Entity Linking with Multiple Knowledge Bases: An Ontology Modularization Approach

Abstract

The recognition of entities in text is the basis for a series of applications. Synonymy and Ambiguity are among the biggest challenges in identifying such entities. Both challenges are addressed by Entity Linking, the task of grounding entity mentions in textual documents to Knowledge Base entries. Entity Linking has been based in the use of single cross-domain Knowledge Bases as source for entities. This PhD research proposes the use of multiple Knowledge Bases for Entity Linking as a way to increase the number of entities recognized in text. The problem of Entity Linking with Multiple Knowledge Bases is addressed by using textual and Knowledge Base features as contexts for Entity Linking, Ontology Modularization to select the most relevant subset of entity entries, and Collective Inference to decide the most suitable entity entry to link with each mention.