01:15:33 dbooth has joined #ledp 01:58:22 timbl has joined #ledp 02:06:21 timbl_ has joined #ledp 03:28:35 ballen has joined #ledp 03:38:39 sspeiche has joined #LEDP 03:40:20 sspeiche has left #LEDP 03:42:11 Arnaud has joined #ledp 04:41:23 dbooth has joined #ledp 13:00:31 RRSAgent has joined #ledp 13:00:31 logging to http://www.w3.org/2011/12/07-ledp-irc 13:00:39 rrsagent, make logs public 13:02:26 Topic: Semantic Web Application Patterns: Pipelines, Versioning and Validation 13:03:44 by David Booth 13:04:59 http://dbooth.org/2011/ledp/ 13:07:10 martynas has joined #ledp 13:07:20 julius has joined #ledp 13:10:01 dbooth is talking 13:10:03 Arnaud has joined #ledp 13:10:40 joined new startup PangenX in healthcare 13:11:01 Addressing the semantic integration problem 13:11:28 create the illusion of a single unified data source 13:11:36 ...using RDF in the middle 13:11:53 sspeiche has joined #LEDP 13:12:01 There's some kind of data production pipeline 13:13:34 Question: Martin Nally: We made people turn data at the edge into http resources 13:14:24 use ontologies/rules etc within the data pipeline 13:14:52 David Wood: Suggests breakout session about minimal server for HTTP+RDF 13:15:15 timbl has joined #ledp 13:15:21 Illustrates with Cleveland Clinic reporting pipeline 13:15:43 A dependency graph 13:16:01 Each node is a consumer and producer of data 13:17:01 Martin Nally: Suggests a comparison of standard ETL techniques with an RDF based approach would be interesting. 13:17:41 Convert into RDF at the edges - but you're not done, plenty more work to do. 13:18:01 David Wood: LeeF is going to talk about data segmentation - also relevant. 13:18:37 Need to be smart about tracking when data changes - not doing redundant work 13:18:56 open source project "RDF Pipeline" PoC 13:19:11 The pipeline (graph) is described in RDF 13:20:03 Each node is a processing stage combining a wrapper and update code 13:20:10 bheitman has joined #ledp 13:20:29 Each node declares its inputs (as an RDF List) 13:20:41 Each node has a URL 13:20:50 ale_de_vries has joined #ledp 13:21:32 The updaters are found in well known places (not described in the RDF) 13:21:48 The updater does the interesting work 13:22:04 This is like RDF Make / Ant 13:22:45 Updaters have their own policy about when to recompute their output (lazy, eager, periodic etc) 13:23:34 The wrappers are typed (eg. FileNode, SparqlGraphNode) 13:24:01 FileNode is simply a wrapper for a shell-script 13:24:23 The SparqlGraphNode inputs and outputs to/from named graphs 13:24:47 Wrappers handle inter-node communication, HTTP by default 13:25:20 If you have 2 nodes in the same JVM you don't need to go out to HTTP, the wrapper can be directly invoked 13:25:49 This is a framework not an API. based on an RDF description. 13:26:07 This is language agnostic - you don't even need to use RDF 13:26:30 part 2: Ontologies and rules for semantic transfromations 13:26:46 uses SPARQL as a rules language (see also SPIN) 13:27:18 (unlike SPIN) this uses SPARQL INSERT 13:27:53 This is more efficient than using CONSTRUCT, because data is handled within the triple-store 13:28:14 Uses dynamic combination of named graphs 13:28:42 equivalent to multiple from named clauses 13:28:54 David Wood: Mulgara does something like this 13:30:08 My question: Why not use the multiple from named construct? Because you might have thousands (e.g. think of patient records) 13:30:33 TBL: Have you considered other graph operations? 13:32:02 David Wood: Algenraic operation s sneak in through real-world use-cases 13:32:28 TBL: Are you looking for the language for writing pipeline operations. eg. In CWM you merge and filter graphs. 13:33:32 The motivation is that ad-hoc pipelines are a nightmare to maintain. 13:35:02 Martin Nally: Was this produced for a customer? Would be more convinced there's a market if a major customer had demanded this. 13:35:47 Oracle has an RDF product (Oracle semantic technology) 13:37:20 Cornelia has joined #ledp 13:37:24 There will be RDF support in DB2 - not a major announcement, just incremental development. 13:38:16 Martin Nally: Is there a role for the W3C to compile information about customer use-cases. 13:39:01 TBL: W3C will do education outreach, including case-studies. 13:39:02 LeeF has joined #ledp 13:39:31 These will explain the ROI 13:40:15 David Wood: Aware of a use-case where they have millions of (named) graphs. They run into exactly this kind of processing problem. 13:40:28 Last points about URI versioning 13:40:36 ballen has joined #ledp 13:40:39 Change the URI or the semantics? 13:41:00 point 1: Publish your versioning policy whatever way you go. 13:41:51 Point 2: Old and new URIs can co-exist (in RDF). Doesn't break extant software 13:43:03 Martin Nally: Works well in publishing web. Read/Write web is more problematic 13:43:12 http://dbooth.org/2011/ledp/ 13:43:51 How to validate in an open world? 13:44:04 AlanYagoda has joined #ledp 13:44:12 1: model integrity: is what I'm producing sensible? 13:44:48 2: Suitability for use - this is defined by the consumer with own expectations. 13:45:19 Producers supply validator for the data they produce, the consumer for the data they expect. 13:45:32 "SPARQL is my hammer" 13:46:34 Martin Nally: How do I publish my validation (rules) ahead of time - not just a runtime task. 13:46:55 dbooth: The validator is a design-time construct serving that purpose 13:47:59 Martin Nally: There's runtime variability. How do we capture that? 13:48:11 [slide 24] 13:48:46 How do you describe that _structure_ (metadata)? 13:49:18 The consumer expects more that just RDF 13:49:43 Sandro: first approx is a description of the vocabulary 13:50:03 s/description/enumeration/ 13:50:32 Then the constraints on the data. 13:51:36 Martin Nally: Schema can be used to form and constrain data. Can SPARQL be used in the same way? 13:52:12 Sandro: Easy to write inscrutable SPARQL 13:52:54 Sandro: Should be possible to use SPARQL (ASK) and schema interchangeably 13:53:37 dboot: SPARQL ASK can be used as a validator 13:54:18 Martin: This is a view definition, it can be used in different ways. 13:55:16 Question: Isn't RDF self-describing? Properties you (the consumer) recognises can be used however they like. What's the role of validation? 13:55:37 TBL: For example, in geodata there's a lat for every long. 13:56:10 dbooth: has finished 13:56:39 I wanna mention SPIN again - it has some vocabulary for validation. Haven't tried it though 13:58:00 One more question: How does this contrast with SPARQL motion? 13:58:21 dbooth: that has a centralized architecture 13:58:52 EricP begins 13:59:13 https://www.w3.org/2011/Talks/1207-acls-egp/ 13:59:21 on slide 2 13:59:32 davidwood has joined #ledp 13:59:54 About Web Access Control 14:00:35 If the atomicity is the document you end up with hundreds of named hraphs expressing the ACLs 14:01:17 Express ACLs for a particular graph pattern, not a particular document. 14:01:42 Analogous to Oracle label security 14:01:51 cmatheus has joined #ledp 14:02:02 Existing Abstract POlicy Language 14:02:20 helthcare dominated by XACML 14:02:59 XACML needs to be profiled, leans heavily on HL7 14:04:34 The expressivity eg. "SalesManager in Boston implies access to regional projections". A bunch of conjunctions 14:05:01 Enforcement ensures request falls within the policy 14:05:18 Expressed in RDF "just for fun" 14:06:00 Enforcement by SPARQL extension functions used in a SPARQL filter. Standard tooling doesn't do anything for you. 14:06:26 Virtual views (SPARQL constructs create a virtual view) 14:06:39 Give syou control over what the consumer can see. 14:07:12 For example you can hide private information 14:08:07 Sprinkle the access controlled data in SPARQL OPTIONALS that will leave the sensitive data unbound. 14:09:17 Each optional has a condition that looks at the access (named) graph for authorisation 14:09:35 slide 13 14:09:52 ACLs are visibly next to the sensitive information. 14:10:05 Makes it easier to debug and inspect 14:11:58 Arbitrary SPARQL expressivity: eg. you can have conditions that are selective on the medication. 14:12:26 XACML doesn't give you that degree of expressivity 14:13:07 Martin: IBM has done some query rewriting for security - ran into performance issues 14:14:19 dbooth: Was the performance issue because they were doing it on the fly. 14:14:45 Martin: No not construction, but execution of the (rewritten) query. 14:15:27 dbooth: What did you do instead? 14:16:13 Martin: fell back to something closer to the virtual graphs. Jena built this at a lower level rather than doing this in-query. 14:17:37 EricP: data obligation (time sensitivity) is hard. 14:18:32 Policy injection (of patterns from a XACML) is future work. 14:19:18 Contrast with Oracle object level security - the users will choose. 14:19:49 Question: Is fine grained access control (at the triple level) really useful? 14:20:01 cmatheus has joined #ledp 14:21:04 s/Question/Ashok/ 14:22:05 Elsevier: We need to create views that particular customers can see. 14:22:34 EricP has finished 14:23:09 LeeF up next 14:24:03 http://dl.dropbox.com/u/11365687/ledp/ledp2011-data-segmenting-in-anzo.pptx 14:24:19 davidwood has joined #ledp 14:24:28 http://dl.dropbox.com/u/11365687/ledp/ledp2011-data-segmenting-in-anzo.pptx 14:24:57 "Data Segmenting in Anzo" 14:25:15 Topic: Data Segmenting in Anzo 14:25:21 by Lee Feigenbaum 14:25:29 Where do we see value in the loinked data space 14:26:06 Cambridge Semantics support deployment of the Anzo platform. 14:26:32 Our hammer is named graphs. 14:27:22 Smallest unit of granularity for versioning, access control. Using TriG 14:28:25 Does everything go into a named graph? single triples (eg. for statement level control) ? 14:29:00 ballen has joined #ledp 14:29:09 Do named graphs correspond to documents? sometimes docs are artificial constructs. 14:30:05 Or all triples sharing the same subject - gets the job done. subject triple closure. 14:30:39 concise bounded description - though the moral is to avoid bnodes where possible. 14:32:33 Use annotation to denote some properties as internal. eg. The wheels of a car are treated as 'internal' to the car (and end up in the same named graph) rather than having separate graphs for each wheel. 14:33:02 TBL: You're treating the direction of the relationship as being significant. 14:33:48 EricP: You're making access control decisions. 14:34:48 TBL: RDF isn't a tree - don't do object oriented programming in disguise. 14:35:02 TBL: with this design, my decision about modeling with "parent" vs "child" becomes an Access Control decision. Direction of link shouldn't matter. 14:35:57 rulesguy has joined #ledp 14:36:31 dbooth: The direction is being used as a heuristic way to organize the data. 14:37:11 Sandro: The unit of AC is a subject oriented graph. 14:39:11 TBL: This makes it harder to navigate to the parent - it ends up in a different named graph. 14:39:13 sandro: do we automatically also get to see who the subject parents are, or who the children are, having been given access o the subject graph...? 14:39:58 The impact is that we end up with millions of small (named) graphs 14:40:30 But this corresponds with the natural granularity for permissions. 14:40:56 The challenge is to find the right graph 14:41:39 ...and sometimes multiple graphs (eg. including both the parents and children of a resource) 14:42:12 The graph name is the same as the resource name 14:42:37 Ora: RDF lets you use the same name for different things so that's OK 14:43:55 Sandro: You mean the name (URI) of the subject? Yes 14:44:21 ora has joined #ledp 14:45:00 Elsevier: We treat named graphs as a simple kind of packaging. They are combined later for a given application. 14:45:49 eg in notepad: ex:Lee { ex:Lee a ex:Person; ex:name "Lee" } 14:46:06 That was TriG, the name ex:Lee serves two roles 14:47:07 not kosher but pragmatic 14:48:03 The fallback is a system-wide sparql query. 14:48:25 ( IMHO, the reason Lee hasn't come across problems with this is they're not really doing decentralization / linked data. ) 14:48:40 mcdonoug has joined #ledp 14:48:59 graphs are replicated (cached) on the client 14:49:51 Now for linked - data... everything exposed as LD. It dereferences URIs. 14:50:40 LD priniciples are not used internally, but for public consumption. 14:52:40 Sandro: Does it do 303 redirects? 14:52:54 davidwood has joined #ledp 14:54:09 Impractical to enumerate millions of FROM..NAMED. Similar to dbooth but we call them 'named datasets'. 14:54:51 TBL: It's a virtual union graph 14:55:41 RDFS, OWL used as expressive data model (not really open world semantics) 14:55:58 publish RDFa and support JSON serializations. 14:56:22 SPARQL rules. Like dbooth, used CONSTRUCT but switched to INSERT 14:56:47 SPARQL ASK for preconditions and validation 14:57:38 Message: Anzo driven by semweb technologies, but it needs to be integrated within conventional software architectures. 14:57:58 Wary of standards that don't affect interoperability 14:57:58 ] 14:58:53 Lee: The time is totally ripe for a new Semantic Web Education / Outreach effort. 14:59:00 time is ripe for education and outreach. We're often asked to use their tools against arbitrary SPARQL endpoint. The answer is NO. 14:59:29 SPARQL 1.1. service description may make this easier. 15:00:10 standards needed for: advertising content of linked data sources, SPARQL endpoints 15:00:35 standards for named datasets and other SPARQL extensions 15:01:02 David: What extensions? 15:01:21 A few hundred function extensions. 15:03:18 Question: Do you have a feeling for the right kind of sizing for graphs in typical application? 15:03:49 Answer: It depends on the application, depending on performance metrics 15:04:33 Have you looked at Kasabi (re advertising content of LD)? Will look into it. 15:06:33 David Wood: Taking an action item to set up group to look into use/extensions/federation of SPARQL 15:07:22 e.g. what functions, extensions are supported by an endpoint. 15:07:38 David Wood: Also which predicates. vocabulary 15:07:46 TBL: That's a different level 15:08:46 There are a few different aspects, different groups are concerned with different aspects, they need to get together and agree, adopt and implement. 15:09:31 David Wood: There's a core of people who would be interested in talking about this and making a member submission. 15:09:54 EricP has picked up the ball and taken the action item. 15:10:22 LeeF has finished 15:11:00 dbooth has joined #ledp 15:11:44 betehess has joined #ledp 15:11:57 scribe: Alexandre 15:12:02 scribenick: betehess 15:12:13 RRSAgent, generate minutes 15:12:13 I have made the request to generate http://www.w3.org/2011/12/07-ledp-minutes.html betehess 15:14:28 title: Linked Enterprise Data Patterns Workshop - day 2 15:14:31 RRSAgent, generate minutes 15:14:31 I have made the request to generate http://www.w3.org/2011/12/07-ledp-minutes.html betehess 15:15:13 chair: Eric Prud'hommeaux 15:16:19 s/title: /meeting: / 15:16:22 RRSAgent, generate minutes 15:16:22 I have made the request to generate http://www.w3.org/2011/12/07-ledp-minutes.html betehess 15:17:08 Zakim has left #ledp 15:23:30 [note: DS will stand for David Schaengold, RevelytiX] 15:23:44 s/RevelytiX]/Revelytix]/ 15:23:56 DS: starting a background on revelytix 15:24:03 Topic: Validation of Distributed Enterprise Data is Necessary, and RIF Can Help 15:24:03 ... we're in DC area 15:24:14 http://www.w3.org/2011/09/LinkedData/ledp2011_submission_16.pdf 15:24:25 ... we're profitable 15:24:31 by David Schaengold (Revelytix) 15:24:34 ... around 35 people 15:24:53 ... will explain our software architecture 15:25:14 ... we have different kind of stores 15:25:22 ... it's not exactly a semweb architecture 15:25:30 ... it's more like data integration 15:25:48 ... we have rdbms, flat files, sometimes triple stores, etc. 15:26:06 ericP has changed the topic to: http://www.w3.org/2011/09/LinkedData/ledp2011_submission_16.pdf 15:26:23 ... we have spider, an implementation of r2rml 15:27:00 ... and then a module for sparql query federation 15:27:25 ... r2rml in an emerging standards (at w3c) 15:27:46 ... relation to rdf ML 15:28:14 Ora: how do you tweack the sql for your database? 15:28:45 ... speaking about case where databases can't do some kind of joins 15:28:56 ashok: you have full power of sql 15:29:32 Ora: let me ask again: there are some stuff you can't do with SQL, like @@@ 15:29:38 "hide" ??? 15:29:57 (something which "can't do self-join") 15:30:17 DS: we can take advantages of indexes, and other things 15:30:42 davidwood has joined #ledp 15:30:53 ... the result set is embedded into triples, that can be given to spider 15:31:16 ... spdier does optimizations during the federation 15:31:27 ... we intend to support virtualized queries 15:31:49 ... related to enterprise ontologies 15:32:40 ... we love sparql, but http is kind of slow 15:32:47 ... Sherpa is an adapter that is faster 15:33:08 ... between sparql federation and spider 15:33:17 Zakim has joined #ledp 15:33:24 Zakim, this is ledp 15:33:24 ok, ericP; that matches SW_(LEDP)7:30AM 15:33:29 ... so, we've run into problems 15:33:46 ... for example at the integration level 15:34:07 ... for example when you unify IDs among different stores 15:34:27 ... so we thought using owl or sparql 15:34:47 ... but they have different modelization capabilities 15:35:14 [DS giving examples of limits of sparql/owl] 15:35:44 DS: we need *one* language 15:35:49 ... for the set of rules 15:36:00 ... we're using RIF 15:36:30 [ enumerating key points for RIF from the slides ] 15:36:51 DS: RIF has 2 dialects 15:38:07 ... SPARQL is not a rule language 15:38:13 ... so we don't validation with it 15:38:44 ... there is no distinction between sparql query and sparql rule 15:38:48 ... too confusing 15:39:38 earlier, the database I was referring to was "Hive" (for Hadoop) 15:39:52 s/like @@@/like Hive/ 15:40:12 ora: is it even considered as SQL-compliant? 15:40:31 s/ora: is/ora, is/ 15:41:20 no, it isn't, it is "SQL-like" 15:41:23 DS: just by clicking a button, we can know which source is concerned 15:41:24 Permantn link to my slies: http://www.slideshare.net/LeeFeigenbaum/data-segmenting-in-anzo 15:41:35 I am curious to see RIF used as a 'rule language' as opposed to a 'rule interchange format' 15:41:42 s/Permantn/Permanent 15:41:54 ... triples are frames, will give an example later 15:42:22 ... here is the example 15:42:32 ... [see slides for the query] 15:42:50 ... it's a disjoint property written in owl 15:43:09 ... if you know rdf, it's easy to write 15:43:30 sandro: it's not the interchange format 15:43:36 ... it's a convenient syntax 15:43:47 davidwood, Arnaud, ashok, et al, proposals for breaks? strawman: http://www.w3.org/2011/09/LinkedData/Report 15:43:57 timbl: but these people wrote a parser for that 15:44:20 sandro: there is some history there, we chose XML and RDF for the exchange format 15:44:48 DS: problem: we don't have NOT 15:45:03 ... I mean not generic negation 15:45:09 ... but we needed it 15:45:32 ... sometimes, we can use built-in predicates 15:45:45 ... if rules are too complex, you need it 15:46:08 ... our solution: implement Not() with rif:error() 15:46:54 ... we anticipate there will be a profusion of zero argument predicate in the future 15:47:25 ericp, We should have a breakout regarding the definition of a minimal RDF/REST server. 15:47:46 sandro: rif:error is not part of RIF core 15:48:03 DS: but we use it a lot, we'll be happy to see it standardized 15:48:34 ... the rules have labels in documents, we see it in the output 15:48:52 ... so we can show you all the uris that triggered an error 15:49:10 ... there is no inference here 15:49:29 timbl: and there are tagged as strings? 15:49:31 ericp, ...and another breakout regarding pattern collection (I think). Let's poll immediately following David S's talk. 15:49:33 DS: not sure 15:49:43 sandro: this is addressed by RDF 1.1 15:50:06 DS: we have entailments as a bonus 15:50:23 cmatheus has joined #ledp 15:50:50 ... we don't use not() here 15:51:06 ... we do entailment before validation 15:51:19 RRSAgent, draft minutes 15:51:19 I have made the request to generate http://www.w3.org/2011/12/07-ledp-minutes.html betehess 15:51:54 ... you can try it out 15:52:01 ... download spyder, it's free 15:52:17 ... not currently open-source, but we're working on it (maybe next year) 15:52:30 ... spinner is the federated query engine 15:52:44 ... and Rex is the RIF-implementation 15:53:12 @@@: was told they would open-source it before semtech in June 15:53:18 ... not sure what that meant 15:53:39 ... looked on the Web right now, there is still a license agreement 15:54:14 ... so maybe there is a misunderstanding about free and open 15:54:26 DS: we want to make the source available 15:54:33 ... I don't know about the license 15:54:42 ... just want people to be able to use it 15:55:22 s/@@@: was/davidwood: was/ 15:55:37 davidwood: maybe it's just a communication issue 15:56:01 DS: there is code overlap between spyder and spinner 15:56:14 ... but you don't see it operationally 15:56:30 Sumalaika has joined #ledp 15:57:04 DS: we haven't implementation the sending of email from the rules yet 15:57:09 DS: thanks 15:57:17 RRSAgent, draft minutes 15:57:17 I have made the request to generate http://www.w3.org/2011/12/07-ledp-minutes.html betehess 15:57:33 davidwood: what will be the breakout sessions? 15:57:42 ... trying to do that before eric comes back 15:58:56 http://www.w3.org/2011/09/LinkedData/Report 15:59:04 davidschaengold has joined #ledp 15:59:08 RRSAgent, draft minutes 15:59:08 I have made the request to generate http://www.w3.org/2011/12/07-ledp-minutes.html betehess 15:59:31 Topic: Breakout discussions 15:59:35 [people trying to organize the breakout sessions] 16:00:37 i/by David Booth/scribenick: SteveBattle 16:00:43 rrsagent, draft minutes 16:00:43 I have made the request to generate http://www.w3.org/2011/12/07-ledp-minutes.html dbooth 16:02:16 i/EricP begins/Topic: Web Access Control 16:06:37 Martin: (1) LDEP Basic Profile / 101 Curriculum. Self-contained set. 16:08:08 -??P19 16:08:37 rrsagent, draft minutes 16:08:37 I have made the request to generate http://www.w3.org/2011/12/07-ledp-minutes.html dbooth 16:09:29 so topics (1) Missing specifications (2) Education (3) ROI 16:13:03 (A) Missing Specifications (B) Education and ROI 16:13:17 sandro: How about we each breakout makes a list of deliverables? 16:16:46 [trying to identity the deliverables a group would create] 16:17:14 timbl: I see 3 deliverables 16:17:24 ... spec of the basic profile 16:17:30 ... which is not best practice 16:17:49 ... then tutorial for developers new to this world 16:18:09 ... and a whiter paper for management 16:18:59 @elsevier@: the tutorial is a set of resources to get started 16:19:13 davidwood: it's already on the web 16:19:21 people: there are too many resources 16:19:59 @elsevier@: the thing is that you still have to make choice 16:20:22 ... it's different from providing a minimal profile to achieve something 16:20:58 ... we have big players in this room, that want to do linked data 16:21:20 ... we can reduce freedom to make easier to start 16:21:45 timbl: pagination is a feature part of the spec 16:22:36 here is the Linked Data Patterns text: linkeddatabook.com/ 16:22:40 http://linkeddatabook.com/ 16:22:50 Cornelia: profile is basically a subset of things that already exist. I see a Linked Data pattern spec being missing. trying to solve the identity for example 16:23:08 http://willyou.typewith.me/p/eldw 16:23:14 Arnaud: they are not always subsets 16:23:27 ... it can be about combining specs 16:23:34 ... I'm ok to choose another name 16:23:45 ... doesn't have to be totally new 16:23:47 ericP, http://willyou.typewith.me/p/eldw 16:23:52 ... it's more about putting things together 16:24:38 s/@elsevier@/brad/ 16:24:42 There are topics that come up over and over again in LD prpojects 16:24:51 They need specific attention 16:25:00 e.g., URL opqueness 16:25:09 Resource Identity 16:25:20 these topics need attention urgently 16:25:22 @@: depending on your community, you starting point is different 16:25:37 ... so we need to identify the enterprise developer we want to reach 16:25:58 ... for example, take the pet store from the jee community 16:26:15