Commodity Recommendation Algorithm Based On User Features And Reinforcement Learning

With the rapid development of mobile e-commerce,the number of users and products are increasing,which makes it impossible for a large number of commodities to quickly reach the target customers.This requires the mobile e-commerce platform to perform high-quality commodity recommendation algorithms and build bridges between products and potential customers.However,most of the current commodity recommendations rely on traditional recommendation algorithms such as collaborative filtering,and the recommended efficiency and accuracy cannot satisfy the requirement.Therefore,in the mobile e-commerce scenario,studying commodity recommendation algorithm combining emerging technologies such as features analysis and reinforcement learning are particularly important for the development of mobile e-commerce.Based on the actual project scenario and the characteristics of mobile e-commerce data,this paper studies a recommendation algorithm combining user feature analysis and dynamic state mining.The main work includes the following points:(1)The characteristics of the mobile e-commerce data collected in the project were studied,the problems existing in the data and the factors affecting the recommendation quality were analyzed,and a series of pre-processing operations were performed on the data.At the same time,in order to improve the efficiency and accuracy of the algorithm,the data is converted in a variety of ways.(2)A recommendation algorithm based on user features and reinforcement learning is proposed.This paper analyzes the user behavior of mobile e-commerce data,constructs a user feature set,trains the user behavior models,and ensembles the models to form a recommendation based on static features.At the same time,this paper designs the decision process of mobile e-commerce and describes the user's dynamically states.A deep decision network is trained based on user feedback to generate dynamic product recommendations.Finally,this paper uses a hybrid strategy to mix the two recommendations as the final recommended output.(3)A user behavior simulator is established to provide feedback information of the user during the training process of reinforcement learning.The behavior data of users and commodities in mobile e-commerce data is sparse and cannot meet the needs of reinforcement learning training.Therefore,based on the user's behavior logic,this paper constructs a user behavior simulator based on recurrent neural network to provide simulated user feedback to reinforcement learning.Based on the mobile e-commerce data in the project,this paper validates the proposed algorithm and compares it with some popular recommendation algorithms.The experimental results show that the proposed algorithm has significant improvement in recommendation quality and efficiency,which proves the effectiveness of the algorithm.