Tutorial - Part IV Applications Serena Villata

Transcription

1 Tutorial - Part IV Applications Serena Villata INRIA Sophia Antipolis, France

2 Licenses in the Web of Data the absence of clarity for data consumers about the terms under which they can reuse a particular dataset, and the absence of common guidelines for data licensing, are likely to hinder use and reuse of data Heath and Bizer, Linked Data: Evolving the Web into a Global Data Space, 2011

3 Licenses in the Web of Data Support for generating RDF licenses Share-Alike statements Licenses compatibility and composition Open challenges

4 Support for generating RDF odrl: l4lod: : :licogl a odrl:set; odrl:permission [ a odrl:permission; odrl:action odrl:distribute; odrl:action odrl:derive; odrl:action odrl:commercialize ] ; odrl:duty [ a odrl:duty; odrl:action odrl:attribute; odrl:action odrl:attachpolicy ]. RESEARCH QUESTION How to support users in defining RDF licenses from natural language ones?

5 Main features 1 RDF representation of licenses - CCRel and ODRL vocabularies, 2 Classification problem in supervised learning - Support Vector Machines, 3 Online service: NLL2RDF (Natural Language License to RDF)

6 Main features 1 RDF representation of licenses - CCRel and ODRL vocabularies, 2 Classification problem in supervised learning - Support Vector Machines, 3 Online service: NLL2RDF (Natural Language License to RDF)

7 Main features 1 RDF representation of licenses - CCRel and ODRL vocabularies, 2 Classification problem in supervised learning - Support Vector Machines, 3 Online service: NLL2RDF (Natural Language License to RDF)

8 Synopsis of the overall framework PREPROCESSING MODULE CLASSIFICATION MODULE TOKENIZATION SVM NATURAL LANGUAGE LICENSES TEXTS LEMMATIZATION PoS TAGGING RDF LICENSES GENERATION MODULE NLL2RDF RDF LICENSES SPECIFICATION

9 NLL2RDF - online demo Test it!

10 Share-Alike statements Goal: model licenses as part of the data to enable easy exchange and automated processing Solution: new policy modelling language to manage Share-Alike statements

11 Model of provenance information subclass of Usage haspurpose Purpose property arrow start: domain arrow end: range wastriggeredby Process used Artefact wasgeneratedby Derivation Policy haspolicy

12 Modeling licenses in OWL DL Public Domain License PD : Usage Derivation. CC Attribution BY : (Usage wastriggeredby.attribution) (Derivation wasgeneratedby 1. haspolicy. containedin.{by }).

13 Modeling licenses in OWL DL CC Attribution-NoDerivs BY ND : C BY C ND. CC Share-Alike BY SA : C BY wasgeneratedby 1. haspolicy.( containedin.{by SA} containedin 1.{BY SA}).

14 Licenses compatibility and composition QUERY RESULT Open Government License QUERY RESULT QUERY RESULT Open Database License CC BY-NC-ND License?????? What is the license associated to the query result? RESEARCH QUESTIONS 1. How to compose in a compliant way the licensing terms to produce a single composite license? 2. How to produce in an automated way the composite license adopting different composition heuristics?

15 Main features 1 Combination of Semantic Web languages (machine-readable licenses) - defeasible deontic logic, 2 Extension of existing proposals for licenses compatibility and composition in service license analysis and CC licenses, 3 Heuristics for licenses combination.

16 Main features 1 Combination of Semantic Web languages (machine-readable licenses) - defeasible deontic logic, 2 Extension of existing proposals for licenses compatibility and composition in service license analysis and CC licenses, 3 Heuristics for licenses combination.

17 Main features 1 Combination of Semantic Web languages (machine-readable licenses) - defeasible deontic logic, 2 Extension of existing proposals for licenses compatibility and composition in service license analysis and CC licenses, 3 Heuristics for licenses combination.

18 Synopsis of the overall framework LICENSES COMPATIBILITY AND COMPOSITION MODULE CLIENT QUERY SELECT... WHERE{...} LICENSES SELECTION COMPATIBILITY and COMPLIANCE EVALUATION SPARQL QUERY RESULT XML + <link URI-Lc> LICENSES COMPOSITION CLIENT QUERY QUERY RESULT

19 The formal language Represent, and reason about two components: 1 describe ontology of concepts involved in LOD licenses, 2 capture the deontic component of those licenses. Rule-based language, Ontology rules: regular defeasible logic rules for deriving plain literals, a 1,..., a n l 1 c b support the conclusion of b, given a 1,..., a n, Logic of deontic rules: constructive account of basic deontic modalities (obligation, prohibition, permission), a, Ob l 2 O p: if a is the case and b is obligatory, then Op holds in license l 2.

20 The formal language Represent, and reason about two components: 1 describe ontology of concepts involved in LOD licenses, 2 capture the deontic component of those licenses. Rule-based language, Ontology rules: regular defeasible logic rules for deriving plain literals, a 1,..., a n l 1 c b support the conclusion of b, given a 1,..., a n, Logic of deontic rules: constructive account of basic deontic modalities (obligation, prohibition, permission), a, Ob l 2 O p: if a is the case and b is obligatory, then Op holds in license l 2.

21 The formal language Represent, and reason about two components: 1 describe ontology of concepts involved in LOD licenses, 2 capture the deontic component of those licenses. Rule-based language, Ontology rules: regular defeasible logic rules for deriving plain literals, a 1,..., a n l 1 c b support the conclusion of b, given a 1,..., a n, Logic of deontic rules: constructive account of basic deontic modalities (obligation, prohibition, permission), a, Ob l 2 O p: if a is the case and b is obligatory, then Op holds in license l 2.

22 The formal language Represent, and reason about two components: 1 describe ontology of concepts involved in LOD licenses, 2 capture the deontic component of those licenses. Rule-based language, Ontology rules: regular defeasible logic rules for deriving plain literals, a 1,..., a n l 1 c b support the conclusion of b, given a 1,..., a n, Logic of deontic rules: constructive account of basic deontic modalities (obligation, prohibition, permission), a, Ob l 2 O p: if a is the case and b is obligatory, then Op holds in license l 2.

23 Composition heuristics OR-composition: if at least one of the licenses involved in the composition owns a clause, then also l c owns it; AND-composition: if all the licenses involved in the composition own a clause, then also l c owns it;

24 Composition heuristics OR-composition: if at least one of the licenses involved in the composition owns a clause, then also l c owns it; AND-composition: if all the licenses involved in the composition own a clause, then also l c owns it;

25 Proof theory Combining licenses, Checking their compatibility, Establishing ontology and deontic conclusions which can be drawn from the composite license, i.e., if l c = l 1 l n obtained from l 1,..., l n then conclusions derived in the logic are those that hold in the perspective of l c. Proof theory: Positive definite provability in the paper

26 Proof theory Combining licenses, Checking their compatibility, Establishing ontology and deontic conclusions which can be drawn from the composite license, i.e., if l c = l 1 l n obtained from l 1,..., l n then conclusions derived in the logic are those that hold in the perspective of l c. Proof theory: Positive definite provability in the paper

27 Proof theory: Positive defeasible provability Defeasible provability (+ Mlc p): M l p is a fact; or there is an applicable strict or defeasible rule r in R x for M l p and, for every rule s in R y for M l p, either s discarded or r is weaker than an applicable strict or defeasible rule t in R x for M l p. OR-composition: R x = R y is the union set of all rules of all licenses in the composition AND-composition: R x consists of all rules shared by all licenses in the composition and R y is the union set of all rules of all licenses in the composition.

28 Proof theory: Positive defeasible provability Defeasible provability (+ Mlc p): M l p is a fact; or there is an applicable strict or defeasible rule r in R x for M l p and, for every rule s in R y for M l p, either s discarded or r is weaker than an applicable strict or defeasible rule t in R x for M l p. OR-composition: R x = R y is the union set of all rules of all licenses in the composition AND-composition: R x consists of all rules shared by all licenses in the composition and R y is the union set of all rules of all licenses in the composition.

29 Proof theory: Positive defeasible provability Defeasible provability (+ Mlc p): M l p is a fact; or there is an applicable strict or defeasible rule r in R x for M l p and, for every rule s in R y for M l p, either s discarded or r is weaker than an applicable strict or defeasible rule t in R x for M l p. OR-composition: R x = R y is the union set of all rules of all licenses in the composition AND-composition: R x consists of all rules shared by all licenses in the composition and R y is the union set of all rules of all licenses in the composition.

30 Proof theory: Positive defeasible provability Defeasible provability (+ Mlc p): M l p is a fact; or there is an applicable strict or defeasible rule r in R x for M l p and, for every rule s in R y for M l p, either s discarded or r is weaker than an applicable strict or defeasible rule t in R x for M l p. OR-composition: R x = R y is the union set of all rules of all licenses in the composition AND-composition: R x consists of all rules shared by all licenses in the composition and R y is the union set of all rules of all licenses in the composition.

31 Proof theory: Positive defeasible provability Defeasible provability (+ Mlc p): M l p is a fact; or there is an applicable strict or defeasible rule r in R x for M l p and, for every rule s in R y for M l p, either s discarded or r is weaker than an applicable strict or defeasible rule t in R x for M l p. OR-composition: R x = R y is the union set of all rules of all licenses in the composition AND-composition: R x consists of all rules shared by all licenses in the composition and R y is the union set of all rules of all licenses in the composition.

32 Proof theory: Positive defeasible provability Defeasible provability (+ Mlc p): M l p is a fact; or there is an applicable strict or defeasible rule r in R x for M l p and, for every rule s in R y for M l p, either s discarded or r is weaker than an applicable strict or defeasible rule t in R x for M l p. OR-composition: R x = R y is the union set of all rules of all licenses in the composition AND-composition: R x consists of all rules shared by all licenses in the composition and R y is the union set of all rules of all licenses in the composition.

33 Proof theory: Positive defeasible provability Defeasible provability (+ Mlc p): M l p is a fact; or there is an applicable strict or defeasible rule r in R x for M l p and, for every rule s in R y for M l p, either s discarded or r is weaker than an applicable strict or defeasible rule t in R x for M l p. OR-composition: R x = R y is the union set of all rules of all licenses in the composition AND-composition: R x consists of all rules shared by all licenses in the composition and R y is the union set of all rules of all licenses in the composition.

34 Example: l 1 and l 2 composition L = {l 1, l 2 } R Ol1 = {r 1 : l 1 O Attribution, r 2 : l 1 O Commercial} R Ol2 = {r 3 : l 2 O Commercial, r 4 : l 2 O ShareAlike, r 5 : l 2 O Derivative} OR heuristics for obligations AND heuristics for permissions + Olc Attribution, + Olc ShareAlike, and + Plc Derivative

35 Example: l 1 and l 2 composition L = {l 1, l 2 } R Ol1 = {r 1 : l 1 O Attribution, r 2 : l 1 O Commercial} R Ol2 = {r 3 : l 2 O Commercial, r 4 : l 2 O ShareAlike, r 5 : l 2 O Derivative} OR heuristics for obligations AND heuristics for permissions + Olc Attribution, + Olc ShareAlike, and + Plc Derivative

36 Evaluation: SPINDle (logic defeasible reasoner)

37 Real life example from the logic to SPINdle F = {Open} L = {l OGL, l ODbL, l BY NC ND } OlOGL R = {r 1 : l OGL O Attribution, r 2 : Open l OGL O Publishing, r 3 : Open l OGL O Distribution, r 4 : Open l OGL O Derivative, r 5 : Open l OGL O Commercial} OlODbL R = {r 6 : l ODbL O ShareAlike, r 7 : l ODbL O Attribution, R Ol BY NC ND r 8 : l ODbL O Sharing, r 9 : l ODbL O Derivative} = {r 10 : l BY NC ND O Attribution, r 11 : l BY NC ND O Commercial, r 12 : l BY NC ND O Derivative, r 13 : l BY NC ND O Sharing} = {l ODbL l BY NC ND }

38 Real life example from the logic to SPINdle >> Open r1: =>[Oc]Attribution r2: Open =>[-Oc] -Publishing r3: Open =>[-Oc] -Distribution r4: Open =>[-Oc] -Derivative r5: Open =>[-Oc] -CommercialExpl r6: =>[Oc] ShareAlike r7: =>[Oc] Attribution r8: =>[-Oc] -Share r9: =>[-Oc] -Derivative r10: =>[Oc] Attribution r11: =>[Oc] -CommercialExpl r12: =>[Oc] -Derivative r13: =>[-Oc] -Share r9 > r12

39 Real life example from SPINdle to RDF AND-composition + Olc Attribution OR-composition is admissible: conflict between r 5 and r 11, and between rule r 12 and rules r 4 and r 9 Deontic conclusions: + Olc Attribution, + Olc ShareAlike, + Plc Publishing, + Plc Distribution, + Plc Sharing, Plc Derivative, Plc Commercial SPINdle it takes 14 milliseconds to produce the following conclusions +d [Oc]Attribution, +d [-Oc]-Distribution, +d [-Oc]-Publishing, +d [-Oc]-Share, +d [Oc]ShareAlike

40 Real life example from SPINdle to RDF AND-composition + Olc Attribution OR-composition is admissible: conflict between r 5 and r 11, and between rule r 12 and rules r 4 and r 9 Deontic conclusions: + Olc Attribution, + Olc ShareAlike, + Plc Publishing, + Plc Distribution, + Plc Sharing, Plc Derivative, Plc Commercial SPINdle it takes 14 milliseconds to produce the following conclusions +d [Oc]Attribution, +d [-Oc]-Distribution, +d [-Oc]-Publishing, +d [-Oc]-Share, +d [Oc]ShareAlike

41 Real life example from SPINdle to RDF AND-composition + Olc Attribution OR-composition is admissible: conflict between r 5 and r 11, and between rule r 12 and rules r 4 and r 9 Deontic conclusions: + Olc Attribution, + Olc ShareAlike, + Plc Publishing, + Plc Distribution, + Plc Sharing, Plc Derivative, Plc Commercial SPINdle it takes 14 milliseconds to produce the following conclusions +d [Oc]Attribution, +d [-Oc]-Distribution, +d [-Oc]-Publishing, +d [-Oc]-Share, +d [Oc]ShareAlike

42 Real life example from SPINdle to RDF AND-composition + Olc Attribution OR-composition is admissible: conflict between r 5 and r 11, and between rule r 12 and rules r 4 and r 9 Deontic conclusions: + Olc Attribution, + Olc ShareAlike, + Plc Publishing, + Plc Distribution, + Plc Sharing, Plc Derivative, Plc Commercial SPINdle it takes 14 milliseconds to produce the following conclusions +d [Oc]Attribution, +d [-Oc]-Distribution, +d [-Oc]-Publishing, +d [-Oc]-Share, +d [Oc]ShareAlike

43 Real life example from SPINdle to RDF SPINdle it takes 14 milliseconds to produce the following conclusions +d [Oc]Attribution, +d [-Oc]-Distribution, +d [-Oc]-Publishing, +d [-Oc]-Share, +d l4lod: : :licc a l4lod:license; l4lod:obliges l4lod:attribution; l4lod:obliges l4lod:sharealike; l4lod:permits l4lod:publishing; l4lod:permits l4lod:distribution; l4lod:permits l4lod:sharing.

44 Real life example from SPINdle to RDF SPINdle it takes 14 milliseconds to produce the following conclusions +d [Oc]Attribution, +d [-Oc]-Distribution, +d [-Oc]-Publishing, +d [-Oc]-Share, +d l4lod: : :licc a l4lod:license; l4lod:obliges l4lod:attribution; l4lod:obliges l4lod:sharealike; l4lod:permits l4lod:publishing; l4lod:permits l4lod:distribution; l4lod:permits l4lod:sharing.

45 1 Enlarge set of composition heuristics: quantitative ones and Constraining-value 2 Data obtained by inference from one or several licensed datasets, i.e., queries going beyond basic SELECT queries, where aggregations are present, e.g., average, sum 3 Temporal terms of the licenses 4 Licensing vocabularies: meaning, implications, statistics.

46 1 Enlarge set of composition heuristics: quantitative ones and Constraining-value 2 Data obtained by inference from one or several licensed datasets, i.e., queries going beyond basic SELECT queries, where aggregations are present, e.g., average, sum 3 Temporal terms of the licenses 4 Licensing vocabularies: meaning, implications, statistics.

47 1 Enlarge set of composition heuristics: quantitative ones and Constraining-value 2 Data obtained by inference from one or several licensed datasets, i.e., queries going beyond basic SELECT queries, where aggregations are present, e.g., average, sum 3 Temporal terms of the licenses 4 Licensing vocabularies: meaning, implications, statistics.

48 1 Enlarge set of composition heuristics: quantitative ones and Constraining-value 2 Data obtained by inference from one or several licensed datasets, i.e., queries going beyond basic SELECT queries, where aggregations are present, e.g., average, sum 3 Temporal terms of the licenses 4 Licensing vocabularies: meaning, implications, statistics.

49 Thanks for your attention!