Domain-Specific Entity Linking Model Based On Entity Synonym Detection And Relevance Estimation

Knowledge bases have changed the way how people manage and make use of knowledge.Entity linking is a key step in the applications of knowledge bases.Entity linking is the process of mapping entity mentions in the text to their corresponding entities in a knowledge base.It plays an important role in many applications,including question answering over knowledge bases,knowledge base population and event analysis and so on.And it is a problem with longterm study.A great portion of previous researches have dealt with open-domain entity linking,while domain-specific entity linking has drawn more attention in recent years.As domain-specific entity linking tasks may have different characteristics from open-domain tasks,traditional entity linking techniques developed for open-domain entity linking cannot be applied effectively.In this paper,we focus on domain-specific microblog entity linking and tackle with two main issues:1.Informal entity mentions and invented entity mentions are popular.Many of these entity mentions are not recorded in knowledge bases and pose a challenge for entity linking.2.Entity popularity is an important feature adopted by many entity linking approaches for entity disambiguation.Since the entity popularity in a specific field is different from the global entity popularity with respect to multiple fields.Existing approaches for entity popularity estimation in open domain cannot be applied effectively to domain-specific entity linking tasks directly.To deal with these issues,we proposed an entity linking model based on entity synonym detection and entity relevance estimation.And we adopted distant supervision technique to reduce the cost of data annotation and make our model more applicable.We created a dataset and conducted extensive experiments to show the effectiveness of our proposed approach.