Abstract

The goal of query conceptualization is to map instances in a query to concepts defined in a certain ontology or knowledge base. Queries usually do not observe the syntax of a written language, nor do they contain enough signals for statistical inference. However, the available context, i.e., the verbs related to the instances, the adjectives and attributes of the instances, do provide valuable clues to understand instances. In this paper, we first mine a variety of relations among terms from a large web corpus and map them to related concepts using a probabilistic knowledge base. Then, for a given query, we conceptualize terms in the query using a random walk based iterative algorithm. Finally, we examine our method on real data and compare it to representative previous methods. The experimental results show that our method achieves higher accuracy and efficiency in query conceptualization.

Thanks for your interests in this paper. Please also pay attentions to our ACL 2016 short text understanding tutorial: Understanding Short Texts – ACL 2016 Tutorial, presented by Zhongyuan Wang and Haixun Wang