend is that social data is regarded as increasingly reliable and as a trusted source that can complement classical survey data and deliver complementary insights.
• Search and search-based applications: providing new search functionalities that go beyond enterprise search or online information retrieval to provide a platform for high-value applications that includes advertising, e-discovery and compliance, business intelligence and customer self-service.
• Health care and clinical medicine: providing new analytic solutions, e.g. supporting diagnosis or analysis of claims. These new solutions are expected to drive the market and go beyond mere text mining of scientific literature.

The key data sources relevant for text analytics solutions are:

• blogs and other social media (mentioned by 61% of respondents)
• news articles (42%)
• comments on blogs and articles (38%)
• online forums (36%)

Clearly, User Generated Content (UGC) scores top and can be assumed to continue to play a key role in the near future.

The top business applications identified are (considering only those mentioned by at least 20% of the respondents):

• voice of the customer (mentioned by 39% of respondents)
• brand/product/reputation management (38%)
• competitive intelligence (33%)
• search, information access or question answering (29%)
• Customer Relationship Management (CRM) (27%)
• Content management or publishing (25%)
• Online commerce (16%)
• Life sciences or clinical medicine (15%)
• E-discovery (14%)
• Insurance, risk management, or fraud detection (13%)

Summarizing the written comments provided by the respondents to the survey, we can identify the following trends, requirements and issues:

• Customization: A major issue is the required effort to customize existing solutions to a particular domain and application. A lack of domain-specific models has been identified.
• Usability: Usability of current text analytic solutions is regarded as generally low, with APIs that are difficult to use. Deployment of text analytics is regarded as requiring in-house expertise on NLP and data mining, which is clearly a bottleneck.
• Limitations: Accuracy of current solutions as well as depth of analysis is quite limited (see also Cieliebak et al. 2013)
• Cost/Effort: The cost and effort needed in selecting appropriate vendors and solutions, integrating their solutions, analysing results and generating valuable insights to obtain a ROI requires huge effort and cost. The learning curve for deployment of content analytic solutions is generally too high.
• Integration: Integration into existing workflows or systems, in particular existing Business Intelligence (BI) solutions, is regarded as difficult and requires a major effort.

The following information types are highly relevant in content analytics:

• topics and themes (66% of respondents)
• sentiments, opinions, attitudes, emotions, perceptions, intent (54%)
• relationships and/or facts (47%)
• named entities (56%)
• concepts (51%)
• document metadata (47%)
• other entities (34%)
• semantic annotation (31%)
• events (33%)

Important properties of content analytics solutions are:

• ability to generate categories or taxonomies (65% of respondents)
• ability to use specialized dictionaries, taxonomies, ontologies or extraction rules (54%)
• broad information extraction capability (53%)
• document classification (53% of respondents)
• deep sentiment, emotion, opinion, intent extraction (45%)
• low cost (44%)
• real time capabilities (43%)
• sentiment scoring (41%)
• support for multiple languages (40%)
• open source (37%)
• predictive-analytics integration (36%)
• big data capabilities (33%)
• ability to create custom workflows (33%)
• Business Intelligence (BI) integration (32%)
• sector adaptation (30%)
• support for data fusion (28%)
• hosted or Web service (on-demand API) (25%)
• media monitoring/analysis interface (22%)

Participants also mentioned the need to process content in languages other than English. Top languages (other than English) mentioned by at least 10% of respondents:

• Spanish (38% of respondents)
• French (36%)
• German (34%)
• Italian (18%)
• Chinese (16%)
• Portuguese (13%)
• Arabic (10%)

Summary

The main applications for textual content analytics are in the fields of i) analyzing the voice of the customer in the context of Customer Relationship Management (CRM), ii) brand, product and reputation management, iii) technology surveying and competitive intelligence, iv) content management and publishing, and v) search, information access and question answering. The main source of information is clearly user-generated content (blogs, social media, online forums etc.). While the main tasks required are rather conventional (topic extraction, document classification, entity extraction, relation and event extraction etc.), there is a clear need for analytics solutions that i) are tuned to the needs of particular domains, and ii) are able to generate and incorporate semantic domain-specific knowledge in the form of taxonomies, terminologies etc. to support domain customization.

The main issue with current text analytics technologies identified are clearly: i) lack of standard and flexible APIs, ii) high effort and resources needed for customization and domain adaptation, iii) level of expertise required to deploy and integrate solutions into own workflows, iv) lack of accuracy and depth in analysis (see also Cieliebak et al. 2013). Multilinguality is a further very important feature. Language coverage needs to be extended to address the most important languages: Spanish, French, German, Italian, Chinese, Portuguese and Arabic. Future generations of text analytics solutions will thus have to provide easy-to-use standard and flexible APIs, increase accuracy and depth of analysis and be able to process the most important languages as mentioned above. The ability to integrate additional background knowledge to make domain adaption easier will also be a key capability. Finally, solutions need to be easily integratable into existing workflows, processes and tool chains.

Application Development & Delivery (AD&D)

This section analyzes the main findings of the TechRadar^TM report by Forrester (Yakkundi et al. 2013). Many digital companies are concerned with the challenge to invest and maintain an ecosystem of technology that supports digital customer experiences that are enjoyable, innovative and contextualized. Forrester defines the sum of these digital customer experience technologies as:

A solution that enables the management, delivery and measurement of dynamic, targeted, consistent content, offers products, and service interaction across digitally enabled customer touch points. (emphasis is ours)

The relevant technologies identified in the TechRadar^TM report are the following:

• A/B and multivariate testing: Multivariate Testing (MTV) allows marketers and digital experience professionals to test different components of a digital experience, such as site design, site usability, campaign landing page with respect to some measure, etc.
• Digital Asset Management: Digital Asset Management (DAM) is software used to manage the creation, production, management, distribution and retention of rich media content including audio, video, graphical images and compound documents.
• eCommerce: eCommerce solutions provide companies with the capability to connect with, market, sell so and serve B2B and B2C customers across many digital touch points.
• Email marketing platforms: These tools help to build, execute and monitor email advertising campaigns.
• Mobile analytics: These analytics tools support the collection, analysis and measurement of mobile app traffic and user data to optimize mobile experience.
• Online video platforms: These are solutions that focus on distribution of video content.
• Optimization: Optimization refers to a set of tools that leverage exploratory, descriptive and predictive statistical techniques to drive relevant content, interactions and offerings to end users.
• Portals: Portals are software platforms that aggregate content, data, and applications and deliver them in a rich, personalized environment that supports digital customer experience.
• Product content management: Product Content Management (PCM) focuses on enabling and organization to identify or derive trusted product data and content across heterogeneous data environments.
• Recommendation engines: These tools help to recommend products, text or other content to visitors on different types of websites with the goal of delivering the most relevant and useful experience to the customer.
• Site search: Tools offering search capabilities over content from different systems in a portal or web site.
• Social depth platforms: Tools that integrate social content and experiences into marketing sites.
• Web analytics: Tools supporting the collection of usage data in web channels.
• Web content management: Web content management (WCM) software supports organizations in creating websites and online experiences as well as to create, manage and publish digital content across websites or multiple channels.

The holy grail to effectiveness identified by Forrester is to create contextualized, personalized multi-channel experiences for customers. The following key success factors can be identified:

• Multi-channel experience: multiple channels and target customers can be targeted to deliver a consistent message, also in different languages and modalities, integrating Web content management with product management and eCommerce solutions.
• Contextualization: Portals should be extended to manage and deliver relevant, tailored experiences including personal or individual information on a per-user basis.
• Holistic recommenders: rather than only performing product recommendations, one should create holistic recommender systems that provide pervasive support to users.
• Breaking down silos: customer data, product data and social media data needs to be brought together to deliver contextual cross-channel experiences.
• Adaptability & Control: marketing experts and line-of-business people need more control over digital channels in order to react dynamically to arising customer needs.

Summary

A key challenge in the area of application development and delivery consists in the combination of multiple channels to effectively target customers across channels. On the one hand, this requires to make sure that consistent messages are delivered across channels, which can be ensured by incorporating terminological and lexical resources for the particular domain and application. A further challenge is to foster the convergence of different subsystems (e.g. content management, product management, customer management) etc. into a rich and seamless semantic ecosystem that can support rich customer experiences and contextualized, personalized and situated media delivery and recommendations. A further key challenge consisting in supporting agility in the capture, integration and provision of content, allowing marketing experts to dynamically react to changing customer and market needs.

4th LIDER Roadmapping Workshop

The LIDER project organized a roadmapping workshop in Leipzig on September 2nd, 2014. This workshop was part of the Multilingual Linked Open Data for Enterprises (MLODE) workshop and collocated with the SEMANTICS conference. The program of the roadmapping workshop can be found online. The workshop featured the contributions of more than 10 companies and users from the area of content analytics, which shared with the workshop participants their view on current challenges and perspectives for the exploitation of Linked Data in content analytics tasks.

The main themes identified at the roadmapping workshop were the following six (prioritized with respect to importance according to the vote of the audience at the workshop): i) Resource Creation and Sharing, ii) Open Linked Publishing and Consumption, iii) Multilingual Semantic Content Analytics and Search, iv) Standardization of APIs, and v) Big text and data analytics.

Resource Creation and Sharing

One of the recurring topics of the roadmapping workshop was the need and lack of appropriate linguistic resources to facilitate training of content analytics solutions and in order to adapt content analytics solutions to specific domains. Industry participants stressed that business-friendly licenses that allow for the exploitation of data in commercial contexts are urgently needed. Resources that were most frequently mentioned as important were: terminologies, lexicographic resources, POS-tagged datasets (especially for user-generated content, e.g. tweets), treebanks, datasets annotated with sentiment, NER-annotated corpora as well as corpora in which named entities are linked to external knowledge bases or resources. In general, it was observed that while for most applications resources of middle-level quality are sufficient, for some applications in which quality is crucial highly curated and verified resources are needed. Lexica are a good example for this with a wide spectrum from largely automatically created resources (e.g. BabelNet or automatically generated WordNets) trough to highly curated and manually validated lexical resources such as Princeton WordNet, KDictionaries etc. An important question is how technology can support the quality vs. cost tradeoff in the creation of such resources and how to define human-machine collaborative workflows that increase quality of resources while at the same time minimizing the amount of work needed by experts. The community of content analytic solution vendors clearly mentioned the need for relevant resources for micro-domains, i.e. the availability of highly domain-specific resources (lexica, terminologies, annotated corpora, ontology of intentions) that would support domain adaptation and that could be widely reused across vendors.

Open Linked Data Publishing and Consumption

With respect to publishing data as open linked data, a nuanced perspective emerged during the LIDER roadmapping workshop. While having high-quality open and reusable data was clearly seen as a benefit (see the idea of an open commons mentioned by Alex Pentland (Pentland 2014)), it was also mentioned that the publication of linked data should take into account the principle of economy and efficiency, in particular taking into account that the cost for publishing, quality control etc. should not exceed the added value provided by the resource. Further, many internal datasets at organizations or companies can not be published because doing so would disclose private information about employees or even customers. Further, many datasets are specific to the particular structures and processes implemented in a certain organization and, when taken out of this context, do not provide any added value. In the future, more experiences are needed to understand the cost and value of publishing a certain dataset as linked data as well as metrics to monitor and quantify its impact and reuse. Methodologies that simplify the process of publishing data as linked data are also needed. In general, it was mentioned that the Linked Data effort should focus more on linking rather than only on publishing isolated datasets on the Web. In general, it was also mentioned that instead of focusing on the publication of large multi-purpose datasets, the focus in the future should be on publishing small and reusable buildings blocks that can be used for a specific but frequently recurring purpose. An important target would be to link concepts across languages, effectively creating a multilingual linked knowledge infrastructure that can be exploited in applications that need to provide multilingual support and cross the borders of languages.

Multilingual Semantic Content Analytics and Search

The need for robust and accurate text analytics solutions is still pressing. As identified by the MetaNet activities, the support for most European languages in terms of linguistic resources and natural language processing (NLP) tools is low. This situation was confirmed by the participants of the LIDER roadmapping workshop. Robust and accurate text analytics solutions are needed at different levels: POS tagging, NER recognition, NE linking, information extraction, sentiment analysis, etc. In general, participants expressed the clear need to go deeper and have text analytic solutions that extract deeper semantics including pragmatics including the intention of a customer. Solutions that can perform cross-lingual normalization of content, terms, and named entities are urgently needed, whereby translation is only one component of the needed technology infrastructure. Semantic search (e.g. via keywords, question answering) at the meaning level rather than at the string level is still one of the major needs in the market, as mentioned by several participants of the roadmapping workshop.

The Human Factor

The human factor was identified as an important aspect for the adoption and proliferation of content analytics solutions. In many cases, the expectations of potential end users lead to disappointment with the current state of the art of content analytics solutions. Instead of focusing and reporting performance figures of single tools, experiences need to be gathered on which performance figures are needed for which type of application, and end users need to be sensitized that tools are far from perfect, but that the performance levels can be sufficient for a particular purpose or application. In general, it was felt that the community of content analytics vendors, developers and researchers needs to invest in raising awareness of the benefits, limitations, success stories but also pitfalls of content analytics solutions and devise effective ways of communicating these aspects to potential customers and users.

Standardization of APIs

An important bottleneck in the field of content analytics providers is that each provider offers their own proprietary API, thus hindering easy exchange of solutions and composition of different solutions into more complex workflows. This makes the comparison of different solutions very difficult and further leads effectively to vendor locking as once a company has adopted a certain provider, given the high costs of adaptation (see Section Survey on Needs in Content Analytics), it is likely to stay with that vendor. The participants of the workshop expressed the desideratum of working towards standardization of APIs in content analytics. For one thing, this would make it easier to integrate different solutions into one complex product and would also support joint partnerships between different providers of content analytics solutions. Second, from the point of view of the customer, it would support cross-vendor comparison of technology to make more informed decisions. Finally, standardized APIs would also facilitate benchmarking and quality assessment of tools and services by running automated tests or evaluations, both by the research community but also as part of an open ecosystem of resources and tools.

Big text & data analytics

Surprisingly, the topic of big data analytics was prioritized lowest by the participants at the LIDER roadmapping workshop. One explanation for this is that Big Data as a hype has generated already some disappointment (this is in line with Gartner predicting Big Data to move down from the peak of inflated expectations (see Section General IT Trends). The need to process large amounts of data is clearly there and likely to move into mainstream and adoption, but the hype seems to be decreasing. The challenge of developing solutions that can scale to large streams of big data comprising unstructured data is still a crucial one. The integration of knowledge across languages and formats at larger scale is an important challenge to address.

Summary

The 4th LIDER Roadmapping Workshop was organized in Leipzig on September 2nd as part of the MLODE workshop and collocated with the SEMANTiCS conference. Several companies and users from the content analytics area joined the Roadmapping Workshop and active discussions happened among and with the participants in order to face current challenges on the exploitation of Linked Data in content analytics tasks.

Six main themes were identified during the workshop, going from Resource creation and Sharing (where participants pointed out the lack and the need of appropriate linguistic resources for - possibly, domain specific - content analytics solutions) to Open linked data publishing (with an outcry for economy and efficiency of the publication of linked data), as well as the need of support for resource-poor European languages and the need to raise awareness about the benefits and limitations of content analytics solutions.

Connecting Europe Facility (CEF)

The goal of the Connecting Europe Facility (CEF) program by the European Commission is to support the development of "high-performing, sustainable and efficiently interconnected trans-European networks in the fields of transport, energy and digital services" (European Commission, 2012). By this, it expects to contribute to increase growth, jobs and competitiveness for Europe. A budget of 50 billion EUR between 2014 and 2020 is foreseen for this.

The stated strategic objective of the CEF is to contribute to the development of a Digital Single Market and to effectively eliminate market fragmentation, making sure that cross-border public services are broadly available and accessible by millions of citizens and companies to connect to the single market.

A further goal is to contribute to the transformation of Europe into a knowledge-intensive, low-carbon and highly competitive economy. The CEF is investing thus in the creation of modern and flexible energy, transport and digital infrastructure networks. The preferred instrument for this are public-private partnerships funded through innovative financial instruments that make investment into infrastructure projects attractive.

With respect to the development of digital service infrastructures, the CEF Digital Service Infrastructures (DSIs) are expected to act as platforms on which innovative applications can be created and deployed, and to facilitate mobility of citizens and working across borders. To overcome service fragmentation and lack of interoperability due to national borders, the goal is to develop pan-European services that interoperate across borders and de-fragment the market, in particular in the areas of eGovernment, eProcurement and eHealth.

Funding via Horizon 2020 and CEF can contribute to reducing fragmentation between content analytics solutions. By implementing the needs for content analytics described in Section Survey on Needs in Text Analytics, it can help to make data and services operable across national borders. Linked Data technologies, in particular linguistic, possibly multilingual, linked data technologies, can be expected to play a major role. They can contribute to the interoperability of services e.g. by providing a means to align different conceptualizations or ontologies. This will improve data exchange and semantic interoperability.

For more information on the CEF digital agenda, see [here].

Linked Data in Research

In order to identify relevant topics that are on the current research agenda, we have analyzed the calls for papers of the year 2014 of four major conference in the fields of Semantic and Linked Data technologies: The World Wide Web Conference (WWW 2014), which featured a dedicated Semantic Web track, the International Semantic Web Conference (ISWC 2014), the European Semantic Web Conference (ESWC 2014), and SEMANTICS 2014.

Besides looking at the major conferences, we have also identified a number of workshops specifically dedicated to Linked Data issues, with particular emphasis on linguistic linked data:

• Linked Data on the Web Workshop (LDOW2014), collocated with WWW 2014
• Workshop on Linked Data in Linguistics (LDL-2014), collocated with LREC 2014
• Workshop on Semantic Web Enterprise Adoption and Best Practice (WASABI)
• Workshop on Linked Data Quality, collocated with SEMANTICS 2014
• 1st Workshop on Linked Data for Knowledge Discovery, collocated with ECML/PKDD 2014
• Workshop on Linked Open Data 2014 : Improving SME Competitiveness and Generating New Value, collocated with SEMANTICS 2014

World Wide Web Conference (WWW 2014)

The relevant topics for WWW 2014 in the Semantic Web track were the following:

• Infrastructure: Storing, querying, searching, serving Semantic Web data
• Linking, joining, integrating, aligning/reconciling Semantic Web data and ontologies from different sources
• Tools for annotation, visualization, interacting with Semantic Web data, building ontologies
• Knowledge Representation: Ontologies, representation languages, reasoning in the Semantic Web
• Applications that produce or consume Semantic Web data, including those in the enterprise, education, science, medicine, mobile, web search, social networks, etc.
• Extracting Semantic Web data from web pages and other sources
• Methodologies for the engineering of Semantic Web applications, including uses of Semantic Web formats and data in the development process itself.

European Semantic Web Conference (ESWC 2014)

ESWC 2014 included the following topics in the Open Linked Data Track:

• Linked Open Data extraction and publication
• Storage, publication and validation of data, links, and embedded LOD
• Linked data integration/fusion/consolidation
• Database, IR, NLP and AI technologies for LOD
• Creation and management of LOD vocabularies
• Linked Open Data consumption
• Linked data applications (e.g., eGovernment, eEnvironment, or eHealth)
• Dataset description and discovery
• Searching, querying, and reasoning in LOD
• Analyzing, mining and visualization of LOD
• Usage of LOD and social interactions with LOD
• Dynamics of LOD
• Architecture and infrastructure
• Provenance, privacy, and rights management; relationship between LOD and linked closed data
• Assessing data quality and data trustworthiness
• Scalability issues of Linked Open Data

International Semantic Web Conference (ISWC 2014)

The call for papers for ISWC 2014 included the following topics:

• Management of Semantic Web data and Linked Data
• Languages, tools, and methodologies for representing and managing Semantic Web data
• Database, IR, NLP and AI technologies for the Semantic Web
• Search, query, integration, and analysis on the Semantic Web
• Robust and scalable knowledge management and reasoning on the Web
• Cleaning, assurance, and provenance of Semantic Web data, services, and processes
• Information Extraction from unstructured data
• Supporting multilinguality in the Semantic Web
• User Interfaces and interacting with Semantic Web data and Linked Data
• Geospatial Semantic Web
• Semantic Sensor networks
• Query and inference over data streams
• Ontology-based data access
• Semantic technologies for mobile platforms
• Ontology engineering and ontology patterns for the Semantic Web
• Ontology modularity, mapping, merging, and alignment
• Social networks and processes on the Semantic Web
• Representing and reasoning about trust, privacy, and security
• Information visualization of Semantic Web data and Linked Data
• Personalized access to Semantic Web data and applications of Semantic Web technologies
• Semantic Web and Linked Data for Cloud environments

Linked Data on the Web Workshop, collocated with WWW 2014

The Linked Data on the Web Workshop (LDOW), collocated with the World Wide Web Conference in 2014, called for the following topics:

• Mining the Web of Linked Data
  • large-scale derivation of implicit knowledge from the Web of Linked Data
  • using the Web of Linked Data as background knowledge in data mining
• Integrating Large Numbers of Linked Data Sources
  • linking algorithms and heuristics, identity resolution
  • schema matching and clustering
  • data fusion
  • evaluation of linking, schema matching and data fusion methods
• Quality Evaluation, Provenance Tracking and Licensing
  • evaluating quality and trustworthiness of Linked Data
  • profiling and change tracking of Linked Data sources
  • tracking provenance and usage of Linked Data
  • licensing issues in Linked Data publishing
• Linked Data Publishing, Authoring and Consumption
  • mapping and publication of various data sources as Linked Data
  • authoring and curation of Linked Data
  • Linked Data consumption interfaces and interaction paradigms
  • visualization and exploration of Linked Data
• Linked Data Applications and Business Models
  • application showcases including browsers and search engines
  • marketplaces, aggregators and indexes for Linked Data
  • business models for Linked Data publishing and consumption
  • Linked Data as pay-as-you-go data integration technology within corporate contexts
  • Linked Data applications for life-sciences, digital humanities, social sciences etc.

Workshop on Linked Data in Linguistics (LDL-2014)

The Workshop on Linked Data in Linguistics (LDL-2014), collocated with LREC 2014, mentioned the following topics in their call for papers:

• Use cases and project proposals for the creation, maintenance and publication of linguistic data collections that are linked with other resources
• Modelling linguistic data and metadata with OWL and/or RDF
• Ontologies for linguistic data and metadata collections
• Applications of such data, other ontologies or linked data from any subdiscipline of linguistics
• Descriptions of data sets, ideally following Linked Data principles
• Legal and social aspects of Linguistic Linked Open Data

Workshop on Semantic Web Enterprise Adoption and Best Practice (WASABI)

The Workshop on Semantic Web Enterprise Adoption and Best Practice (WASABI), collocated with EKAW 2014, had a special focus on Linked Data Lifecycle Management and called for the following topics:

• Surveys or case studies on Semantic Web technology in enterprise systems
• Comparative studies on the evolution of Semantic Web adoption
• Semantic systems and architectures of methodologies for industrial challenges
• Semantic Web based implementations and design patterns for enterprise systems
• Enterprise platforms using Semantic Web technology as part of the workflow
• Architectural overviews for Semantic Web systems
• Design patterns for semantic technology architectures and algorithms
• System development methods as applied to semantic technologies
• Semantic toolkits for enterprise applications
• Surveys on identified best practices based on Semantic Web technology
• Linked Data integration and change management

Workshop on Linked Data Quality

The Workshop on Linked Data Quality, collocated with SEMANTICS 2014, mentioned the following topics of interest in their call for papers:

• approaches targeting Linked Data in the areas of:
  • quality assessment
  • inconsistency detection
  • cleansing, error correction, refinement
  • versioning
• reputation and trustworthiness of web resources
• quality of ontologies
• quality modelling vocabularies
• frameworks for testing and evaluation
• data validators
• best practices for Linked Data management
• user experience
• empirical studies

1st Workshop on Linked Data for Knowledge Discovery (LD4KD)

The 1st Workshop on Linked Data for Knowledge Discovery (LD4KD), collocated with ECML/PKDD, included the following topics of interest in the call for papers:

• Linked Data for data pre-processing: cleaning, sorting, filtering or enrichment
• Linked Data applied to Machine Learning
• Linked Data for pattern extraction and behaviour detection
• Linked Data for pattern interpretation, visualization or optimization
• Reasoning with patterns and Linked Data
• Reasoning on and extracting knowledge from Linked Data
• Linked Data mining
• Link prediction or links discovery using KDD
• Graph mining in Linked Data
• Interacting with Linked Data for Knowledge Discovery

Linked Open Data 2014 : Improving SME Competitiveness and Generating New Value

The Workshop on Linked Open Data: Improving SME Competitiveness and Generating New Value, collocated with SEMANTICS 2014, included the following topics in their call for papers:

• Linked Data for SMEs
• Managing the Data Life cycle in SME environments
• Analytics for Improving Business Knowledge using Linked Data
• Transforming Data to Open and Linked formats
• Business Collaboration through Data Sharing and Alignment

Summary

Research in Linked Data is increasingly focusing on issues related to the consumption and application of Linked Data, also in commercial and SME contexts. Key aspects related to the consumption and exploitation of linked data include i) description and discovery of linked data, ii) validation and quality assurance and iii) provenance, privacy and rights management. The research community is thus increasingly working on the above mentioned issues and will eventually contribute to a linked data ecosystem where data trust, provenance and licensing information is taken into account and appropriately represented, and infrastructure is available to ensure for the low-cost publishing, discovery, validation and reuse of linked datasets. Further, current efforts consider how SMEs can exploit Linked Data technology as part of their workflows and in enterprise applications.

Roadmap

We structure the roadmap into three key application areas: i) Global Customer Engagement Use Cases; ii) Public Sector and Civil Society Use Cases; iii) Linguistic Linked Data Life Cycle and Linguistic Linked Data Value Chain.

The goal of the LIDER project is to identify actions and developments that need to happen in order for Linguistic Linked Data (LLD) technologies to impact the three fields identified above. We frame our predictions by indicating:

• Horizon: distinguishing between 1-2 years, 3-5 years, and 5-10 years from the publication of this report, with the horizon of 1-2 years corresponding essentially to initiatives that have already been started (e.g. European research projects funded under FP7 or started in 2014 under Horizon 2020), 3-5 years corresponding to inititatives planned but not started yet, e.g. as foreseen in the Horizon 2020 work program. The horizon 5-10 years corresponds to activities that are not yet foreseen in any plan or work program but are expected to emerge.
• Main Actors: identifying the main actors involved in the initiative and who will push the initiative. We distinguish here between the following actors: academica, industry, SMEs as well as partnerships between these actors.
• Means: instruments by which the initiative can be realized: collaborative research projects, cooperation between academia and industry, standardization, industrial pilots, mainstream adoption, academic proof-of-concept, etc.

Global Customer Engagement Use Cases

Media Publishing & Content Management

As identified by Forrester (Yakkundi et al. 2013), one of the key challenges in media publishing and content management for the years to come is the convergence of technologies to deliver a homogeneous multichannel experience to customers, or as stated explicitly in the report by Forrester "The dynamic nature of digital experiences requires technology that enables business users to manage, measure, and optimize what happens on web, mobile, and social channels". The key objective to achieve here is to combine different technologies to provide a contextualized digital experience to users, bringing together content, product and customer relationship management in one ecosystem. Portals are expected to continue playing an important role in delivering a personalized environment that supports digital customer experiences. The important issue is to realize a certain degree of agility at the content level to be able to quickly integrate new (external) data resources to react to new needs and interests of customers.

The key impact avenues for linked data based content analytics in media publishing and content management are thus the following:

• relying on Linked Data technologies to create unified information spaces of linked datasets bringing together datasets that have been isolated so far (e.g. product content, product data, customer data, social data etc.) to contribute to a unified experience (so called vertical clouds)
• exploit terminologies and ontologies available as linguistic linked data to achieve terminological consistency across channels
• agile import of datasets into portals to support changing customer needs and interests
• support for localization of content across channels by exploiting multilingual linguistic linked data resources

We expect that, in the future, linked data technologies for content publishing will mature to make multimodal and multilingual repurposing of content and storytelling feasible. It will be crucial in this to support access to knowledge -- and not only data -- by non-experts, e.g. content creators and consumers, developing interfaces that abstract from technical aspects, data models and query languages. This access should be in particular across media and across natural languages. This will ultimately lead to visual story generation from multiple sources including text, video and other modalities as well as new methods for re-purposing and composing heterogeneous content for different challenges, natural languages and audiences.

We need further best practices for linked data based content publishing as well as experience reports on the adoption of such best practices in verticals such as energy efficiency management, smart cities, healthcare, etc. together with best practices and business models for generating value ou of such resources. In particular, we advance that enterprises will realize in the coming years the potential of linked data to connect different media beyond separate annotations (cross-media links) and the potential of linked data to share such data across companies.

The integration of content comprises the creation of a seamless network of data and knowledge that spans multiple modalities as well as open and closed datasets in a way that is respectful to IP and corresponding licenses. Assuming that, machines will be mainly responsible in the future for discovering and mashing up content from heterogeneous sources in different formats, modalities and languages in such a way that copyright, IP, confidence and provenance of data is taken into account. Technically, this will require linked data-aware licensing servers which will be responsible for delivering information taking into account access rights, supporting machine-mediated access, extraction, aggregation, composition and repurposing of data.

A further challenge will be to deploy linked data technology to capture social interactions at large scale, annotating multiple media streams with respect to who says what to whom, what reactions are triggered by which content. This includes capturing emotions across modalities, bringing different modalities and languages together to analyze emotions and sentiment effectively.

In the long term we foresee that linked data will converge to create a seamless ecosystem in which structured and unstructured information from different modalities and languages will be integrated and linked, thus being exploitable by a new generation of methods and services that will support storytelling, and question answering over all these sources. This assumes that the technology stack for analysing content using natural language processing techniques at large scale has matured enough.

Supporting exploration of data by citizens and the larger public including analysts, journalists etc., is another challenge. This will effectively contribute to dealing with data and information overload.

Using terminologies and ontologies available as linguistic linked data to achieve terminological consistency across channels would certainly contribute to exploiting data in commercial contexts.

Roadmap for Media Publishing & Content Management

Horizon	Prediction	Actors	Means
1-2 Years	Increased linked data publishing in verticals and for different use cases, first ROI models emerge	Industry-Academia Partnerships	Productive Systems & Integrated Projects
1-2 Years	Increased multilingual terminologies and ontologies in verticals and for different use cases and channels, first ROI models emerge	Industry-Academia Partnerships	Productive Systems & Integrated Projects
1-2 Years	Effective solutions and interfaces for access to data for non-experts (content creators that are not experts) are in place	Academic and Industrial Cooperation	Pilots
1-2 Years	Increased awareness about potential of linked data to connect different media beyond monomodal annotations (cross-media links)	Start-ups and Academia	Hand-on Workshops, Tutorials, Webinars
3-5 Years	Robust and accurate techniques for linking ontologies across languages are available and successfully applied	Industry	Productive Systems
3-5 Years	An increasing number of media publishing companies publish their data about programs, background information as part of the Web, following the early adopters such as BBC, NYT, etc.	Industry	Productive Systems
3-5 Years	Multilingual and cross-media access and exploration of data by citizens becomes possible	Industry-Academia Partnerships	Research Projects and Pilots
3-5 Years	Techniques for bringing together data from multiple modalities for holistic sentiment analysis mature and become productive	Industry	Productive Systems
3-5 Years	Standards for annotation and exchange of multimodal datasets emerge	Industry-Academia Partnerships	Research Projects & Pilots
3-5 Years	Methods for multimodal storytelling by repurposing, summarizing and composing existing and heterogeneous multimodal content are available	Industry-Academia Partnerships	Research Projects & Pilots
3-5 Years	First solutions to combine open and public datasets with appropriate handling of IP and licenses emerge as well as first solutions to discover and assess trust and quality of linked data	Industry-Academia Partnerships	Research Projects and Pilots
3-5 Years	Techniques for large-scale capturing of social interactions and their semantis across modalities emerge	Industry-Academia Partnerships	Research Projects and Pilots
5-10 Years	Non-public data will be available as linked data on the Web; linked data-aware licensing servers take into account and reason about access rights in delivering content	Industry-Academia Partnerships	Research Projects & Pilots
5-10 Years	Linked Data technology supports the seamless integration of structured and unstructured data available on the Web as well as querying across languages of this integrated data	Industry-Academia Partnerships	Research Projects & Pilots
5-10 Years	Linked Data based multimodal and multilingual storytelling matures and is adopted	Industry	Productive Systems

Marketing and Customer Relationship Management

Extrapolating from the importance of social aspects described in Section General IT Trends and the needs expressed both by the survey carried out by Alta Plana (see Section Survey on Content Analytics Needs, our own surveys (LIDER Project deliverable D1.1.1), as well as the findings from the 4th LIDER Roadmapping Workshop), we can assume that the analysis of the voice of the customer will play a major role in guiding marketing and advertising activities in the future. What will be needed in the future are robust techniques to extract and interpret the voice of the customer with i) a high level of accuracy, ii) across natural languages and modalities, and iii) analyzing sentiment at deeper levels beyond mere polarity to recognize also the intent of a user as a basis to generate actionable knowledge. The insights generated by methods to analyze the voice of the customer need to be converted into appropriate metrics that can be integrated into standard BI solutions to be correlated with other measures and in order to measure the impact and ROI of a certain marketing campaign. The holy grail of the advertising industry is to aggregate data from potential and actual customers ubiquitously as a basis to create deep personal profiles in order to provide personalized and contextualized recommendations that permeate the whole life of a user. An important challenge herein is to be able to process large amounts of big data in real time (see Section Big Data issue). With respect to creating deep personalized profiles of users, domain-specific background knowledge available as Linked data can be exploited to create rich semantic profiles that support contextualized, personalized and situated recommendation and interaction. Linked Data technologies can also play a key role in linking profiles of users across channels and sites. This needs to take into account the right of people for privacy (see Section Privacy).

In particular, linked data can contribute to this challenge by the following:

• sentiment lexica in different languages, including polarity information and link to intentions are available as part of the Linguistic Linked Open Data (LLOD), so that these lexica are easily integratable into standard sentiment analysis tools and workflows
• ontologies modeling intentions for a number of micro-domains become part of the LLOD
• datasets annotated with sentiment, subjectivity, polarity and potentially for irony for many languages become part of the LLOD
• robust methods for linking and identifying users across channels and sites as a basis to create aggregated user profiles
• robust and accurate methods for detecting sentiment, subjectivity and polarity and even irony for the major European languages as LLOD-aware services
• providing ontologies/taxonomies/terminologies that can be used to represent semantic profiles of users
• developing robust and accurate methods to identify users across sites and channels
• providing ontologies for modeling situations and contextual parameters to provide situated and contextualized recommendations
• providing domain-specific terminologies and ontologies lexicalized in multiple languages and across modalities to provide consistency across channels, languages and modalities

In general, we can expect disruptive changes to the current paradigm for marketing and customer relationship management. Marketing and advertising activities can be assumed to become totally transparent to the user, moving from an active push over advertising strategies that are embedded in social conversations and communications through to the recognition and fulfillment of intentions and needs in real time. Explicit marketing and advertising activities will loose importance as customer targeting and product placement becomes a commodity service that is perceived as a real added value by customers. This has several implications:

• Machine-to-machine communication: As advertising and marketing moves from a push through to a commodity service that recognizes and fulfills customer needs in real time, we can expect that both businesses and consumers will move from being real physical entities to being digital agents that interact and negotiate directly. This will radically change current business models.
• Rich semantic user profiles: Real-time recognition of needs and intentions requires rich linked information including semantic information about objects, individuals, groups, intentions, contexts, cultures, etc. This will require standardized ways for representing and linking such information.

As more and more private data is linked and profiles become more and more important, and online reputation becomes increasingly relevant for online transactions, new business models will emerge, such as offering to manage online profiles and reputation.

We also foresee that advertising and product placement will become a commodity service that fullfills needs in real time, thus becoming completely transparent, personalized and contextualized. However, this will lead to a situation in which users experience a lack of control and a most likely to a pushback in which users demand for more control of their personal data. Consequently, this will require new solutions and paradigms to empower users in revoking their personal data and solutions for unlinking datasets, thus giving back control to users about their personal data (see Section Privacy)

Roadmap for Marketing and Customer Relationship Management

Horizon	Prediction	Actors	Means
1-2 Years	Paradigm shift from physical interaction with real customers to M2M (machine-2-machine) interaction	Industry	Paradigm Shift
1-2 Years	Paradigm shift away from CRM as a push activity to a (moderated) only conversation	Academic and Industrial Cooperation	Paradigm Shift
1-2 Years	Need for standardized vocabularies for describing user profiles, product information and their relations	Academia-Industry Partnerships	Standardization activites
3-5 Years	Robust and accurate techniques for linking terminologies and ontologies across languages are available and successfully applied	Industry	Productive Systems
3-5 Years	Advertising industry exploits rich semantic interlinked personal and product profiles to provide personalized and contextualized recommendations	Industry	Productive Systems
3-5 Years	First solutions for centralized and trusted management and storage of personal data emerge that consider provenance and licensing terms and conditions and ensure compliance with these	Academia-Insdustry Partnerships	Research Projects & Pilots
3-5 Years	Paradigm shift moving away from explicit advertising to transparent advertising, recognizing intentions and needs in real time to fulfill them, becoming a commodity beyond mere recommendations	Industry-Academia Partnerships	Research Projects and Pilots
3-5 Years	Services for personal data tracking and revokation start to become available	Academica-Industry Parterships	Research Projects & Pilots
3-5 Years	New business models emerge, e.g. for online reputation management and optimization of individuals, companies etc.	Start-ups, Academia	Pilots
5-10 Years	Personalization in advertizing is increasingly perceived as a threat by end users and as interfering with their free choice, demand for more control over data arises; advanced solutions allowing users to manage their personal data arise	Industry-Academia Partnerships	Research Projects and Pilots
5-10 Years	Methods and best practices for retracting and unlinking personal data emerge	Industry-Academia Partnerships	Research Projects and Pilots

Public Sector and Civil Society Use Cases

Supporting the Creation of a Single Digital Market and the Connecting Europe Facility (CEF)

One of the main stated goals of the Connecting Europe Facility (CEF) is to overcome service fragmentation and lack of interoperability due to national borders, with the objective to develop pan-European services that interoperate across borders and de-fragment the market, in particular in the areas of eGovernment, eProcurement and eHealth.

For this, datasets exchanged across borders need to be harmonized both at a syntactic and semantic level. While interoperability at the syntactic level is being addressed already to some extent, establishing interoperability at the semantic level involves a long-term effort involving the alignment of concepts used in different countries and jurisdictions.

With its strong tradition working on ontology alignment and linking, the Semantic Technologies community has the potential to contribute to:

• the development of shared ontologies of key administrative and legal concepts across Europe
• linking of vocabularies and ontologies existing in different countries and jurisdiction to foster interoperability
• development of declarative specifications of workflows and processes, so that tools can reason about these, composing them to achieve some task
• collaborative ontology creation across languages and countries
• exploitation of terminologies and ontologies to ensure consistency of communication in public administration

Roadmap for Supporting the Creation of a Single Digital Market and the Connecting Europe Facility (CEF)

Horizon	Prediction	Actors	Means
1-2 Years	Shared ontologies and terminologies for key administrative, financial and legal sub-domains emerge as part of the Linguistic Linked Open Data Cloud; these ontologies are lexicalized in multiple languages following standard vocabularies and best practices such as the lexicon model for ontologies	Academia-Industry Parternships	Research Projects & Pilots
1-2 Years	Ontologies are increasingly linked across national contexts and published on the Linked Open Data cloud	Academic and Industrial Partnerships	Research Projects and Pilots
1-2 Years	Most important domains in which terminological standardization across languages is needed to realize the vision of a Single Digital Market are identified	Academic and Industrial Partnerships	Research Projects and Pilots
1-2 Years	A taxonomy of types of links that matches needs of the Single Digital Market is developed	Academia-Industry Partnerships	Standardization activites
1-2 Years	Strive to reduce the amount of unstructured data exchanged by exploiting cross-lingual ontologies and terminologies that diminish the need to exchange unstructured messages; identify those fields where language-independent communication is applicable and feasible	Academia-Industry Partnerships	Standardization activites
3-5 Years	First standards and best practices including ontologies to describe services, and products emerge that can be exploited in machine-2-machine negotiation and matchmaking (e.g. matching job offers to profiles)	Industry	Standardization and Pilots
3-5 Years	Robust and accurate techniques for linking ontologies across languages are available and successfully applied	Industry	Productive Systems
3-5 Years	The need to formally monitor compliance of actors across countries with European and other regulatory frameworks becomes obvious, and first solutions for expressing policies and regulations become available (e.g. using advanced logics such as deontic logics)	Academia	Research Projects
5-10 Years	A rich ecosystem of reference and localized ontologies describing key domains in which data exchange across language and national boundaries is crucial has emerged together with an ecosystem in which validated and trusted mappings at different levels of trust and provenance are available	Industry-Academia Partnerships	Research Projects and Pilots
5-10 Years	Robust methodologies for collaborative development of shared ontologies across cultural contexts emerge and are successfully applied	Academia-Insdustry Partnerships	Research Projects & Pilots
5-10 Years	Solutions for monitoring compliance of data and services to regulations are available and deployed at a large scale	Industry	Productive Systems

Localization & Translation

Linked Data Technologies have definitely the potential to impact the current localization and translation market and processes by providing more flexible ways of publishing and exploiting multilingual datasets including parallel corpora, terminologies and translation memories, but also other multilingual datasets of common interest (e.g. DBpedia and BabelNet). Best practices and standards for publishing parallel texts as part of the Linguistic Linked Data cloud and their exploitation in standard localization and translation workflows are needed.

Linked data has the potential to be exploited by translators but also content creators. New paradigms in which content creation and translation are intertwined in the sense that machine translation can be exploited in boostrapping content creation and the other way round will become feasible in the near future. Standards and best practices for licensed linked data need to be developed so that the localization industry can trustfully work with linked data. In the long-term, we foresee that linked data will be an enabler of high-quality personalized translation.

Translation of terminologies and speech-to-speech translation will be two important application fields of MT. Translation of terminologies is crucial to realize the idea of a single digital market and in order to ensure terminological consistency across players and stakeholders and national sites. Speech-to-speech translation is going to play a key role in content creation and consumption, allowing to translate educational offering, social applications, tv programs etc. in real time.

In the longer term, a crucial issue will be to extend coverage to non-European languages and extend the experiences to the languages and contries in which these languages are spoken.

Roadmap for the role of Linked Data in translation and localization

Horizon	Prediction	Actors	Means
1-2 Years	Linked data is increasingly exploited as background knowledge and context by translators and content creators	Industry	Productive Systems
1-2 Years	Best practices and standards for publication of parallel texts, terminologies and translation memories as Linked Data emerge	Academic and Industrial Partnerships	Standardization
1-2 Years	Best practices and standards for publication of speech-to-speech translation data emerge and are increasingly applied	Academic and Industrial Partnerships	Standardization
3-5 Years	A synergetic feedback loop between translation and content creation is developed in many working systems, thus bootstraping multilingual content creation by MT services and feeding back corrections to the MT services to improve in the longer term	Industry & Non-profit organizations	Productive Systems
3-5 Years	Localization industry is increasingly aware of linked data; licensed linked data solutions make it possible for localization industry to share and exploit linked data content	Academia-Industry Partnerships	Standardization activites
3-5 Years	First solutions for live updates of resources by online communities emerge that take into account IP, licensing and provenance aspects adequately	Academia-Industry Partnerships	Research Projects and Pilots
3-5 Years	Large-scale terminology translation and alignment contributes to establish a single digital market across languages (EU members languages)	Academia-Industry Partnerships	Research Projects & Pilots
3-5 Years	Robust techniques for speech-to-speech translation emerge and are deployed in commercial settings and applications	Industry	Productive Systems
5-10 Years	Terminologies for relevant domains are standardized, supporting automatic consistency checks across legislations become possible	Academia	Research Projects
5-10 Years	Fully personalized and contextualized translation supported by linked data is deployed in commercial systems and existing localization workflows	Industry	Productive Systems

Open Data Commons, Data Quality and Data Lifecycle

Alex Pentland has advocated the creation of a data commons that is available to partners under a lightweight legal agreement, such as the trust network agreements. Open data can generate great value by allowing third parties to improve services. (Pentland 2014). This vision is certainly compatible with the ideas behind the Linked Open Data (LOD) project. Further, Alex Pentland states that "A key insight is that our data are worth more when shared because they can inform improvements in systems such as public health, transportation, and government. Using a "digital data commons" can potentially give us unprecedented ability to measure how our policies are performing so we can know when to act quickly and effectively to address a situation." Such a creative commons can be used as a basis to optimize important aspects of our life, including transportation, traffic, energy networks and it can also build the basis for the vision of smart cities (Kaminos 2009) to improve local and regional governance.

A crucial issue in building and using the data commons is ensuring high-quality and up-to-dateness of the data available as part of the data commons. This requires methods for ensuring quality of datasets over their whole lifecycle, monitor usage etc. but also to ensure appropriate access control if data access has to be restricted. In addition, it will be important that all data parts of the data commons are enriched by corresponding licensing information that states the terms and conditions under which the resource can be used, and for which purposes. Such licensing information needs to become an integral part of the data and be in machine readable format to allow for automated discovery, reasoning and filtering. In building a data commons, it is important that datasets are linked across languages to support cross-country and cross-lingual comparisons.

Mention: methods for assessing quality and confidence and data, certification, validation services as well as methodologies so that errors can be spotted and corrected and possibly be feedback to the original source

Roadmap for Supporting the Open Data Commons, Data Quality and Data Lifecycle

Horizon	Prediction	Actors	Means
1-2 Years	Linked Data Network is considered as a viable option for the realization of the data commons	Academia-Industry Parternships	Research Projects & Pilots
3-5 Years	Most Linked Data endpoints implement an access control layer	Academic and Industrial Partnerships	Research Projects and Pilots
5-10 Years	Services for monitoring quality, evolution and uptime of datasets as part of the data commons are in place, mature and widely used	Academic and Industrial Partnerships	Research Projects and Pilots

Linguistic Linked Data Life Cycle and Linguistic Linked Data Value Chain

Linguistic Resource Development and Sharing

Many of the use cases and needs mentioned in the sections above require the availability of linguistic data, corpora and lexical resources in multiple languages. Such resources are generally scarce, as identified by MetaNet.

Towards increasing the coverage and quality of linguistic resources, the following aspects need to be considered:

• Licensing Information and Provenance: Licensing and provenance information need to be attached to the data, ideally in machine readable form, define the conditions under which data can be used, also in commercial settings in which revenue is obtained from the use of the resource.
• Resource Market: A business-to-business and research-to-business market for high-quality data trading needs to emerge; one possibility is to market data per API and establish pay-per-use models.
• Quality and Availability: Resources need to have high availability and be of high quality. Frameworks for monitoring availability and quality of resources need to be established.
• Stimulation of resource development: The creation of resources can be stimulated by initiating partnerships between academic and industrial consortia that jointly work on a dataset that all can exploit for their purposes at zero cost under the condition that the resource is licensable by third parties under fair conditions.
• Sharing and Discovery: The easy sharing and discovery of resources needs to be ensured by appropriately extending the functionality of current metadata repositories for linguistic data.

Roadmap for Language Resource Development and Sharing

Horizon	Prediction	Actors	Means
1-2 Years	Vocabularies for providing licensing information in machine readable form are standardized	Academia-Industry Parternships	Standardization
1-2 Years	Vocabularies for describing linguistic datasets become standardized	Academia-Industry Parternships	Standardization
3-5 Years	Metadata repositories implement improved functionalities for discovery of relevant resources by means of current Web standards (e.g. SPARQL); these repositories adopt the standardized vocabularies developed by relevant communities in addition to their own vocabularies based on internal datamodels to allow for interoperability	Academic and Industrial Partnerships	Research Projects and Pilots
3-5 Years	Metadata repositories ensure that the link to the actual data is available, following a number of best practices	Academic and Industrial Partnerships	Research Projects and Pilots
3-5 Years	Agent-to-agent negotiation for data exploitation is realized in pilots	Academic and Industrial Partnerships	Pilots
3-5 Years	Aggregator services that collect, aggregate and index metadata about linguistic resources emerge providing added value as brokers and are increasingly used to discover linguistic resources	Academia-Industry Partnerships	Pilots & Research Projects
5-10 Years	A cross-border market for linguistic resources is established and in operation	Industry, Academia and Non-profit organizations	Productive Systems
5-10 Years	An ecosystem of services that validate and benchmark data are available and widely used	Industry, Academia and Non-profit organizations	Productive Systems

Linguistic Linked Data Value Chain

A crucial aspect in the future will be the establishment of a value chain and appropriate ecosystem and infrastructure for the creation, marketing, exchange, consumption and modification of linguistic data. The Linguistic Linked data cloud can provide the basis for such an ecosystem, but the following aspects need to be taken into account:

• Business models: Business models for all the actors along the chain need to be developed, that is for resource creators, resource traders or brokers including discovery platforms, as well as those actors that enrich, manually validate and improve datasets; non-monetary rewards and transactions in terms of community recognition or increase of reputation need to be explored
• Trust & Rating: An ecosystem for assessing trust and rating stakeholders based on reputation etc. is needed
• Data quality and benchmarking: Methods for independent measurement and benchmarking of data quality need to be developed, quality models for comparing different resources are needed; it is important to stress that data quality is nevertheless difficult to measure inherently and to a large extent is determined by the value that the data generates in terms of ROI for certain applications and is thus something that the market should agree upon; data quality is in many cases determined by factors external to the data itself, i.e. documentation, recommendations of data by others etc.
• Standardization & Plurality: Standardized formats for resources, services and APIs need to be agreed upon as a community effort and based on open standards, vocabularies and best practices; at the same time, plurality and backwards compatibility needs to be supported, having converters from legacy formats into standard formats available as part of the value chain
• SW extensions: extensions to SPARQL and other SW technology is needed as SPARQL does not match perfectly all use cases for querying parallel text, speech data etc.
• Data curation and improvement: Community contributions to improve a certain resource are in line with the distributed nature of linked data, but infrastructure, tools and workflows to support this need to be developed.
• Call for tenders: Call for tender models in which potential customers can call for the development of a certain resource will likely play an important role in the future
• Integration: Plug&play integration of external datasets with internal datasets should be supported, meshing up and combining datasets should be supported by agreed-upon common formats and appropriate tooling
• Architecture for LLOD-aware services: Service architectures that scale and rely on distribution and stream processing and exploit open Web protocols and Semantic Web standards (JSON-LD) to provide content analytics services and exploit the LLOD as background knowlede are needed

Roadmap for Linguistic Linked Data Value Chain

Horizon	Prediction	Actors	Means
1-2 Years	New business models for linguistic resource creation also based on non-monetary transactions and rewards emerge	Academia-Industry Parternships	Research Projects and Pilots
1-2 Years	Proposal for an architecture for LLOD-aware services building on open Web and Semantic Web Standards emerge	Academia-Industry Parternships	Research Projects and Pilots
3-5 Years	Business models for data creation, brokering and improvement are established	Academic and Industrial Partnerships	Productive Systems
3-5 Years	Models for the creation of linguistic resources following the call for tenders paradigm are explored in pilots	Academic and Industrial Partnerships	Research Projects and Pilots
3-5 Years	First quality models for linguistic linked data emerge	Academic and Industrial Partnerships	Research Projects & Pilots
3-5 Years	Principles and best practices for measuring and representing trust and quality of datasets and services by independent parties that provide independent certification services for linguistic resources emerge	Academia-Industry Partnerships	Pilots & Research Projects
3-5 Years	Best practices and standards for publishing the most important types of linguistic resources are available and widely used; convertors to migrate legacy data into the LLOD are available	Academia-Industry Partnerships	Pilots & Research Projects
5-10 Years	Models based on call for tenders for the development of linguistic resources are in place and widely used	Academia, Industry, Non-profit organizations	Productive Systems
5-10 Years	LLOD-aware services following architectural best practices and open Semantic Web vocabularies and Web protocols that can be easily composed and integrated into worfklows are widely deployed	Industry, Academia and Non-profit organizations	Productive Systems

Orthogonal Topics

Lying in the intersection of the three topics in Figure VENN DIAGRAM, a set of common developments has to be addressed, including data privacy and data protection, data provenance and data licensing.

Data privacy and data protection

As more and more (personal) data is collected from users, the issue how to ensure responsible and trustful handling of this data emerges as a crucial one. In this regard, the report Rethinking Personal Data: A New Lens for Strengthening Trust published by the World Economic Forum mentions that: "The growth of data, the sophistication of ubiquitous computing and the borderless flow of data are all outstripping the ability to effectively govern on a global basis", with the result that "Industry, government and civil society are all uncertain on how to create a personal data ecosystem that is adaptive, reliable, trustworthy and fair". (Kearney 2014).

Especially critical here is passively generated data, e.g. data generated by sensors, devices or wearables, as users are typically unaware of the data generated nor have they typically provided consent for the use of the data.

In fact, the World Economic Forum has started a global dialogue on the topic of how to ensure responsible use of personal data, identifying the following three key issues:

• Delivering meaningful transparency: giving individuals higher transparency with respect to how data is used, simplifying the way in which data practices are communicated to individuals.
• Strengthen Accountability: creating an ecosystem and incentive infrastructures to ensure principled and enforceable data use. This implies that there needs to be verifiable evidence by stakeholder organizations that relevant measures are being taken to ensure compliance with data usage best practices.
• Empower Individuals: Individuals should be empowered on the one hand to be able to decide about how their data is used but also to be able to use their data for their own purposes. On the other hand, they should be able to understand and manage the impact of data usage.

It has been argued that a first step in developing good practices for responsible and trustful data usage is to have a taxonomy of types of personal data (Kearney 2014, Chapter Near-Term Priorities for Strengthening Trust) that distinguishes at the very least:

• individually provided data: data voluntarily provided by individuals through forms, surveys, social media applications etc.
• observed data: data generated by some mobile or wearable device or any other sensor
• inferred data: data generated as part of some data mining or machine learning algorithm on the basis of either individually provided or observed data.

While the user is aware and actively participates in the provision of individually provided data, this is not the case for observed or inferred data where there is typically no awareness from the side of an individual on which data is being collected and for which purpose.

A reference model for personal data management has been outlined in (Kearney 2014, Chapter Long-Term Issues and Insights) and distinguishes the following three layers:

• Infrastructure: The infrastructure layer contains the technology, services and applications required to assure the availability, confidentiality, security and integrity of the data, both while in transit and at rest.
• Data management: The data management layer focuses on the transfer and exploitation of personal data as specified in corresponding permissions and policies. Metadata is crucial to enrich the data by a layer that allows to express permissions and provenance as a basis to ensure compliance with agreed-upon policies and terms and conditions. The metadata regarding licensing and provenance needs to remain attached to the data during the whole data lifecycle.
• User interaction: The user interaction layer facilitates a transparent interaction of individuals with service providers regarding the terms and condition of their personal data.

In addition, Alex Pentland (2014) mentions a set of five key policy recommendations for large organizations with respect to data management:

• Distribution: Large data systems should store data in a distributed manner, separated by type (e.g., financial vs. health) and real-world categories (e.g., individual vs. corporate).
• Provenance and Views: Data sharing should always maintain provenance and permissions associated with data, and should support automatic, tamper-proof auditing. Best practice here would be to share answers only to questions about the data, e.g. by relying on views rather than sharing the data themselves, whenever possible.
• Secure external data sharing: External data sharing should take place only between data systems that have similar local control, permissions, provenance, and auditing, and should include the use of standardized legal agreements such as those employed in trust networks.
• Best practices for data flows: Best practices for data flows to and from individual citizens and businesses is to require them to have secure personal data stores and be enrolled in a trust network data sharing agreement
• Secure identification protocols: All entities should employ secure identity credentials at all times.

In a data economy in which data is the new oil and many processes and services are optimized in a data-driven fashion and thus large amounts of data -- also pertaining to individuals -- is accumulated, the issue of how to ensure trust and preserve privacy are issues of utmost importance. Working out models that allow to create value out of data while at the same time respecting privacy rights is thus an important topic on the political agenda.

The center of the European agenda of actions to improve the data protection and the privacy of EU citizens is the General Data Protection Regulation (GDPR) expected to be issued in 2014. This regulation extends the scope of the EU data protection law to all foreign companies processing data of EU residents, harmonizes the data protection regulations throughout the EU (supported by the European Data Protection Board) and proposes a single, centralized Data Protection Agency to be responsible for taking legally binding decisions against a company (GDPR, 2012).

In the new regulation, the notice requirements will be strenghtened, and additional information will have to provided regarding the retention time for personal data, the contact information of both the data controller and the data protection officer. Data controllers will have to prove that they have the person's consent and will have to inform faster if a data breach has occurred, informing also the person. The right to erasure (Art. 27) will improve the users' privacy (see the case affecting Google, ECLI:EU:C:2014:317) but will also add a burden on the data management procedures.

There are several technical measures that have been identified as clearly contributing to the creation of a trustful data ecosystem that respects the privacy rights of individuals and to which Linked Data technologies can clearly offer a contribution to:

• Distribution: partitioning data and physically distributing it across sites is a measure to make aggregation of person data more difficult and thus ensure that no single individual has access to all datasets about a person in one place (Pentland 2014). Linked Data can strongly contribute to this as it inherently relies on data distribution across a network of physically distributed but logically linked datasets. Linked Data could thus provide the basis for data distribution as envisioned by Alex Pentland (2014).

• Access Control: appropriate access control and authentication services are needed to ensure that data access is limited to those users that are allowed. The Linked Data technology stack thus needs to be extended by an access control and authentication layer.

• Views: Alex Pentland argues that most data access should be in the form of views rather than through direct access to the data. This is also compatible with Linked Data technologies where SPARQL can be seen as a language to define views and expose the results of this view over the Web. Combined with appropriate mechanisms for access control, Linked Data could thus provide the basis for the view-based access to data that Pentland advocates.

• Machine-readable provenance and licenses: it is crucial to ensure compliance with licenses and that license and provenance information are attached to the data and remain so during the whole data lifecycle to prevent misuse of the data. It is crucial to make the terms and conditions specifying how the data can be used transparent in all steps of the data lifecycle. Machine readable licenses that can be attached to the data (e.g. in RDF) are thus needed and can be made an integral part of the data.

Copyright

Much of the linguistic data is indeed subject to the Intellectual Property laws. Regardless of whether the language resource is expressed as Linked Data or not, most of the published resources have a protection that has to be respected by data consumers. This protection comes from the EU Copyright Directive (Directive 2001/29/EC) if the resource is a work, and from the EU Database Directive (Directive 1996/9/EC) if the resource is simply regarded as data.

Even if a large percent of the published material is published as Open Data (namely, with light requirements imposed by the rightsholder for the resource to be used, derived or redistributed), a non-negligible portion of the resources is published with strong restrictions on its use.

The landscape is likely to be modified in the forthcoming years. A Public Consultation on the review of the EU copyright rules was held during 2013 and 2014 and mid-term changes in EU copyright law may be on the horizon. These changes will probably integrate a better handling of digital resources and their licenses published online, more uniform rules throughout Europe and a more transparent management of the copyright collecting societies. This may imply forcing a swifter management of the rights and an increased importance of automated rights management systems.

Roadmap for Privacy & Trust

Horizon	Prediction	Actors	Means
1-2 Years	Regulation on the processing of personal data (GDPR) enters in force	Legal bodies	New regulation
1-2 Years	Proposals for access control over Linked Data are consolidated and standardized	Academia-Industry Parternships	Research Projects & Pilots
3-5 Years	Most Linked Data endpoints implement access control layer	Academic and Industrial Partnerships	Research Projects and Pilots
3-5 Years	A new Copyright Directive empowers automated right management processing	Legal bodies	A new Directive
3-5 Years	Standards for provision of licensing and provenance information in RDF are available	Academia-Industry Partnerships	Standardization
5-10 Years	Inclusion of licensing and provenance information in machine readable form using standard W3C vocabularies is widely adopted	Industry	Productive Systems

Big Data & Content Analytics

In the application fields mentioned above, processing large amounts of heterogeneous as well as multilingual data is a must. The challenge for the years to come consists in developing robust techniques that can process and sift through large amounts of data to generate insights in near-real time. As most of the content generated is still of an unstructured nature, big data techniques need to be applied to efficiently process large amounts of unstructured data and combine it with existing structured data as well as data in other modalities (audio, video, image, 3D data, etc.). Following the trend identified by Gartner to provide analytics services on the clouds, it seems reasonable to assume that these solutions for processing unstructured content at large scale will be deployed and run on the cloud at the place where the data resides. Currently, linguistic data provision and provision of analytics services is fragmented, with many services existing using different standards, data formats, licenses, etc. It is crucial to develop an ecosystem of linguistic data and content analytics services where data and services operate together seamlessly. Adoption of open web standards (RDF, SPARQL, OWL, JSON) as well as modern web service technology (e.g. REST) is key towards accomplishing this ecosystem. Towards processing Big Data, workflows need to be distributed so that processing can be efficient. For this, deploying worfklows and NLP services on the cloud needs to be facilitated, allowing people to easily configure a workflow and then executing the workflow on the cloud. Standard interfaces and APIs needs to be provided and implemented by legacy and new services to make them easily integreatable and composable to address more complex tasks. In order to realize such an ecosystem of high-throughput services that implement compatible APIs and interfaces and are thus composable into complex workflows, the following is needed:

• Standard APIs and interfaces: defining standard APIs and interfaces based on standard W3C and other standardized vocabularies defined by the Linked Data community (e.g. NIF, lemon, etc). This will allow to easily combine and exchange services easily and make the task of integrating them into existing workflows easier.
• Best practices for publishing of linguistic linked data resources: best practices for publishing and sharing linguistic linked data resources are developed for the most important types of resources
• Best practices for implementation of NLP services: Best practices for the implementation of content analytics services that are directly layered on top of the current web architecture are developed, requiring only HTTP as protocol but no additional protocols such as SOAP or other RPI methods.
• Flexible LLOD-aware service architectures: new LLOD-aware architectures for service deployment and composition that build on current Web standards such as JSON-LD are developed; this new generation of LLOD-aware services can consume LLOD resources and produce new resources that are dynamically added to the LLOD.
• Queryable repositories of linguistic linked data: Well-known repositories and providers of linguistic resources expose metadata following Web standards such as SPARQL and standard W3C vocabularies to support querying repositories by machines.
• Service deployment on the cloud: Infrastructure is created to instantiate and deploy content analytics workflows on the cloud as well as to train NLP components with datasets as part of the LLOD and deploy these on the cloud.
• Certification and validation of data and services: Infrastructure for validating, benchmarking and certifying the performance and quality is available and implemented by independent parties, e.g. aggregators of services.
• Multilinguality: Tools that extract all the relevant entities (see Section Survey on Needs in Content Analytics) in all European languages need to be available.
• Variety: Approaches that can efficiently process different types of resources including text, structured as well as other modalities (video, audio, images) become available.
• Scalability: in order to scale to large amounts of content, content analytic services need to effectively rely on distribution and parallelization as well as optimally balance offline and online computation to support real-time performance.

Roadmap for Big Data & Multilingual Content Analytics

Horizon	Prediction	Actors	Means
1-2 Years	Standard interfaces for NLP services based on open standards and vocabularies are defined	Academia & Industry	Standardization
1-2 Years	Best practices for publishing linguistic resources fostering exploitation by content analytics services emerge	Academia-Industry Parternships	Research Projects & Pilots
1-2 Years	Proposals for flexible architecture for LLOD-aware services emerge	Academia-Industry Parternships	Research Projects & Pilots
3-5 Years	Infrastructure and standards for easy deployment of services on the cloud emerge and are increadingly used by content analytics providers and users	Standardization, Academia-Industry Partnerships	Research Projects and Pilots
3-5 Years	New approaches for scalable content analytic services that are deployed on the cloud and rely on distribution and parallelization emerge and are increasingly used	Industry	New Products and Services
3-5 Years	Standard Interfaces and APIS for NLP and content analytic services are widely adopted	Industry	Productive Systems
5-10 Years	Best practices and approaches for balancing online and offline computation in content analytics to provide answers in real time emerge and are applied.	Academia & Industy	Integrated Projects & Productive Systems
5-10 Years	Robust approaches to process heterogeneous data sources efficiently are developed and used	Academia & Industy	Integrated Projects & Productive Systems
5-10 Years	A landscape of mature tools that support extraction of relevant entities (topics, NE, facts, relations etc.) are available for all European languages and are used in industrial contexts	Industry	Productive Systems

Conclusion

In this report we have analyzed the potential impact of linked data technology, in particular, linguistic linked data, on contant analytics. We have in particular identified a number of key application fields in which linked data technology can be expected to have a major impact, and we have identified the steps together with their timeframes required to unleash this potential impact and provide a clear added value in the application areas identidied.

In the application field of Global Customer Engagement, linked data has a strong potential to support the creation of rich multimedia experiences that deliver content to users as a multichannel and mutlimodal experience, supporting also storytelling. Linked will contribute to this vision techniques for integrating and linking data across sites, media and languages and it will support repurposing of content with respect to language, modality and audience. It will support the semantic description of people, products, contexts and situations and intentions and it will exploit these descriptions to provide tailored, personalized and contextualized user experiences and interactions. We forsee that current advertising and customer interaction models in which content or recommendations are pushed to customers will radically change towards models in which single and collections of customers are addressed via a (moderated) dialogue. This will require advanced techniques for analyzing robustly the voice of the customers including their intents and desires, for all European languages, but also techniques that automatically generate content in real time, also in multiple languages. Lexicalized and multilingually linked terminologies and ontologies will play a major role here to ensure consistency of message as well as in supporting the repurposing of content across languages, modalities and audiences.

In the area of public sector and civil society, linked data will contribute to create an ecosystem of data that is partially open and partially closed but is extended with appropriate provenance and licensing information as well as mechanisms for representing and dealing with trust and confidence, so that the public as well as private companies can exploit the data within their purposes and applications. Simplifying access to data by appropriate interfaces, e.g. based on natural language, is a crucial goal to achieve. Linked Data is a crucial part to create the digital commons and to distribute data and make it accessible over SPARQL views, thus fullfilling basic criteria to develop secure and trustful information networks Most importantly, linguistic linked data technology has the potential to contribute to the creation of a single digital market in which services operate across the borders of languages and nations. For this, relevant application fields in which communication can be elevated to a language-independent, semantic level need to be identified and appropriate vocabularies need to be developed. Further, linked data can contribute to linking vocabularies, terminologies and ontologies from different linguistic and national contexts as a basis to achieve interoperability as well as as a basis for the localization of services.

With respec to the Linguistic Linked Data Life Cycle and Linguistic Linked Data Value Chain, linked data technology has the potential to effectively improve the way that linguistic data is provided, by making it better discoverable and exploitable in current workflows and applications. For this, current hosters of linguistic metadata should adopt semantic and linked data technologies to explose metadata in addition to provenance and licensing information so that discovery of relevant datasets becomes possible, also by machines. Adoption of standardized vocabularies for metadata description but also for datasets (e.g. lemon) are crucial to support interoperability here. In order to cater for the large need for linguistic resources, a language resource market that builds on linked data needs to emerge, creating new models of revenue that include non-monetary benefits for those who create, curate, refine or extend existing resources. Call for tenders for the creation of linguistic resources could turn to be a flexible and efficient mechanism. Collaborative methodologies for the creation of linguistic resources in which each actor is recompensed in some way for their contribution will need to emerge.

Building the above sketched ecosystem for linguistic linked data, a landscape of LLOD-aware natural language processing and content analytics that exploit linguistic linked data effectively and in addition: i) are discoverable, ii) rely on standardized APIs and interfaces, and iii) are this easily integratable into standard workflows as well as composable and are iv) scalable. Scalability is particularly important to make this ecosystem of LLOD-aware services ready to support big data application in which large amounts of unstructured data (including texts, videos, images but also other modalities) need to be anaylzed efficiently. Build on distribution of services over the cloud and data as well as supporting stream scenarios is crucial in this context.

A crucial issue for the future is to appropriately deal with intellectual property, copyright but also privacy. In all these fields we forsee a strong potential for linked data technology as a way to ensure that provenance and licensing information remains attached to the data over the whole lifecycle, thus supporting awareness of the copyright and provenance of data, also when it is aggregated and meshed-up with other data, as well as to monitor compliance. Linked Data vocabularies as well as the RDF language as a way to make such information machine readable will play a crucial role here, supporting machine-to-machine negotiation.

Especially to boost adoption by SMEs, the cost-effectiveness of solutions is critical. There also need to be clear guidelines for exploring data sources as linked open data. Proof-of-concept implementations and implementations and show cases can help to analyse benefits and cost for a larger audiences, in joint partnerships between academia and industry.

References

1. Robert Pepper and John Garrity. The internet of everything: How the network unleashes the benefits of big data. In The Global Information Technology Report 2014. World Economic Forum, 2014.
2. M. Palmer. Data is the new oil, November 2006,
3. C.P. Alexander. The new economy. Time Magazine, May 1983.
4. SINTEF. Big data, for better or worse: 90% of world’s data generated over last two years. ScienceDaily, May 2013
5. J. Gantz and D. Reinsel. The digital universe in 2020: Big data, bigger digital shadows, and biggest growth in the far east, December 2012
6. Beñat Bilbao-Osorio and Soumitra Dutta and Bruno Lanvin (eds.), The Global Information Technology Report, World Economic Forum, 2014
7. Anjali Yakkundi, David Aponovich, and Mark Grannan. Techradar^TM For AD&D Pros: Digital Customer Experience Technologies, QS 2013. Forrester Research, 2013.
8. Alex Pentland. Big data: Balancing the risks and rewards of data-driven public policy. In The Global Information Technology Report 2014. World Economic Forum, 2014.
9. Marc Teerlink, Paula Wiles Sigmon, Brett Gow and Kingshuk Banerjee. The new hero of big data and analytics: The Chief Data Officer, IBM Global Business Services, Executive Report, 2014.
10. Mark Cieliebak, Oliver Dürr, and Fatih Uzdilli. Potential and Limitations of Commercial Sentiment Detection Tools. In: Proc. of the Workshop on Emotion and Sentiment in Social and Expressive Media: approaches and perspectives from AI, 2013
11. LT Innovate. The LT-Innovate Innovation Manifesto’. LT Innovate, 2014.
12. A.T. Kearney (ed.). Rethinking Personal Data: A New Lens for Strengthening Trust. World Economic Forum, 2014.
13. European Commission. Connecting europe facility: Investing in europe’s growth, September 2012
14. Komninos, Nicos. doi:10.1504/ijird.2009.022726 Intelligent cities: towards interactive and global innovation environments. International Journal of Innovation and Regional Development (Inderscience Publishers) 1 (4): 337–355(19), 2009

1. Meta Technology Council. Strategic Research Agenda for Multilingual Europe 2020. Springer, 2012.
2. LIDER project. [, https://www.w3.org/community/ld4lt/wiki/Lider_roadmapping_activities Lider roadmapping activities], 2013

1. European Commission 2012, Proposal for a REGULATION OF THE EUROPEAN PARLIAMENT AND OF THE COUNCIL on the protection of individuals with regard to the processing of personal data and on the free movement of such data (General Data Protection Regulation), available at http://ec.europa.eu/justice/data-protection/document/review2012/com_2012_11_en.pdf