Chinese Query Reformulation And Variation: a Case Study in Sogou Log

摘要

Query reformulation and variation are two of the most difficult and essential aspects of information seeking and retrieval. Current literatures only investigate changes in number of terms in successive queries, but they don't analyze modified positions and reformulation patterns when users reformulate queries. In this paper, we first propose 8 reformulation classes according to different modified positions and changes in number of query terms to analyze query reformulation. Experimental results show that Chinese users prefer to partially reformulate queries to shorter ones. Second, we define reformulation pattern and try to find the most frequently used reformulation pattern by Chinese users. We find that if a user only reformulates query once in one session, he is most likely to add terms to the query at the two ends. Finally, we denote the conception of query variation the query differences with the same search intent. According to Sogou log, different abstractive level of queries can most likely lead to the same click-through set with different queries. The findings of this paper are beneficial to the study of information retrieval modeling and higher preference prediction accuracy.