Chinese latent relational search based on relational similarity

Abstract

Latent relational search is a new way of searching information in an unknown domain according to the knowledge of a known domain from the Web. By analyzing the analogous relationship between word pairs, the latent relational search engine can tell us the accurate information that we need. Given a Chinese query, {(A, B), (C, ?)}, our aim is to get D which is the answer of “?”. In this paper, we propose an approach to Chinese Latent relational search for the first time. Moreover, we classify the relation mappings between two word pairs into three categories according to the number of relations and the number of target words corresponding to each relation. Our approach firstly extracts relation- representing words by using the preprocessing modular and two clustering algorithms, and then the candidate word set of D corresponding to each relation-representing word can be obtained. The proposed method achieves an MRR of 0.563, which is comparable to the existing methods.