E-commerce personalized recommendation method based on context multi-arm gambling machine

Abstract

The invention discloses an e-commerce personalized recommendation method based on a context multi-arm gambling machine, and the method comprises the steps of calculating the similarity between the context information and the context multi-arm gambling machine action characteristics, enabling the similarity and a model action estimation value to be combined into a new action estimation value, and selecting a new action with the maximum action estimation value for recommendation, wherein the context gambling machine model is used for selecting the action with the maximum similarity and recommending according to the context information; the explored part is to select the action with the maximum estimated value, correspond to the hot recommendations in a recommendation system, and finally update a model according to feedback of a user. According to the invention, the context information and the action estimation value are simultaneously used for action selection, and the immediate reward is maximized through utilization, the diversity of recommended content is increased by exploring and discovering information of potential preferences of users. And meanwhile, when the context information is unknown, the recommendation is performed through exploration, so that the cold start problem in the recommendation field is effectively solved.

Description

technical field [0001] The invention relates to a personalized recommendation method for e-commerce, in particular to a personalized recommendation method for e-commerce based on context multi-armed gambling machines. Background technique [0002] Reinforcement learning is the learning of agents from environmental states to behavioral mappings for solving sequential decision-making problems. Any decision-making problem involves exploration and utilization, in which utilization is to select the optimal strategy based on the current known knowledge experience, and exploration is to try other suboptimal strategies. Utilization can obtain the maximum immediate reward, but when the learning is insufficient, the algorithm will fall into a local optimum, and exploration can fully learn the rewards of each strategy, and find the optimal strategy without causing the agent to fall into a local optimum, which helps to maximize Cumulative returns, but exploration requires more learning...

AI Technical Summary

Problems solved by technology

There are two main problems in the traditional recommendation system: (1) cold start problem, new users do not have a scoring matrix, and cannot make recommendations (2) the diversity of recommendation information, traditional recommendation algorithms are difficult to explore the potential interest preferences of users, only based on Historical information, recommendation results are not novel

Images