Péter Mika, Aldo Gangemi 

Descriptions of Social Relations

Peter Mika
Department of Business Informatics &
Department of Management and Organizations
Free University Amsterdam, The Netherlands
pmika@cs.vu.nl
Aldo Gangemi
Laboratory for Applied Ontology (LOA)
Institute for Cognitive Sciences and Technology
National Research Council, Rome, Italy
gangemi@ip.rm.cnr.it

 

Summary

This note presents a proposal for a minimal extension of social ontologies such as FOAF with the reification (first-order treatment) of social relationships. Such a treatment is required to capture much of the richness in the characterization of relationships as discussed in the Social Science literature and is also necessary for a semantics-based integration of social network information coming from different sources and contexts. We provide a philosophically-motivated conceptual model as well as a quick-and-dirty RDF representation. The work also adheres to the design principle of minimality as observed by FOAF in that only the most general, but necessary concepts are introduced, leaving to the community the extension of this framework with concrete relationships and qualities.

Motivation

We argue that ontological representations of social networks such as FOAF would need to be extended with a framework for modelling and characterizing social relationships for two principle reasons: (1) to support the automated integration of social information on a semantical basis and (2) to capture established concepts in Social Network Analysis, which provides the most significant toolkit for processing social networks with the purpose of understanding social structure and its effects.

While the area of social-semantic applications is still in the period of rapid formation, social relationships as metadata are increasingly featured in a number of software categories:

Social networks in personal content applications. Personal content applications use social ontology for the annotation and retrieval of multimedia documents. Tidepool from Immuexa is a personal application that lets users collect and organize their 'personal memories' in digital form. Tidepool adds RDF-based annotations (FOAF, in specific) to multimedia documents and allows users to navigate through content by browsing through ontological terms. Tidepool also combines this technology with instant messaging for annotating images collectively and sharing content and metadata with friends and family.
The earlier co-depiction experiment of the RDFWeb project demonstrated how personal information in the form of FOAF profiles can be used to annotate digital photos in a completely distributed fashion.

Social ontology for online communities and social networks. The purpose of networking sites is to let users organize their network connections (by creating profiles and linking to profiles of others), discovering new possible ties in the process and recovering connections to old-time friends or other relations. The functionality imitates the local search process of real social networks (by letting users browse the friendship network) and relies on the high clustering of social networks (the friends of our friends are likely to be friends as well). Popular networking sites such as Friendster or Orkut (and the professionally oriented LinkedIn) are mostly closed systems that take a centralized approach in storing network data, while FOAF uses Semantic Web technology for interoperability and promotes a decentralized approach.

Social networks in enterprise Knowledge Management. In the domain of KM, the categories of Enterprise Document Management and Collaboration software are rapidly merging into integrated solutions where metadata regarding the personal profiles and social networks of experts is combined with metadata about the documents and other content of the enterprise. Social network analysis packages from vendors such Entopia and Verity build on this metadata to provide overview and analysis of the human capital of the enterprise. (The rapid emergence of the new category of social software prompted Business 2.0 magazine to elect Social Network Applications as the software category of the year in 2003.)

Besides the increasing amount of social network metadata generated by these applications, the emerging field of social network mining provides methods for recovering social interactions and networks from legacy sources such as web pages, databases, mailing lists, personal emails etc. With the variety of sources and contexts that social information is coming from, the problem has now become to integrate and consolidate this information on a semantical basis before applying methods of network analysis. Two key problems in this area is the disambiguation of identities (people in the different sources may be referred to in different ways) and the aggregation of network relationships. In the following we focus on the second problem, which requires the engineering method of decomposition: the representation of social relationships needs to be fine-grained enough so that we can capture all the detail from the individual sources of information in a way that these can be later recombined and taken as an evidence of a certain relationship.

The second motivation for a richer representation of social relationships comes from the need to accomodate concepts from Social Network Analysis (SNA). SNA is distinguished from other fields of sociology by (1) a focus on relationships between actors rather than attributes of actors, (2) a network view (sense of interdependence) and (3) a belief that structure affects substantive outcomes (emergent effects). The models of SNA are based on graphs, with graph measures such as centrality that are defined using a sociological interpretation of graph structure. These measures are then often used to correlate with measurable effects of social structure such as the observed status of the individual within a community, see Mika (2004). Methods of SNA have been traditionally applied to data collected by survey techniques, but its measures are increasingly employed in the analysis of electronic data, see e.g. Tyler et al. (2003), Mika (2004).

For illustration, we list below some of the most commonly discussed characteristics of social relationships. (We focus on interpersonal-relations in specific, ignoring social relationships at different level of analysis, such as institutional relationships or institutional trust.)

  • Sign: (aka valence) A relationship can represent both positive and negative attitudes such as like or hate. The positive or negative charge of relationships is important on its own for the study of balance within social networks.
  • Strength: The notion of tie strength was first introduced by Granovetter (1973) in his groundbreaking work on the benefits of weak ties. Tie strength itself is a complex construct of several characteristics of social relations. In her survey, Hite (2003) lists the following additional aspects of tie strength discussed in the literature:
    • Affect/philos/passions (Granovetter, 1985; Krackhardt, 1992; Uzzi, 1999)
    • Frequency/frequent contact (De Burca et al., 2001; Granovetter, 1985)
    • Reciprocity (Granovetter, 1985; Portes and Sensenbrenner, 1993; Powell,
      1990; Uzzi, 1999)
    • Trust/enforceable trust (Portes and Sensenbrenner, 1993; Powell, 1990; Uzzi, 1996)
    • Complementarity (Powell, 1990)
    • Accommodation/adaptation (Powell, 1990)
    • Indebtedness/imbalance (Powell, 1990)
    • Collaboration (Powell, 1990)
    • Transaction investments (Powell, 1990)
    • Strong history (Powell, 1990)
    • Fungible skills (Powell, 1990)
    • Expectations (Portes and Sensenbrenner, 1993)
    • Social capital (Portes and Sensenbrenner, 1993)
    • Bounded solidarity (Portes and Sensenbrenner, 1993)
    • Lower opportunistic behavior (Provan, 1993)
    • Density (Staber, 1994)
    • Maximize relationship over org. (Powell and Smith-Doerr, 1994)
    • Fine-grained information transfer (Uzzi, 1996)
    • Problem solving (Uzzi, 1996)
    • Duration (De Burca et al., 2001; Uzzi, 1999)
    • Multiplexity (De Burca et al., 2001; Uzzi, 1999)
    • Diffusion (MacLean, 2001)
    • Facilitation (MacLean, 2001)
    • Personal involvement (De Burca et al., 2001)
    • Low formality (few contracts) (De Burca et al., 2001)
    • Connectedness (De Burca et al., 2001)
    • ...
    As of yet no agreement has been reached in the field as to the importance of these individual aspects of tie strength (Marsden and Campbell, 1984), likely because researchers tend to ignore aspects that are irrelevant to their actual study. More unfortunate is the fact that no agreed upon operationalization have emerged yet for measuring them, which means that researchers in the field use different ellicitation methods and questions when it comes to determining tie strength either as a numerical value or as a binary distinction between weak and strong ties.
  • Provenance: A social relationship may be viewed differently by the individual participants of the relationship, sometimes even to the degree that the tie is unreciprocated, i.e. perceived by only one member of the dyad. Similarly, outsiders may provide different accounts of the relationship, which is a well-known bias in SNA.
  • Relationship history: Social relationships come into existence by some event (in the most generic, philosophical sense) involving two individuals. (Such an event may not require personal contact (e.g. online friendships), but it has to involve social interaction. Note that the 'knows' notion of FOAF is somewhat misleading in this sense, e.g. I know (cognitively recognize) George Bush, but I certainly never had any social interaction with him.) From this event, social relationships begin a lifecycle of their own during which the characteristics of the relationship may change through interaction or the lack of (see e.g. Hite & Hesterly, 2001). Recording the temporal dimension of the relationship may be important for social-semantic applications where past experience in the relationship needs to be taken into account.
  • Relationship roles: A social relationship may have a number of social roles associated with it, which we call relationship roles. For example, in a student/professor relationship within a university setting there is one individual playing the role of professor, while another individual is playing the role of a student. Both the relationship and the roles may be limited in their interpretation and use to a certain social context (see below). Social roles, social contexts and their formalization are discussed in (Masolo et al., 2004)

In summary, a rich ontological characterization of social relationships is needed for the characterization and analysis of individual social networks as well as the consolidation (merging or syndication) of social network information that comes from multiple sources and possibly different contexts, which is the typical scenario of the Web. (For example, Orkut allows to describe the strength of friendship relations on a 5-point scale from "haven't met" to "best friend", while other sites may choose other scales or terms.) Even if no shared typology of social relations or shared characterizations emerge in the short term, minimally a mechanism is required to represent such relations and qualifications, in order to facilitate eventual ontology mapping.

Conceptual model

The importance of social relationships alone suggests that they should be treated on the first-order. Social relations are (mostly binary) predicates, their instances being the concrete relations among the participants of the relationship. Social relations are also socially-constructed objects in the sense of Masolo et al. Much like social roles, social relationships have a strong contextual dependence in that they own their definition (the ability to identify them) to the social context in which they are interpretable. For example, a student/professor relationship at the Free University of Amsterdam (and the attached role of student and professor) is defined by the social context of the university and this kind of relationship may not be recognizable outside of the university. (In another sense, we may talk about student as the entire class of roles of students at learning institutions around the world.) Similarly, friendship is interpreted in context, so much so that a wide body of sociological literature is concerned with the interpretation of friendship in different contexts. Intuitively, those of us who have lived in different cultures for extended periods have all experienced the differences in attitudes toward friendship even within Europe.

In summary, the definition and use of a social construct such as a social relationship is limited to a social context. Only within this social context are we able to identify and interpret a certain interaction-pattern among individuals as a certain kind of relationship. This process, which is known as cognitive structuring works by applying the generic pattern we associate with such a relationship to the actual state-of-affairs we observe. However, the same observed pattern may be interpreted according to another theory as a different kind relationship. The individual relationships and their generic description are thus clearly separate. The generic pattern of the relationship comprises those and only those aspects that are shared among particular occurences of the relationship (for example, there are always two distinct roles in the case of a student/professor relationship with certain requirements for playing those roles). The description is partial in the sense that it allows variation in the particular relations between individuals.

The representation of context and the separation of the level of state-of-affairs (observations of objects and sequences of events) from the higher level of descriptions (contexts) that can be used to interpret those state-of-affairs turns out to be a common problem in the representation of much of human knowledge. A solution proposed by (Gangemi and Mika, 2003) is the Descriptions and Situations ontology design pattern that provides a model of context and allows to clearly delineate these two layers of representation.

D & S is a generic pattern for modelling non-physical objects whose intended meaning results from statements, i.e. it emerges in combination with other entities. For example, a norm, a plan, or a social role is usually represented as a set of statements and not as a concept. On the other hand, non physical objects may change and be manipulated similar to physical entities, and are often treated as first-order objects. That means that an ontology should account for such objects by modelling the context or frame of reference on which they depend. D & S is an ontology-design pattern in the sense that it is used as a template for creating domain ontologies in complex areas. D &D has been successfully applied in a wide range of real-life ontology engineering projects from representing Service Level Agreements (SLAs) to the descriptions of Web Services (Mika, 2004b).

D & S builds on some basic categories from the DOLCE foundational ontology, namely the notions of Objects, Events and Regions. (These concepts represent the top level ontological choice in almost all Foundational Ontologies.) As depicted in the Figure, the notion of Context in D & S is composed of a set of Parameters, Functional Roles and Courses of Events. Axioms enforce that each descriptive category acts as a selector on a certain basic category of DOLCE: Parameters are valued-by Regions, Functional Roles are played-by Objects (endurants) and Courses of Events sequence Events (perdurants). The elements of the context thus mirror the elements of the state-of-affairs (a set of objects, events and their locations), but add additional semantics to them. Note also that these levels of description and situation are clearly separate in that the same state-of-affairs may be interpreted according to another theory by mapping the elements of that other theory to the same set of objects and events. D & S captures the intuition that multiple overlapping (or alternative) contexts may match the same world or model, and that such contexts can have systematic relations among their elements.

D&S has been already used by Masolo et al. for the representation of social contexts and social roles. Their arguments about the context dependence of social roles equally hold for social relations and we follow their approach in using D&S for the design of our conceptual model for the representation of social relationships. In particular, we model a Social Relationship as a subclass of Context and particular social relationships such as Friendship a subclass of this generic concept. As contexts, Social Relationships can have a number of Parameters, Roles and single Course as components.

A typical Role is the Relationship Role, a subclass of the Social Role concept introduced by (Masolo et al., 2004). An example of a relationship role is (the trivial) Friend role in a friendship relation, the Student and Professor roles in a student/advisor relationship and the Uncle/Nephew roles of kinship. Relationship Roles are restricted to be played by the Natural Persons.

The course of the relationship captures the generic characteristics for the course of a certain relationship, i.e. the kinds of event and their possible sequences that characterize a certain kind of relationship. The course is related to the actual events in a particular relationship by the sequences relationship.

Characteristics of relationships such as the ones mentioned above are conceptualized as parameters, mostly a requisite for the course of the relationship. For example, frequency may be axiomatized as the average number of events in the course of the relationship within a given time unit. We recognize that softer qualities of relationships (such as the emotional content of the relationship) may be harder to capture precisely, but the engineer should strive in any case to relate it to other components of the relationship. (If the semantics cannot be captured precisely, at least the ellicitation question(s) that were used to determine the quality should be documented.)

The Figure shows the representation of the friendship relation (some property instances are not shown for visualization purposes). Friendship in general is a social relationship with a single role called Friend, played by actual persons such as Jack and Jill. Friendship also has a typical course; an event such as a dinner where both Jack and Jill have participated may be related to this course, which would indicate at least that it has a significance to the development of a friendship between Jack and Jill. (Friendship is difficult to capture more precisely in this respect in that there is hardly a typical course for a friendship. Nevertheless, one may discern typical events, such as the point that the participants consider as the "beginning" of their friendship.)

RDF(S) representation

Social relationships are described in FOAF as RDF statements with a subject and an object of type foaf:Person and a predicate that is a subproperty of the foaf:knows relationship. Qualifying a statement in RDF requires reification and for this purpose a minimal vocabulary is provided by the RDF Schema specification. (Note that reification may be required for a different reason other than qualifying the relationship: while most relationships are binary, some social relationships may in fact have an arity higher than 2. For example, brokering is a relationship between two people and a broker. Representing such relationships again requires some form of reification.) This vocabulary includes the rdf:Statement class and the rdf:subject, rdf:predicate and rdf:object properties. Note that these concepts have no semantics in OWL DL. (Their use is allowed, i.e. it is not considered an extension of the otherwise protected rdf namespace.)

 

We propose two alternative RDF(S) representation of relationships, both using the reification mechanisms of RDF(S) to reify the original triple asserting the existence of the relationship, in other words relationships become subclasses of the rdf:Statement class. Common also to both representations is that the new Relationship class (as a subclass of rdf:Statement and dolce:social-description) is related to a general Parameter class (a subclass of dolce:parameter) by the hasParameter relationship (a subproperty of dolce:temporary-component). Relationship types such as Friendship are subclasses of the Relationship class, while their parameters (such as strength or frequency) are subtypes of the Parameter class. Note that the hasParameter metaproperty cannot be defined in OWL DL (its domain is rdf:Statement while its range is owl:Class or some subclass of it).

The two alternatives differ in the representation of parameters. The first scheme borrows from the design of OWL-S for representing service parameters, as used in the specification of the profile of a Web Service. Here, parameters are related by the valued-by metaproperty to their range (owl:Thing or a datatype, depending on whether the parameter takes objects or datatypes as values). For example in an application Strength may be a subclass of Parameter valued-by integers. The disadvantage of this solution is that specifying values requires two statements or the introduction of a constructed property (the necessary axiom in not expressible in OWL).

 

The second alternative differs in that the 'native' method of RDF is used for representing parameters: the generic Parameter class is defined as a subclass of rdf:Property. This model has the advantage that it becomes more natural to represent parameter values and restrictions on them. The disadvantage is that this solution is not compliant with OWL DL: declaring properties ranging on properties and creating subclasses rdf:Property are not allowed in this species of OWL.

Conclusion

In this paper we first discussed the need for a shared and agreed model of social relations and then moved to propose a conceptual model and a way for extending social ontologies such as FOAF with this model. We hope this paper will serve the purpose of stimulating a discussion around this issue and an agreement on the best practice in order to facilitate the interoperability of the growing number of applications using social ontologies. Acknowledgement Research for this paper has been generously funded by VUBIS, the VU Research School for Business Information Sciences.

Download

Model 1 (RDFS) Model 2 (RDFS)

References

Granovetter, M. (1973). "The Strength of Weak Ties." American Journal of Sociology, 78 (May): 1360-1380.

Granovetter, M. (1985). "Economic Action and Social Structure: The Problem of Embeddedness." American Journal of Sociology, 91(November): 481-510.

Hite, Julie M. (2003). “Patterns of multidimensionality in embedded network ties: A typology of relational embeddedness in emerging entrepreneurial firms.” Strategic Organization!, 1(1), 11-52.

Hite, Julie M. & Hesterly, William S. (2001). “The evolution of firm networks.” Strategic Management Journal, 22(3), 275-286

Masolo, C. et al. (2004). "Social roles and their descriptions." Proceedings of the Ninth International Conference on the Principles of Knowledge Representation and Reasoning. AAAI Press, 2004.

Gangemi, A. & Mika, P. (2003). "Understanding the Semantic Web through Descriptions and Situations." Proceedings of the DOA/CoopIS/ODBASE
2003 Confederated International Conferences. LNCS 2888. Springer Verlag, 2003.

Mika, P. (2004). "Social Networks and the Semantic Web: An experiment in online social network analysis." To appear in Proceedings of the IEEE/WIC/ACM International Conference on Web Intelligence, IEEE Computer Society Press, 2004.

Mika, P. et al. (2004b). "Foundations for Service Ontologies: Aligning OWL-S to DOLCE." In Proceedings of the Thirteens International World Wide Web Conference (WWW2004), ACM Press, 2004.

Marsden, PV & Campbell, KE (1984). "Measuring tie strength." Social Forces, 63, 482-501.

Tyler, J., Wilkinson, D. and Huberman, B. (2003). "Email as Spectroscopy: Automated Discovery of Community Structure within Organizations." http://www.hpl.hp.com/research/idl/papers/email/

 



Home    Research    Personal    Links    Contact
© 2002 Peter Mika