HIT Approaches to Entity Linking at TAC 2011

This paper describes the system of HIT at the 2011 Text Analysis Conference (TAC) Knowledge Base Population (KBP) track English Entity Linking task. Based on structured and unstructured information extracted from Wikipedia, this system predicts the most probable entity that a query mention might refer to. A similarity score is assigned to the candidate entity by computing the the relatedness between the query and the entity, and augmented by the popularity of the entity. We model the query context as a graph of the entities and utilize the referential relationship between the context entities and the candidate entities in Wikipedia to measure the relatedness between the query context and the candidate entity. Evaluation results show the performance of our system reaches the median value of all the participating systems.