Selective Exploration of Commercial Documents in Web Search

Implicit user feedback is known to be a strong signal of user preferences in web search. Hence, solving the exploration-exploitation dilemma [5] became an important direction of improvement of ranking algorithms in the last years. In this poster, in the case of commercial queries, we consider a new negative effect of exploration on the user utility -- distracting and confusing users by shifting well-known documents from their common positions -- and propose an approach to take it into account within Multi-Armed Bandit algorithms, usually applied to solve the dilemma.