Search Facet Creation from Click Logs

Sumio Fujita, Tatsuya Uchiyama, Georges Dupret, Ricardo Baeza-Yates

SIGIR2010 Workshop on Query Representation and Understanding, 2010/7


Information Retrieval

According to user behavior against search results, not only the specific queries, which are generated by search assist functionality baseed on query expansion, but also topically shifted queries are useful as recommended queries. In order to recommend such queries with different characteristics, we combine click-based, topic-based and session based recommendations by a supervised learning in order to maximize semantic similarities between the query and the recommendation. According to the evaluation using query/click logs of Japanese web search, the combination of three methods improve the ranking siginificantly better than the ranking of single method.