Relevance Feedback using Query Logs

A search engine retrieves the documents based on the query su bmitted to it. However, incorporation of user modelling, by the inclusion of p ast information (like the previous queries submitted and the titles of the documents c licked) is expected to increase the accuracy of the search results. Especially, in the case of short term history, such history information is highly related with th e current user query and can help in explaining the user information needs in a batter way. In order to do the same, we develop and experiment with some “history incor poration and term reweighting” techniques that incorporate the user history along with the current query. These techniques expand the current query by includi ng terms from the history queries and the document titles, and reweight the te rms. The results confirm that the incorporation of history along with the current query is considerably better in performance than the usage of only the current quer y. We compare the retrieval models against each other and also analyze the combi nations of the retrieval models with the incorporation and reweighting techniques.