Research On Recommendation Algorithm Based On User Behavior And Three-way Decision

Influenced by the development of e-commerce,many traditional retail enterprises try new retail business models by using online stores and personalized recommendation technologies on the basis of offline physical retailing.Collaborative Filtering Recommendation algorithm(CFR)is the earliest and most widely used recommendation algorithm,and is often used for personalized recommendation in e-commerce.However,some problems in the application of collaborative filtering recommendation algorithm and the operation characteristics of retail enterprises lead to the problems of sparse data,low recommendation accuracy and high operating cost of the algorithm in the recommendation system of retail enterprises,which affect the overall recommendation efficiency of the recommendation system.Therefore,it is necessary to improve the algorithm to overcome the defects of its application in the recommendation system of retail enterprises.This paper takes supermarket chains in retail enterprises as an example,combines the operating environment and characteristics of supermarket chains,studies the improvement of three recommendation steps of collaborative filtering recommendation algorithm,and proposes a three decision collaborative filtering recommendation algorithm(UBT-TWDCFR)based on online and offline user behavior and time influence.Considering the characteristics of the business environment in the retail industry,this paper improves the three recommendation steps of collaborative filtering recommendation algorithm,namely,constructing the “User-Item”matrix,finding the nearest neighbor and making the recommendation decision.The algorithm constructs the scoring matrix of “User-Item” by combining explicit and implicit behaviors of users and emotional characteristics of comments.When calculating user similarity,the time attenuation function is used as the time weight to measure the deviation of user interest with time.Use the three-way decision rule as the decision strategy in the recommendation decision.In the experimental part of this paper,the operation data of supermarket XX for two months were used,and the recommendation process of UBT-TWDCFR improved algorithm was introduced to calculate the personalized recommendation results based on the actual operation data.Then,the recommendation results of UBT-TWDCFR model are compared with the recommendation algorithms of CFR,C-CFR and T-CFR,and the improvement effect of the algorithm is tested from the quality of the recommendation results.In this paper,the performance of the four algorithms in matrix data sparsity,RMSE,accuracy and recommendation cost is compared and analyzed.The experimental results show that the improved UBT-TWDCFR algorithm has obvious effect on reducing the matrix sparsity and improving the accuracy of the recommendation algorithm,and also reduces the recommendation cost of the recommendation algorithm to some extent.It is proved that the improved UBT-TWDCFR algorithm proposed in this paper improves the recommendation efficiency of the collaborative filtering recommendation algorithm,improves the recommendation accuracy and reduces the data sparsity of the algorithm.