Active Exploration to Improve Learning Rate from Click through Data

A search engine is a system that tries to retrieve documents that the user needs, given the user’s query. No search engine is able to do this perfectly, for many reasons. Learning systems have previously been examined to get better retrieval, but with mixed results. The main goal of the report is to show that active exploration is necessary for learning in order to get good results. An algorithm is proposed and evaluated, that alters the presented search result in order to receive user data that is of more use in learning. Simulations show that learning may stop early if active exploration is not used. They also show that the active part affects the search results negatively, wherefore caution should be taken when choosing a strategy for active presentation.