Research on Personalized Product Recommendation Algorithm for User Implicit Behavior Feedback

Abstract

Many e-commerce platforms use various discount promotions, celebrity endorsements, international brand settlement, advertising, and other techniques to increase overall sales and client ownership. Indeed, this type of “thousands of faces” recommendation technology is achieved through a reasonable recommendation technology, allowing platform users to explore the things they are interested in without having to leave their homes. By analyzing the improvement direction of personalized recommendation algorithms in practice, the coordination mode of big data processing framework and computing framework, and the overall efficiency improvement of recommendation system, this paper aims to build a set of personalized product recommendation algorithms with high data utilization, timely recommendation feedback, and stable calculation core. The existing BPR algorithm is improved and optimized in this paper, and the MBPR algorithm is used to calculate the weight of implicit feedback behavior using the entropy method in the subjective empowerment method and the order relationship analysis method in the objective empowerment method, and to quantify the weight of heterogeneous user implicit feedback behavior using the combined empowerment method. The user preference model used in the recommendation system is then created by taking into account the time weight decay factor in the quantitative data. The model accurately estimates the changed user's implicit feedback behavior. In comparison to the 68.61% and 18.68% performance of the conventional BPR algorithm, the MBPR method designed in this work attained an accuracy of 77.21% and 25.17% in the AUC index and 25.17%, respectively.