Extending AIDA framework by incorporating coreference resolution on detected mentions and pruning based on popularity of an entity

Named Entity Disambiguation (NED) is gaining popularity due to its applications in the field of information extraction. Entity linking or Named Entity Disambiguation is the task of discovering entities such as persons, locations, organizations, etc. and is challenging due to the high ambiguity of entity names in natural language text. In this paper, we propose a modification to the existing state of the art for NED, Accurate Online Disambiguation of Entities (AIDA) framework. As a mention’s name in a text can appear many times in shorter forms, we propose to use coreference resolution on the detected mentions. Entity mentions within the document are clustered to their longer form. We use the popularity of candidate entities to prune them and based on the similarity measure of AIDA the entity for a mention is chosen. The mentions are broadly classified into four categories person, location, organization and miscellaneous and the effect of coreference and pruning were analyzed on each category.