Improving Web Page Retrieval using Search Context from Clicked Domain Names

Search context is a crucial factor that helps to understand a user’s information need in ad-hoc Web page retrieval. A query log of a search engine contains rich information on issued queries and their corresponding clicked Web pages. The clicked data implies its relevance to the query and can be used to define the topical context. However, the log is usually not completely available due to privacy concerns. In this paper, we derive clicked pages from clicked domains and use the surrounding query context to enhance retrieval performance. One strategy is to promote clicked pages directly in the initial retrieval result. Another strategy is to expand the original query using selected terms from the clicked pages. Our experimental results on the TREC GOV2 data and a query log of a major search engine show that both strategies can boost retrieval performance compared to the standard language model and pseudo relevance feedback (PRF) model. Their good performance on early precision allows us to apply PRF further for even more accurate result that is comparable to the performance of true relevance feedback.