Diversified recommendation method and system based on reinforcement learning and storage medium

Abstract

The invention provides a diversified recommendation method based on reinforcement learning, and the method is characterized in that the method comprises the following steps: S1, obtaining a training sample, and determining and initializing network parameters; S2, executing a strategy generation action; S3, evaluating and optimizing the strategy; S4, supervising the loss through the comment network; and S5, updating the exploration probability. According to the method, the acquisition of the optimal recommendation sequence can be driven through the reward in the long-stage operation process, the high reward is obtained from the good recommendation action through trial and error, and finally the optimal recommendation list in various states is learned to maximize the long-term income.

Description

technical field [0001] The invention relates to the field of e-commerce item recommendation, in particular to a reinforcement learning-based diversified recommendation method, system and storage medium. Background technique [0002] Today's e-commerce platforms widely use the recommendation system, which predicts the user's preference for certain items by analyzing the user's historical behavior information, and recommends a group of items to them to achieve the purpose of information filtering among massive items. The early methods of the recommendation system are mainly content-based recommendation or collaborative filtering recommendation. There are many specific implementation methods, some use the similarity between users, some use the similarity between items, and some use user characteristics. match with product characteristics. The early method has great limitations, and a large number of improved algorithms have been proposed in the follow-up, and the evaluation of...

AI Technical Summary

Problems solved by technology

Although this method is simple, it is very difficult to define a general scoring formula. It is often necessary to manually adjust the parameters according to the usage environment, and it is difficult to achieve the global optimum.

There are also supervised learning methods to carry out diversified recommendations, but it is very difficult to obtain sufficient training samples. Even if they are obtained, there may be a large difference between them and the actual running samples. In addition, the diversified evaluation indicators cannot be directly used to guide the training process.

Images