Interpreting Fine-Grained Categories From Natural Language Queries Of Entity Search

The fine-grained target categories/types are very critical for improving the performance of entity search because they can be used for retrieving relevant entities by filtering irrelevant entities with a high confidence. However, most solutions of entity search face an urgent problem, i.e., the lack of fine-grained target categories of queries, which are hard for users to explicitly specify. In this paper, we try to interpret fine-grained categories from natural language based queries of entity search. We observe that entity search queries often contain terms specifying the contexts of the desired entities, as well as a topic of the desired entities. Accordingly, we propose to interpret fine-grained categories of entity search queries from the context perspective and the topic perspective. Therefore, we propose an approach by formalizing both context-based category model and topic-based category model, to tackle the category interpreting task. Extensive experiments on two widely-used test sets: INEX-XER 2009 and SemSearch-LS, indicate significant performance improvement achieved by our proposed method over the state-of-the-art baselines.