Integration of knowledge bases and external sources using user-defined predicates and adaptive entity linking

Yuuki Ohmori, Hiroyuki Kitagawa, Toshiyuki Amagasa, Akiyoshi Matono

Abstract: In recent years, various types of knowledge have been published as knowledge bases (KBs) in the resource description framework (RDF), and their use in knowledge processing has been expanding. An RDF KB statically represents knowledge as a subject-predicate-object triple graph and can be queried using SPARQL. On the other hand, various domain-specific databases are not provided in RDF due to limited demand or expertise. In addition, dynamic data (e.g., weather or traffic) is unsuitable for static accumulation. Moreover, sources like maps require domain-specific processing to extract relevant information. These limitations prevent RDF KBs from fully handling diverse and dynamic information, motivating the need for its integrated use with external sources. Two key challenges arise to achieve such integration: first, each external source has its own access method and response format; second, it is necessary to identify the corresponding KB entity for each external object obtained from the external source. Both challenges are non-trivial. Moreover, given the diversity of external sources and target domains, the method used for entity identification or linking should also adapt to this diversity. To address these challenges, we propose an integration architecture named Knowledge Mediator. Knowledge Mediator enables query execution as if external sources were a part of the knowledge base, by employing an adaptive entity linking approach using the Magic Property mechanism of SPARQL. Users can register linking functions that specify the most appropriate entity linking methods. We implemented a prototype and experimentally demonstrate the advantages of Knowledge Mediator.