Selection and reasoning
Work in progress:Workshop in Lyon
Unifying reasoning and search
Atanas Kiryakov, Unifying reasoning and search paper (draft)
Yi Zeng, Yan Wang, Zhisheng Huang, Ning Zhong. Unifying Web-scale Search and Reasoning from the Viewpoint of Granularity In: Proceedings of the 2009 International Conference on Active Media Technology, Lecture Notes in Computer Science 5820, 418-429, Springer, Beijing, China, October 22-24, 2009.
Interleaving reasoning and selection
[WP2, WP4 and WP7a] Unifying Search and Reasoning Based on User Interests, work by WICI, ?OntoText, and Astrazeneca.
Here are some typical tasks which need interleaving reasoning and selection to solve the scalability:
(T1) Concept Classification with Consistent Ontologies: To check whether or not a concept C1 is a subConcept of C2 when the ontologies are consistent.
(T2) Concept Classification with Inconsistent Ontologies: To check whether or not a concept C1 is a subConcept of C2 when the ontologies are inconsistent.
(T3) Concept Classification with Dynamic Ontologies: To check whether or not a concept C11 is a subconcept of C2 by partially reusing the previous reasoning results after the ontologies are changed.
(T4) Concept Retrieval with Consistent ontologies: to return a set of super/sup-concepts of a given concept C when the ontologies are consistent.
(T5) Concept Retrieval with Inconsistent ontologies: to return a set of super/sup-concepts of a given concept C when the ontologies are inconsistent.
(T6) Concept Retrieval with Dynamice ontologies: to return a set of super/sup-concepts of a given concept C by partially reuse the previous reasoning results when the ontologies are changed.
(T7) Instance Checking with consistent ontologies: to check whether or not an individual a is an instance of a concept when the ontologies are consistent.
(T8) Instance Checking with inconsistent ontologies: to check whether or not an individual a is an instance of a concept when the ontologies are inconsistent.
(T9) Instance Checking with dynamic ontologies: to check whether or not an individual a is an instance of a concept by reusing partially the previous reasoning results after the ontologies are changed.
(10) Parent concept retrieval for individual with consistent ontologies: given an individual a, to return its parent concept C (i.e. no other subconcept C1 of C such that the individual a is the instance of the subconcept C1) when the ontologies are consistent.
(11) Parent concept retrieval for individual with inconsistent ontologies: given an individual a, to return its parent concept C (i.e. no other subconcept C1 of C such that the individual a is the instance of the subconcept C1) when the ontologies are inconsistent.
Notes from integration session in Berlin
- Naso: spreading activation can be used by selection but also some of the produced facts could remain to be used by reasoner; at present, plugins clear up everything, but we could change the platform/plugins and make them state-aware so that some information can remain for the next plugin in the pipeline;
- RDF Search: done by Onto and USFD is in deliverables but not implemented
- Spreading activation can be used for reasoning in the datalayer; currently, SA is in the datalayer, but the reasoning is not;
- how do you search RDF graph?
- o the same as you search on Google: keywords are URIs (local name) or literals or RDF molecules (all literals of the URI)
- Agreed: write a (white) paper on combining reasoning and search
- o we need clearly defined tasks for reasoning and search, defined by WP4? o challenge problems: wp4 to come up with 20 of them
- + what would be important to evaluate them? quality measured by reducing the time to do some reasoning task
- o we need clearly defined tasks for reasoning and search, defined by WP4? o challenge problems: wp4 to come up with 20 of them
- deductive methods+reasoning; inductive and deductive reasoning is a good idea
- d4.3.1.:interleaving reasoning and selection
- PION: returns meaningful answer for inconsistent ontologies
- selection and heuristics
- create user context and use it for reasoning
- could be seen as query extension but wici think it's reasoning