Processes and products for generation of completion rules of knowledge graphs
Provided is a computer-implemented technology which generates rules for completion of a knowledge graph by producing, with a generic machine learning model or one that is trained on the knowledge graph, inferred triples, optionally refines and filters the produced rules along predefined user settings and provides the resulting rules, along with the inferred facts covered by the rules, as candidates for completion of the knowledge graph.
1. A computer-implemented method for generation of completion rules for a knowledge graph, comprising the following steps:
using a machine learning model to produce a set of inferred triples from RDF data, wherein the model is a generic model or a model trained and/or retrained on the knowledge graph or a subset of the knowledge graph;
generating completion rules with a same functionality as SPARQL queries of the form
INSERT ?subject ?predicate ?object
WHERE {
triple_pattern_1 .
triple_pattern_2 .
...
}
as follows:
adding to completion rules, that would produce triples that are not part of the set, triple patterns that result in the exclusion of these triples;
allowing to combine alternatives with a function corresponding to OR-disjunction in inductive logic programming;
stopping the generating once a pre-defined ratio of coverage of the set is reached or a user-defined execution time timeout is met; and
providing the resulting completion rules as candidates for completion of the knowledge graph;
wherein the two triple patterns
{?x property1 ?y. ?y property2 ?z.}
are replaced with the new triple pattern
?x property1/property2 ?z
and / or whereby the two triple patterns
{?x property1 ?y. ?z property2 ?y.}
are replaced with the new triple pattern
?x property1/{circumflex over ( )}property2 ?z.
2. The method according to claim 1 , wherein one or more properties are selected or in the absence of a selection all properties occurring are considered candidates for completion rule generation.
3. The method according to claim 1 , wherein the set of inferred triples is optionally filtered and/or post-filtered based on user settings.
4. The method according to claim 3 , wherein the likelihood of an inferred triple being true, as provided by the machine learning model, is used for the filtering by omitting all inferred triples whose likelihood is below a pre-defined threshold.
5. The method according to claim 1 , wherein elements with a same functionality as other elements of SPARQL WHERE clauses, such as FILTER statements, are learned.
6. The method according to claim 1 , wherein rules that have been approved by an expert on the knowledge domain, are applied on the RDF data.
7. The method according to claim 6 , wherein the prediction machine learning model is re-trained on the resulting combined RDF data.
8. The method according to claim 1 , wherein inferred triples that are covered by resulting rules are provided along with the resulting rules, and/or wherein a function corresponding to OR-disjunction is either providing multiple completion rules or combining WHERE bodies of rules with a UNION statement.
9. The method according to claim 1 , wherein resulting rules are provided as candidates for completion of the knowledge graph to an expert on the knowledge domain for evaluation.
10. A computer program product, comprising a computer readable hardware storage device having computer readable program code stored therein, said program code comprising program instructions that cause, when the program is executed by a computer, the computer to carry out the method according to claim 1 .
11. A computer-implemented method for generation of completion rules for a knowledge graph, comprising the following steps:
using a machine learning model to produce a set of inferred triples from RDF data, wherein the model is a generic model or a model trained and/or retrained on the knowledge graph or a subset of the knowledge graph;
generating completion rules with a same functionality as SPARQL queries of the form
INSERT ?subject ?predicate ?object
WHERE
{
triple_pattern_1 .
triple_pattern_2 .
...
}
as follows:
adding to completion rules, that would produce triples that are not part of the set, triple patterns that result in the exclusion of these triples;
allowing to combine alternatives with a function corresponding to OR-disjunction in inductive logic programming;
stopping the generating once a pre-defined ratio of coverage of the set is reached or a user-defined execution time timeout is met; and
providing the resulting completion rules as candidates for completion of the knowledge graph; wherein rules that have been approved by an expert on the knowledge domain are applied on the RDF data; and wherein the prediction machine learning model is re-trained on the resulting combined RDF data.
12. The method according to claim 11 , wherein one or more properties are selected or in the absence of a selection all properties occurring are considered candidates for completion rule generation.
13. The method according to claim 11 , wherein the set of inferred triples is optionally filtered and/or post-filtered based on user settings.
14. The method according to claim 11 , wherein the likelihood of an inferred triple being true, as provided by the machine learning model, is used for the filtering by omitting all inferred triples whose likelihood is below a pre-defined threshold.
15. The method according to claim 11 , wherein elements with a same functionality as other elements of SPARQL WHERE clauses, such as FILTER statements, are learned.
16. The method according to claim 11 , wherein in an optional post-processing step the WHERE statements are refined by introducing property paths and/or wherein the two triple patterns
{?x property1 ?y. ?y property2 ?z.}
are replaced with the new triple pattern
?x property1/property2 ?z
and / or wherein the two triple patterns
{?x property1 ?y. ?z property2 ?y.}
are replaced with the new triple pattern
?x property1/{circumflex over ( )}property2 ?z .
17. The method according to claim 11 , wherein inferred triples that are covered by resulting rules are provided along with the resulting rules.
18. The method according to claim 11 , wherein a function corresponding to OR-disjunction is either providing multiple completion rules or combining WHERE bodies of rules with a UNION statement.
19. A computer program product, comprising a computer readable hardware storage device having computer readable program code stored therein, said program code comprising program instructions that cause, when the program is executed by a computer, the computer to carry out the method according to claim 11 .
20. A computer-implemented method for generation of completion rules for a knowledge graph, comprising the following steps:
using a machine learning model to produce a set of inferred triples from RDF data, wherein the model is a generic model or a model trained and/or retrained on the knowledge graph or a subset of the knowledge graph;
generating completion rules with a same functionality as SPARQL queries of the form
INSERT ?subject ?predicate ?object
WHERE {
triple_pattern 1 .
triple_pattern 2 .
...
}
as follows:
adding to completion rules, that would produce triples that are not part of the set, triple patterns that result in the exclusion of these triples;
allowing to combine alternatives with a function corresponding to OR-disjunction in inductive logic programming;
stopping the generating once a pre-defined ratio of coverage of the set is reached or a user-defined execution time timeout is met; and
providing the resulting completion rules as candidates for completion of the knowledge graph; wherein in an optional post-processing step the WHERE statements are refined by introducing property paths; and
wherein the two triple patterns
{?x property1 ?y. ?y property2 ?z.}
are replaced with the new triple pattern
?x property1/property2 ?z
and / or wherein the two triple patterns
{?x property1 ?y. ?z property2 ?y.}
are replaced with the new triple pattern
?x property1/{circumflex over ( )}property2 ?z.