Entity Linking in Enterprise Search: Combining Textual and Structural Information

Abstract

Fast and correct identification of named entities in queries is crucial for query understanding and to map the query to information in structured knowledge base. Most of the existing work have focused on utilizing search logs and manually curated knowledge bases for entity linking and often involve complex graph operations and are generally slow. We describe a simple, yet fast and accurate, probabilistic entity-linking algorithm used in enterprise settings where automatically constructed, domain specific Knowledge Graphs are used. In addition to the linked graph structure, textual evidence from the domain specific corpus is also utilized to improve the performance.