Deep reinforcement learning interactive recommendation system and method based on knowledge enhancement

Abstract

The invention provides a deep reinforcement learning interactive recommendation system and method based on knowledge enhancement, and relates to the technical field of recommendation. The system comprises a data acquisition and cleaning module, an environment simulator construction module, a knowledge graph construction module, a graph convolution module, a user state representation module, a strategy network module and a value network module. According to the method, rich semantic information in a knowledge graph is combined, a graph convolutional network structure is utilized, embedded representation of adjacent entities is propagated recursively along high-order connectivity, a graph attention network thought is adopted, item representation is enhanced by utilizing the rich semantic information in the knowledge graph, and meanwhile, a user-item bipartite graph is fused, so that the method is more efficient and efficient. The potential relationship is fully mined from collective user behaviors, so that the dynamic preference of the user is accurately captured, and the optimal recommendation strategy is autonomously learned by using deep reinforcement learning, so that the recommendation accuracy is improved.

Description

technical field [0001] The invention relates to the technical field of recommendation, in particular to a deep reinforcement learning interactive recommendation system and method based on knowledge enhancement. Background technique [0002] With the rapid development of mobile applications such as Douyin, Pandora, and Instagram Feeds, existing recommendation systems are under enormous pressure to cope with the constant emergence of new users, changing user interests, and dynamic changes in the environment. However, traditional recommendation methods, such as content-based recommendation methods and matrix factorization-based recommendation methods, all assume that user interests are static, and learn user preferences for items from historical interaction data between users and items. . However, these methods often ignore the dynamic changes of user interests and environments. Furthermore, most of the existing recommender systems aim to maximize the immediate (short-term) r...

AI Technical Summary

Problems solved by technology

The existing interactive recommendation methods based on reinforcement learning still have two limitations: (1) Most of the existing interactive recommendation methods use historical interaction records to learn strategies. However, due to the sparse user feedback and the huge action space, it is difficult to Efficiently learn the optimal recommendation strategy; (2) The existing methods based on reinforcement learning represent the state / action of each individual user in isolation, and have not considered the potential relationship between different users

Images