Leveraging related entities for knowledge base acceleration

Abstract

Knowledge bases such as Wikipedia have been shown to be effective to improve the performance in many information tasks. Clearly, the effectiveness is based upon the quality of these knowledge bases. A high-quality knowledge base should have up-to-date complete information. However, constructing a high-quality knowledge base is not an easy task because it would require significant manual efforts to collect relevant documents, extract valuable information and update the knowledge bases accordingly. In this paper, we aim to automate this labor-intensive process. Specifically, we focus on how to collect relevant documents with regard to an entity from sheer volume of Web data automatically. To solve the problem, we propose to construct the profile of the entity by leveraging a set of its related entities and then discuss how to use the training data to weight the related entities. Experiments over the TREC 2012 KBA collection shows that the proposed method can outperform state-of-the-art methods.