Research on Retrieval Ranking Based on Deep Reinforcement Learning

Abstract

Retrieval ranking technology is the core technology for evaluating information retrieval results. The advantages and disadvantages of retrieval ranking algorithms directly affect the retrieval effect of the system. The traditional retrieval ranking algorithm treats each ranking decision step independently. Moreover, the traditional retrieval ranking algorithm does not consider personalized retrieval ranking for different types of users. The recommendation algorithm based on deep reinforcement learning has made a lot of research on this problem, and it is a new attempt to apply the idea of recommendation algorithm in the retrieval ranking scene. In this paper, the related algorithms of recommendation and retrieval ranking algorithms based on deep reinforcement learning are reviewed in recent years. By considering the ranking process as the Markov decision process, using reinforcement learning to solve the problem of correlation between decision-making steps, an interactive retrieval model is constructed. Interactive retrieval can guide users to define their own needs, by introducing a personalized user simulator to simulate different types of environments, and using reinforcement learning to train personalized retrieval goals. Combining the retrieval rank with the recommendation algorithm will improve the retrieval effect of the user essentially.