Research On Collaborative Filtering Recommendation Method Integrating User Ratings And Item Quality Information

The rapid development of computer and Internet technology has greatly expanded the channels for people to obtain information and effectively promoted the sharing of diversified information.However,while people are experiencing the convenience of the Internet in the face of massive and complex information,they are also increasingly finding it more difficult to obtain useful information efficiently.In order to alleviate the problem of information overload,the recommendation system emerges at the right moment and rapidly develops into a key support for accurate information delivery in e-commerce platforms,social media and other fields.However,it needs to be emphasized that despite a series of important research advances around recommender systems in academia and industry,data sparsity remains one of the major challenges it faces.In the context of sparse data,the phenomenon of few common interest items among users often leads to the failure or inaccuracy of classical similarity measures.Therefore,starting from the in-depth analysis of the characteristics of user behavior items and mining the deeper information of user items,it naturally becomes an important research direction of recommendation methods.In this paper,based on the in-depth mining of user-item two-dimensional perspective information,a collaborative filtering recommendation method based on similarity improvement was established by comprehensively considering the key factors such as user rating bias,the proportion of common interest items,popular items penalty factor,and item quality deviation.The main research contents are as follows:(1)This paper proposes and improves the similarity measure criterion based on user rating difference.Starting from the perspective of users,this paper considers the scoring differences between users on common interest items and user rating bias behavior,and integrates popular items penalty factor,the percentage of common interest items and other information,which provides an effective measurement criterion for more detailed description of the correlation between users.(2)This paper proposes and improves the similarity measure guidelines based on item quality differences.Taking the item perspective as an entry point and considering item quality bias as an important influencing factor,the dispersion coefficient is introduced to portray the dispersion degree of item ratings;by fusing with the percentage of common interest items and the classical correlation coefficient,a similarity measure criterion based on item feature analysis is provided for the portrayal of correlation relationships between users.(3)Nearest neighbor recommendation method based on user rating and item quality information fusion.Based on the information fusion of "user-item" two-dimensional perspective and the goal of minimizing MAE value,two weight measurement mechanisms of similarity are proposed.By fusing the similarity information based on user score and project quality,a nearest neighbor recommendation method is proposed to describe the characteristics of users and projects more comprehensively.Overall,this paper focuses on more detailed and comprehensive analysis and mining of user behavior and item characteristics,and establishes a neighbor-based collaborative filtering method that integrates user ratings and item quality information.Taking Movie Lens-100 K containing 943 users,1682 movies,and 100,000 data sets as the experimental data set,the effectiveness of the method in this paper is verified through experimental tests and comparative analysis.The research results are enriched and the method system of the recommendation system has been developed,which has application value in various recommendation scenarios such as e-commerce websites.