Open Semantic Revision Control with R43ples Extending SPARQL to access revisions of Named Graphs

Transcription

1 Elektrotechnik & Informationstechnik, Institut für Automatisierungstechnik, Professur für Prozessleittechnik / AG Systemverfahrenstechnik Open Semantic Revision Control with R43ples Extending SPARQL to access revisions of Named Graphs SEMANTICS 2016 Markus Graube Stephan Hensel Leon Urbas

2 Agenda Introduction Concept SPARQL Interface Approaches to retrieve old revisions Implementation and Evaluation Conclusion and Discussion SEMANTICS 2016 Open Semantic Revision Control with R43ples 2

3 ComVantage - Motivation Project-based collaborations enable extended services require flexible, interlinked information models and trustful and convenient access to the information ComVantage Finished EU FP7 Project product-centric collaboration space for dynamic and flexible information exchange build on top of best practices from the Semantic Web for providing productcentric and workflow-based mobile apps SEMANTICS 2016 Open Semantic Revision Control with R43ples 3

4 Revision Control System - Motivation Revision Control (Wiesner, 2010) Record Changes and revise old version Track quality Branching / Merging Top requirement for industrial use (Schmidt, 2014) Challenges for Linked Data Open nature of LD as challenge Complete checkout not possible Approaches Temporal RDF Reified Statements (Auer and Herre, 2007) Aggregated Deltas (Im et. Al., 2012) File based (partly with Git, SVN, ) Problems No extensive use of open semantics No standardized and uniform access SEMANTICS 2016 Open Semantic Revision Control with R43ples 4

5 Concept of R43ples (Revision for Triples) Concept Revision Control on Graph level Based on (Vander Sande et. al., 2013) R43ples as proxy SPARQL Endpoint Generation of old revisions on demand Information space on external triplestore Delta compression in Named Graphs Add sets Delete sets Revision Information in additional graph Revision Management Ontology (RMO) Some revisions as fully materialized graphs Tag, Branch, Master SEMANTICS 2016 Open Semantic Revision Control with R43ples Folie 5

6 Information model Overview about revision graphs stored in eatld:r43ples-revisions Revision history for revisioned graph in named graph Use of PROV-O as base Extension to Revision Management Ontology (RMO) Change sets as deltas between revisions stored in additional named graphs Add-Sets Delete-Sets Terminal revision in a branch is fully materialized Tag, Branch, Master SEMANTICS 2016 Open Semantic Revision Control with R43ples 6

7 Revision Graph SEMANTICS 2016 Open Semantic Revision Control with R43ples Folie 7

8 Example revision graph (according to Van der Sande et. al., 2013) SEMANTICS 2016 Open Semantic Revision Control with R43ples Folie 8

9 Access to revisions Support of SPARQL as popular access mechanism Revision information as part of the query Extension necessary to R43ples queries Specification of revisions for used graphs Transparent for standard SPARQL Queries Compatible to SPARQL 1.1 No specification of Revision => MASTER SEMANTICS 2016 Open Semantic Revision Control with R43ples 9

10 Approaches for SPARQL interface (Revision 2 in Graph Virtual Graph SELECT?s?o WHERE { GRAPH < {?s foaf:knows?o. Virtual Service SELECT?s?o WHERE { GRAPH?graph_rev2 {?s foaf:knows?o. SERVICE <r43ples:revision> {?graph_rev2 a rmo:revision; rmo:revisionof < rmo:revisionnumber "2". Additional Keywords SELECT?s?o WHERE { GRAPH < REVISION "2" {?s foaf:knows?o. Graph URI Function SELECT?s?o WHERE { GRAPH r43ples:revision(< 2) {?s foaf:knows?o. SEMANTICS 2016 Open Semantic Revision Control with R43ples 10

11 SPARQL Interface Decision for Additional Keyword Transformation between approaches possible Multiple Graphs SELECT * WHERE { GRAPH <graph> REVISION 23 {?s?p?o. GRAPH <graph> REVISION branch-new-feature {?s?p?o. SEMANTICS 2016 Open Semantic Revision Control with R43ples Folie 11

12 Querying Revisions Query on a fully materialized graph can be directly performed Leaf revision of a branch, master, tag Otherwise reconstruction of old revision for every used graph Generation of a temporary graph Query Rewriting SEMANTICS 2016 Open Semantic Revision Control with R43ples 12

13 Retrieve old revision via generation of a temporary graph 1. Extract revision information and find shortest path to a materialized graph 2. Semantic description allows simple reconstruction of old revisions Generation from newest revision to older ones Use of SPARQL COPY GRAPH <delete_set> TO <temp> DELETE { GRAPH <temp>{?s?p?o WHERE { GRAPH <add_set {?s?p?o 3. Perform query on this temporary graph SEMANTICS 2016 Open Semantic Revision Control with R43ples 13

14 Rewriting of SPARQL query for joining changesets 1. Extract revision information and find shortest path to a materialized graph 2. Rewrite Query 3. Let SPARQL engine let join all the changesets for us SEMANTICS 2016 Open Semantic Revision Control with R43ples 14

15 Joining Changesets q a f R i : Set of triples in revision i q: Revision to be queried f: Revision with fully materialized graph a: Revision in between (q<a<f) A: Union of all add sets between f and q D: Union of all delete sets between f and q R * f: Non-changed triples between f and q D * : Set of triples from D, which are not in earlier add sets R q = R f D R q = R f \ A i D f+1 R f f i=q+1 f j=q+1 D j \ j A k k=q+1 R q = D j \ j=q+1 k=q+1 SEMANTICS 2016 Open Semantic Revision Control with R43ples 15 f+1 j A k

16 Application of Query Rewriting For each TriplePattern in query Example Graph: < Master Revision: 5 Query on Revision 2: SELECT DISTINCT?s?p?o WHERE { GRAPH < REVISION 2 { # only one TriplePattern?s foaf:knows?o. SEMANTICS 2016 Open Semantic Revision Control with R43ples 16

17 Query Rewriting SELECT?s?o WHERE { R q GRAPH < REVISION "2" { = R f D R q = R f \ D f+1 R f?s foaf:knows?o. f i=q+1 A i f j=q+1 D j \ j k=q+1 A k PREFIX rev: < SELECT?s?o WHERE{ GRAPH?g_st1 {?s foaf:knows?o. GRAPH r43ples:revisions { { BIND(rev:5 AS?r_st1) [] rmo:references?r_st1; UNION { MINUS { rmo:fullgraph?g_st1.?r_st1 rmo:deltaremoved?g_st1. FILTER(?r_st1 IN (rev:5,rev:4,rev:3)) GRAPH?g_st1_add {?s foaf:knows?o. GRAPH r43ples:revisions {?r_st1 prov:wasderivedfrom*?r_st1_add. f = rev:5?r_st1_add rmo:deltaadded?g_st1_add. f+1 j q = rev:2 FILTER(?r_st1_add IN A ->rmo:deltaadded R q = D j \ A (rev:5,rev:4,rev:3)) k D ->rmo:deltaremoved f+1 j=q+1 k=q+1? g del D i SEMANTICS 2016 i=q Open Semantic Revision Control with R43ples 17

18 Updating revisions 1. Extract revision information Only branches can be updated Updates on other branches are rejected 2. INSERT and DELETE parts Inserted into new changesets graphs Applied on materialized graph 3. Updating revision graph Insert rmo:commit and rmo:revision Moving branch reference MESSAGE change dc:title to rdfs:label DELETE{ GRAPH<graph> REVISION master {?s dc:title?title. INSERT { GRAPH<graph> REVISION master {?s rdfs:label?title. WHERE { GRAPH <graph> REVISION v0.1 {?s dc:title?title SEMANTICS 2016 Open Semantic Revision Control with R43ples 18

19 SPARQL Interface Further commands follow same schema TAG GRAPH <graph> REVISION 23 TO v1.4 BRANCH GRAPH <graph> REVISION 23 TO branch-b2 MERGE GRAPH <graph> BRANCH branch-b2 TO master SEMANTICS 2016 Open Semantic Revision Control with R43ples Folie 19

20 Revision informationen for the client Direct access to revision graph via SPARQL Parts of revision graph in response header as RDF [] rdf:type rmo:master, rmo:branch ; rdfs:label "master" ; rmo:references [ rmo:revisionof test:r43ples-dataset-merging ; rmo:revisionnumber "1". ]. [] rdf:type rmo:branch ; rdfs:label "B1" ; rmo:references [ rmo:revisionof test:r43ples-dataset-merging ; rmo:revisionnumber "1.0-1". ]. [] rdf:type rmo:branch ; rdfs:label "B2" ; rmo:references [ rmo:revisionof test:r43ples-dataset-merging ; rmo:revisionnumber "1.1-2" SEMANTICS 2016 Open Semantic Revision Control with R43ples 20 ].

21 SPARQL 1.1. Service Description How do clients know that an endpoint support revision control? SPARQL 1.1. Service Description Extension of sd:feature with sd:r43ples Extension of sd:supportedlanguage with ns3: sd: rdf: < < a sd:service ; sd:endpoint < ; sd:feature sd:r43ples ; sd:resultformat ns3:rdfa, ns3:n3, ns3:turtle, ns3:sparql_results_xml, ns3:sparql_results_csv, ns3:rdf_xml, ns3:n-triples, ns3:sparql_results_json ; sd:supportedlanguage sd:sparql11update, sd:r43plesquery, sd:sparql10query, sd:sparql11query ; sd:url < SEMANTICS 2016 Open Semantic Revision Control with R43ples 21

22 Implementation Java Prototyp Jersey (REST) Grizzly aswebserver Jena ARQ for SPARQL Open Source EUPL licence GitHub: Test System SEMANTICS 2016 Open Semantic Revision Control with R43ples 22

23 Evaluation Response time dependant on dataset size revision path length size of changes Additional Storage overhead 12 triples + changes for update Full graph for tag and branches SEMANTICS 2016 Open Semantic Revision Control with R43ples 23

24 Conclusion Revision Control on Triple level Change sets in named graphs Blank Nodes problem Semantic Description No implicit knowledge necessary SPARQL interface Extendend keywords Two approaches Generation of temporary graph JOIN via Query Rewriting R43ples Endpoint transparent for SPARQL clients Default revision = MASTER Current and Future Work Semantic Merging Interface Concept of semantical diffs Extended evaluation SEMANTICS 2016 Open Semantic Revision Control with R43ples 24

25 Thank you for your attention! Prof. Dr.-Ing. habil. Leon Urbas Contact me: Technische Universität Dresden Fakultät Elektrotechnik und Informationstechnik Institut für Automatisierungstechnik Tel.: Fax: Besucheradresse: Barkhausen-Bau Georg-Schumann-Str Dresden Postanschrift (Briefe): Technische Universität Dresden Fakultät Elektrotechnik und Informationstechnik Institut für Automatisierungstechnik Dresden SEMANTICS 2016 Open Semantic Revision Control with R43ples 25