Minimal Deductive Systems for RDF

Transcription

1 Minimal Deductive Systems for RDF Sergio Muñoz, Jorge Pérez, Claudio Gutierrez Department of Computer Science Universidad de Chile Pontificia Universidad Católica de Chile Center for Web Research European Semantic Web Conference 2007 S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 1 / 24

2 RDF Language and Model Theory looks a little bit complicated at first sight... S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 2 / 24

3 RDF Language and Model Theory looks a little bit complicated at first sight... S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 2 / 24

4 RDF Language and Model Theory looks a little bit complicated at first sight... S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 2 / 24

5 RDF Language and Model Theory looks a little bit complicated at first sight... S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 2 / 24

6 RDF Language and Model Theory looks a little bit complicated at first sight... S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 2 / 24

7 Several dozens of reserved keywords... rdfs:resource [res] rdf:type [type] rdfs:isdefinedby [isdefined] rdf:property [prop] rdfs:domain [dom] rdfs:comment [comment] rdfs:class [class] rdfs:range [range] rdfs:label [label] rdfs:literal [literal] rdfs:subclassof [sc] rdf:value [value] rdfs:datatype [datatype] rdfs:subpropertyof [] rdf:nil [nil] rdf:xmlliteral [xmllit] rdf:subject [subj] rdf: 1 [ 1] rdfs:container [cont] rdf:predicate [pred] rdf: 2 [ 2] rdf:statement [stat] rdf:object [obj]... rdf:list [list] rdfs:member [member] rdf: i [ i] rdf:alt [alt] rdf:first [first]... rdf:bag [bag] rdf:rest [rest] rdf:seq [seq] rdfs:seealso [seealso] rdfs:containermembershipproperty [contmp] S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 3 / 24

8 ...plus axiomatic triples... (type, type, prop) (subj, type, prop) (pred, type, prop) (obj, type, prop) (first, type, prop) (rest, type, prop) (value, type, prop) ( 1, type, prop) ( 1, type, contmp) ( 2, type, prop) ( 2, type, contmp)... ( i, type, prop) ( i, type, contmp)... (nil, type, prop) (xmllit, type, datatype) S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 4 / 24

9 ...plus more axiomatic triples... (type, dom, res) (dom, dom, prop) (range, dom, prop) (, dom, prop) (sc, dom, class) (subj, dom, stat) (pred, dom, stat) (obj, dom, stat) (member, dom, res) (first, dom, list) (rest, dom, list) (seealso, dom, res) (isdefined, dom, res) (comment, dom, res) (label, dom, res) (value, dom, res) ( 1, dom, res) ( 2, dom, res)... ( i, dom, res)... (type, range, class) (dom, range, class) (range, range, class) (, range, prop) (sc, range, class) (subj, range, res) (pred, range, res) (obj, range, res) (member, range, res) (first, range, res) (rest, range, list) (seealso, range, res) (isdefined, range, res) (comment, range, literal) (label, range, literal) (value, range, res) ( 1, range, res) ( 2, range, res)... ( i, range, res)... S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 5 / 24

10 ...plus more axiomatic triples... (alt, sc, cont) (bag, sc, cont) (seq, sc, cont) (contmp, sc, prop) (xmllit, sc, literal) (datatype, sc, class) (isdefined,, seealso) S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 6 / 24

11 ...and on top of this a (slightly) non-standard model theory A notion of interpretation (Res, Prop, Class, Lit, PExt, CExt, Int) including subsets of the universe denoting properties, classes and literals, and mapping defining their extensions. Notion of interpretation of blank nodes Definition of reflexivity, transitivity and semi-extensionality of subclass and subproperty Typing restrictions S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 7 / 24

12 But...we need a workable language to bring to reality the vision of the Semantic Web Would like to: Have a simple user-language to be able to popularize RDF among Web users. Have a simple ecification to allow sound development work. Have a language in streamlined form to make it easy to formalize and prove results about its properties. S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 8 / 24

13 What is to be done?: To simplify the language S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 9 / 24

14 What is to be done?: To simplify the language There is a minimal fragment of the theory preserving the essential core of RDFS S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 9 / 24

15 What is to be done?: To simplify the language There is a minimal fragment of the theory preserving the essential core of RDFS Basic idea: Separate user-language from features and constructors which define and ecificy the language. Concentrate in vocabulary with non-trivial semantics. S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 9 / 24

16 Main contributions & Outline Identify a fragment of RDFS that covers the crucial vocabulary and preserves the original RDFS semantics. Study dependencies among vocabulary and develop sound and complete proof systems for each fragment. Present algorithms to modularize reasoning according to relevant vocabulary. S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 10 / 24

17 Main contributions & Outline Identify a fragment of RDFS that covers the crucial vocabulary and preserves the original RDFS semantics. Study dependencies among vocabulary and develop sound and complete proof systems for each fragment. Present algorithms to modularize reasoning according to relevant vocabulary. S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 10 / 24

18 Main contributions & Outline Identify a fragment of RDFS that covers the crucial vocabulary and preserves the original RDFS semantics. Study dependencies among vocabulary and develop sound and complete proof systems for each fragment. Present algorithms to modularize reasoning according to relevant vocabulary. S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 10 / 24

19 The core vocabulary rdfs:resource [res] rdf:type [type] rdfs:isdefinedby [isdefined] rdf:property [prop] rdfs:domain [dom] rdfs:comment [comment] rdfs:class [class] rdfs:range [range] rdfs:label [label] rdfs:literal [literal] rdfs:subclassof [sc] rdf:value [value] rdfs:datatype [datatype] rdfs:subpropertyof [] rdf:nil [nil] rdf:xmlliteral [xmllit] rdf:subject [subj] rdf: 1 [ 1] rdfs:container [cont] rdf:predicate [pred] rdf: 2 [ 2] rdf:statement [stat] rdf:object [obj]... rdf:list [list] rdfs:member [member] rdf: i [ i] rdf:alt [alt] rdf:first [first]... rdf:bag [bag] rdf:rest [rest] rdf:seq [seq] rdfs:seealso [seealso] rdfs:containermembershipproperty [contmp] S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 11 / 24

20 The core vocabulary rdfs:resource [res] rdf:type [type] rdfs:isdefinedby [isdefined] rdf:property [prop] rdfs:domain [dom] rdfs:comment [comment] rdfs:class [class] rdfs:range [range] rdfs:label [label] rdfs:literal [literal] rdfs:subclassof [sc] rdf:value [value] rdfs:datatype [datatype] rdfs:subpropertyof [] rdf:nil [nil] rdf:xmlliteral [xmllit] rdf:subject [subj] rdf: 1 [ 1] rdfs:container [cont] rdf:predicate [pred] rdf: 2 [ 2] rdf:statement [stat] rdf:object [obj]... rdf:list [list] rdfs:member [member] rdf: i [ i] rdf:alt [alt] rdf:first [first]... rdf:bag [bag] rdf:rest [rest] rdf:seq [seq] rdfs:seealso [seealso] rdfs:containermembershipproperty [contmp] S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 11 / 24

21 The core vocabulary rdfs:resource [res] rdf:type [type] rdfs:isdefinedby [isdefined] rdf:property [prop] rdfs:domain [dom] rdfs:comment [comment] rdfs:class [class] rdfs:range [range] rdfs:label [label] rdfs:literal [literal] rdfs:subclassof [sc] rdf:value [value] rdfs:datatype [datatype] rdfs:subpropertyof [] rdf:nil [nil] rdf:xmlliteral [xmllit] rdf:subject [subj] rdf: 1 [ 1] rdfs:container [cont] rdf:predicate [pred] rdf: 2 [ 2] rdf:statement [stat] rdf:object [obj]... rdf:list [list] rdfs:member [member] rdf: i [ i] rdf:alt [alt] rdf:first [first]... rdf:bag [bag] rdf:rest [rest] rdf:seq [seq] rdfs:seealso [seealso] rdfs:containermembershipproperty [contmp] ρdf = {,sc,type,dom,range} S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 11 / 24

22 Blank node rule G H if there is a homomorphism µ : H G S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 12 / 24

23 Core rules Subproperty (transitivity, definition) (A,,B) (B,,C) (A,,C) (A,,B) (X,A,Y) (X,B,Y) Subclass (transitivity, definition) (A,sc,B) (B,sc,C) (A,sc,C) (A,sc,B) (X,type,A) (X,type,B) Typing (domain, range) (A,dom,B) (X,A,Y) (X,type,B) (A,range,B) (X,A,Y) (Y,type,B) Implicit Typing (strange case...) (A,dom,B) (C,,A) (X,C,Y) (X,type,B) (A,range,B) (C,,A) (X,C,Y) (Y,type,B) S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 13 / 24

24 Core rules Subproperty (transitivity, definition) (A,,B) (B,,C) (A,,C) (A,,B) (X,A,Y) (X,B,Y) Subclass (transitivity, definition) (A,sc,B) (B,sc,C) (A,sc,C) (A,sc,B) (X,type,A) (X,type,B) Typing (domain, range) (A,dom,B) (X,A,Y) (X,type,B) (A,range,B) (X,A,Y) (Y,type,B) Implicit Typing (strange case...) (A,dom,B) (C,,A) (X,C,Y) (X,type,B) (A,range,B) (C,,A) (X,C,Y) (Y,type,B) S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 13 / 24

25 Core rules Subproperty (transitivity, definition) (A,,B) (B,,C) (A,,C) (A,,B) (X,A,Y) (X,B,Y) Subclass (transitivity, definition) (A,sc,B) (B,sc,C) (A,sc,C) (A,sc,B) (X,type,A) (X,type,B) Typing (domain, range) (A,dom,B) (X,A,Y) (X,type,B) (A,range,B) (X,A,Y) (Y,type,B) Implicit Typing (strange case...) (A,dom,B) (C,,A) (X,C,Y) (X,type,B) (A,range,B) (C,,A) (X,C,Y) (Y,type,B) S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 13 / 24

26 Core rules Subproperty (transitivity, definition) (A,,B) (B,,C) (A,,C) (A,,B) (X,A,Y) (X,B,Y) Subclass (transitivity, definition) (A,sc,B) (B,sc,C) (A,sc,C) (A,sc,B) (X,type,A) (X,type,B) Typing (domain, range) (A,dom,B) (X,A,Y) (X,type,B) (A,range,B) (X,A,Y) (Y,type,B) Implicit Typing (strange case...) (A,dom,B) (C,,A) (X,C,Y) (X,type,B) (A,range,B) (C,,A) (X,C,Y) (Y,type,B) S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 13 / 24

27 Core rules Subproperty (transitivity, definition) (A,,B) (B,,C) (A,,C) (A,,B) (X,A,Y) (X,B,Y) Subclass (transitivity, definition) (A,sc,B) (B,sc,C) (A,sc,C) (A,sc,B) (X,type,A) (X,type,B) Typing (domain, range) (A,dom,B) (X,A,Y) (X,type,B) (A,range,B) (X,A,Y) (Y,type,B) Implicit Typing (strange case...) (A,dom,B) (C,,A) (X,C,Y) (X,type,B) (A,range,B) (C,,A) (X,C,Y) (Y,type,B) S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 13 / 24

28 Reflexivity rules If (A, type, Property) then (A,, A) Subproperty Reflexivity (X,A,Y) (A,,A) (A,,B) (A,,A) (B,,B) (p,,p) (A,dom,X) (A,,A) for p ρdf (A,range,X) (A,,A) Subclass Reflexivity (A,sc,B) (A,sc,A) (B,sc,B) (X,dom,A) (A,sc,A) (X,range,A) (A,sc,A) (X,type,A) (A,sc,A) S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 14 / 24

29 Reflexivity rules If (A, type, Property) then (A,, A) Subproperty Reflexivity (X,A,Y) (A,,A) (A,,B) (A,,A) (B,,B) (p,,p) (A,dom,X) (A,,A) for p ρdf (A,range,X) (A,,A) Subclass Reflexivity (A,sc,B) (A,sc,A) (B,sc,B) (X,dom,A) (A,sc,A) (X,range,A) (A,sc,A) (X,type,A) (A,sc,A) S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 14 / 24

30 Soundness and Completeness Let = denote the standard RDFS entailment, and ρdf a proof system based on the rules presented. Theorem Let G and H be graphs in ρdf then G = H if and only if G ρdf H. S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 15 / 24

31 Blank Nodes Modularization Blank nodes can be treated in an orthogonal form to ρdf vocabulary. Theorem Let G and H be graphs in ρdf and G = H, then there is a proof of H from G where the blank rule is used at most once and at the end. S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 16 / 24

32 The role of reflexivity The only consequence of reflexivity of and sc in RDFS is the possible entailment of triples of the form (x,,x),(x,sc,x). Theorem Let G and H be ρdf graphs. Assume that H does not contain triples of the form (x,,x) and (x,sc,x). Then, G ρdf H without using reflexivity rules. (Also, by not imposing reflexivity, axiomatic triples can be completely avoided.) S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 17 / 24

33 Dependence diagram among ρdf vocabulary type V 0 sc dom range To determine G = H it is enough to test G = H where G is the subgraph of G which involves only nodes in voc(h) and their dependencies in the diagram. S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 18 / 24

34 It is possible to avoid the closure to test RDFS entailment A naive approach to test G = H is: (pre-)compute the closure of G check if H is contained in the closure of G. Theorem The size of the closure of G is O( G 2 ), and this bound is tight. S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 19 / 24

35 It is possible to avoid the closure to test RDFS entailment A naive approach to test G = H is: (pre-)compute the closure of G check if H is contained in the closure of G. Theorem The size of the closure of G is O( G 2 ), and this bound is tight. Alternative: to use a goal oriented approach based on the dependencies diagram. S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 19 / 24

36 Goal oriented entailment algorithm Does the graph entails (x,d,y)? Look for triples of the form (x,a,y) and -paths from a to d. x a b c d y S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 20 / 24

37 Goal oriented entailment algorithm Does the graph entails (x,d,y)? Look for triples of the form (x,a,y) and -paths from a to d. x a b c d y S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 20 / 24

38 Goal oriented entailment algorithm Does the graph entails (x,d,y)? Look for triples of the form (x,a,y) and -paths from a to d. x a b c d y S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 20 / 24

39 Goal oriented entailment algorithm Does the graph entails (x,d,y)? Look for triples of the form (x,a,y) and -paths from a to d. x a b c d y S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 20 / 24

40 Goal oriented entailment algorithm Does the graph entails (x,d,y)? Look for triples of the form (x,a,y) and -paths from a to d. x a b c d y S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 20 / 24

41 Goal oriented entailment algorithm Does the graph entails (x,d,y)? Look for triples of the form (x,a,y) and -paths from a to d. x a b c d y S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 20 / 24

42 Goal oriented entailment algorithm Does the graph entails (x,d,y)? Look for triples of the form (x,a,y) and -paths from a to d. x a b c d y x a y b e c d S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 20 / 24

43 Goal oriented entailment algorithm Does the graph entails (x,d,y)? Look for triples of the form (x,a,y) and -paths from a to d. x a b c d y x a y b e c d S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 20 / 24

44 Goal oriented entailment algorithm Does the graph entails (x,d,y)? Look for triples of the form (x,a,y) and -paths from a to d. x a b c d y x a y b e c d S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 20 / 24

45 Goal oriented entailment algorithm Let G = {...,(x,p,z),(p,,q),(q,,r),(r,,s),(s,dom,c), (c,sc,d),(d,sc,y),... } Does the graph entails (x,type,y)? x z p q r s dom sc c sc d y S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 21 / 24

46 Goal oriented entailment algorithm Let G = {...,(x,p,z),(p,,q),(q,,r),(r,,s),(s,dom,c), (c,sc,d),(d,sc,y),... } Does the graph entails (x,type,y)? x z p q r s dom sc c sc d y S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 21 / 24

47 Goal oriented entailment algorithm Let G = {...,(x,p,z),(p,,q),(q,,r),(r,,s),(s,dom,c), (c,sc,d),(d,sc,y),... } Does the graph entails (x,type,y)? x z p q r s dom sc c sc d y S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 21 / 24

48 Goal oriented entailment algorithm Let G = {...,(x,p,z),(p,,q),(q,,r),(r,,s),(s,dom,c), (c,sc,d),(d,sc,y),... } Does the graph entails (x,type,y)? x z p q r s dom sc c sc d y S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 21 / 24

49 Goal oriented entailment algorithm Let G = {...,(x,p,z),(p,,q),(q,,r),(r,,s),(s,dom,c), (c,sc,d),(d,sc,y),... } Does the graph entails (x,type,y)? x z p q r s dom sc c sc d y S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 21 / 24

50 Goal oriented entailment algorithm Let G = {...,(x,p,z),(p,,q),(q,,r),(r,,s),(s,dom,c), (c,sc,d),(d,sc,y),... } Does the graph entails (x,type,y)? x z p q r s dom sc c sc d y S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 21 / 24

51 Goal oriented entailment algorithm Let G = {...,(x,p,z),(p,,q),(q,,r),(r,,s),(s,dom,c), (c,sc,d),(d,sc,y),... } Does the graph entails (x,type,y)? x z p q r s dom sc c sc d y S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 21 / 24

52 Goal oriented entailment algorithm Let G = {...,(x,p,z),(p,,q),(q,,r),(r,,s),(s,dom,c), (c,sc,d),(d,sc,y),... } Does the graph entails (x,type,y)? x z p q r s dom sc c sc d y S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 21 / 24

53 Goal oriented entailment algorithm Let G = {...,(x,p,z),(p,,q),(q,,r),(r,,s),(s,dom,c), (c,sc,d),(d,sc,y),... } Does the graph entails (x,type,y)? x z p q r s dom sc c sc d y S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 21 / 24

54 Entailment can be done in O(n logn) time Theorem The goal oriented algorithm takes O( G log G ) time in testing the entailment G = t. Correctness follows by the dependencies diagram. Essentially time G log G in constructing the necessary graph data-structures. Time G in traversing these data-structures. There is no more efficient approach to test G = t! S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 22 / 24

55 Entailment can be done in O(n logn) time Theorem The goal oriented algorithm takes O( G log G ) time in testing the entailment G = t. Correctness follows by the dependencies diagram. Essentially time G log G in constructing the necessary graph data-structures. Time G in traversing these data-structures. There is no more efficient approach to test G = t! S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 22 / 24

56 The O(n log n) bound is tight. Theorem Testing G = t takes time Ω( G log G ). Idea: Coding the set-disjointness problem, which is Ω(n log n) S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 23 / 24

57 Conclusions Core fragment of RDFS well-behaved, representative, and simple. S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 24 / 24

58 Conclusions Core fragment of RDFS well-behaved, representative, and simple. Efficient algorithm to check entailment S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 24 / 24

59 Conclusions Core fragment of RDFS well-behaved, representative, and simple. Efficient algorithm to check entailment Suggestions Add missing rules Eliminate reflexivity Treat bnodes orthogonal to rest S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 24 / 24

60 Conclusions Core fragment of RDFS well-behaved, representative, and simple. Efficient algorithm to check entailment Suggestions Add missing rules Eliminate reflexivity Treat bnodes orthogonal to rest Next: Navigational language based on testing algorithm S. Muñoz, J. Pérez, C. Gutierrez Minimal Deductive Systems for RDF 24 / 24