Recommendation method, device, electronic device and storage medium based on action pruning

Abstract

The present invention provides a recommendation method, device, electronic device and storage medium based on action pruning. The status and rating prediction model corresponding to the recommended content predicts the rating of each content to be recommended, and recommends to target users based on the rating of each content to be recommended; the rating prediction model is obtained by reinforcement learning. In the process of reinforcement learning, the rating prediction The model obtains the regret value of each candidate score in the current sample state from the regret value set, and predicts the score based on the candidate score whose regret value is greater than the preset threshold. The regret value set stores the historical state and its corresponding regret value. The regret value is based on The advantages of each candidate score in the historical state are determined. The historical state is the sample state before the current sample state, which speeds up the convergence speed of reinforcement learning and realizes personalized and accurate recommendations for users.

Description

technical field [0001] The present invention relates to the field of computer technology, in particular to a recommendation method, device, electronic equipment and storage medium based on action pruning. Background technique [0002] Due to the advantages of being able to perceive the dynamic environment and obtain rewards from the environment to continuously adapt to the environment, reinforcement learning is especially suitable for business scenarios involving interactions, such as recommending content to users. [0003] However, in practical applications, the traditional reinforcement learning algorithm only improves the policy based on the immediate reward obtained from the environment, which has the problem of slow convergence speed, and can only be applied to learning tasks dealing with small-scale action spaces. Therefore, how to improve the convergence speed of reinforcement learning is an important issue to be solved urgently in the industry. Contents of the inve...

AI Technical Summary

Problems solved by technology

[0004] The present invention provides a recommendation method, device, electronic equipment and storage medium based on action pruning, which is used to solve the defect of slow convergence speed of reinforcement learning in the prior art, and to improve the convergence speed of reinforcement learning

Images