Due to the proliferation and abundance of information on the World Wide Web (WWW), finding information for user’s query has become increasingly tedious. Hence, search engines have an important role in the current internet. Search engines find information on the Web for users who had legitimate needs for information. The role of ranking algorithms as important part of search engines is crucial: selecting the pages that are most likely be able to satisfy the user’s needs, and bring them in the top positions. In this paper, with formulation ranking as a reinforcement learning problem, a new query-dependent ranking algorithm is proposed which called RURL. Reinforcement learning problems are structured around estimating value functions. In our algorithm, each web page is considered as a state and its score is as value function of the state. The proposed method is evaluated using dotIR, LETOR benchmark datasets and found to be competitive with some famous ranking algorithms. Experimental results show that RURL algorithm outperforms other ranking algorithms and the accuracy of ranking web pages can be enhanced by employing reinforcement learning concepts.
search engin; ranking algorithm; reinforcement learning; value function; state