Supervised Learning for Linking Named Entities to Knowledge Base Entries

This paper addresses the challenging information extraction problem of linking named entities in text to entries in a knowledge base. Our approach uses supervised learning to (a) rank candidate knowledge base entries for each named entity, (b) classify the top-ranked entry as the correct disambiguation or not, and (c) group together the named entities without a corresponding entry in the knowledge base. We analyze the fundamental design challenges involved in the development of a learningbased entity-linking system, and we provide extensive experimental results for a wide range of methods and feature sets. Our experiments over the datasets from the Text Analysis Conference (TAC) Entity Linking Task demonstrate the effectiveness of supervised learning methods, showing that out-ofthe-box algorithms and relatively simple to compute features can obtain very competitive results.