Effective entity linking for news articles

Abstract

This master's thesis presents an entity-linking tool for detecting entity-bearing words in news articles and linking them to the corresponding entries in a knowledge base. Given the vast volume of news articles produced every day, such a tool needs to be not only effective but also effcient. The approach consists of three steps. First, entities are spotted on the basis of capitalized (uppercase) letters in words. Next, entities with surface forms matching the mention are considered. Finally, in the disambiguation phase, a new method based on local relatedness is employed. Using the Entity Recognition and Disambiguation 2014 Challenge platform, we demonstrate that the efficiency of this solution is competitive with other available methods.