Research On User Personalized Recommendation Considering Rating And Popularity Bias

In the field of recommender systems,most scholars devote themselves to optimizing the recommendation model in order to improve the recommendation accuracy.In recent years.while meeting the recommendation results and accurately matching users’ needs,the bias problem in the recommender systems and the improvement of users’ personalized experience,including the degree of novelty and surprise,have gradually come into the research field of scholars.However,due to the rating bias caused by the user’s own evaluation behavior and the popularity bias generated by the recommendation model in the recommender systems,the recommendation results tend to be fuzzy or concentrated on a few high popularity items,thus affecting the recommendation results and reducing the user’s personalized experience.Based on this situation,this paper takes the public data set as the research object,deeply studies the theory and method of unbiased recommendation of items in personalized recommendation,and proposes a series of solutions to the problem of rating bias and popularity bias,which strikes a balance between improving the recommendation accuracy and providing users with better personalized experience services.The main contents of this paper include the following points:Firstly,aiming at the problem of scoring bias caused by users’ fuzzy evaluation of items and low accuracy of model recommendation,a collaborative filtering recommendation model based on single-valued neutrosophic sets was proposed.Based on the user-items scoring matrix,this paper introduces the single-valued neutrosophic sets,designs the rating membership transformation formula,and integrates the extreme evaluation and Jacquard correlation coefficient to obtain the collaborative filtering recommendation model based on the singlevalued neutrosophic sets,which can effectively reduce the influence of fuzzy evaluation,reduce the rating bias and improve the recommendation accuracy.Secondly,to address the issue of popularity bias in collaborative filtering recommendation models based on matrix factorization,a debiasing model has been proposed that compensates for user preferences and ratings.This model incorporates causal inference,and theoretical analysis indicates that it can significantly enhance the recommendation effectiveness.Empirical results also demonstrate that the model effectively mitigates prevalence bias and improves recommendation utility.Finally,the impact of rating bias and popularity bias on the recommender systems was reduced,so that the recommendation model could balance the recommendation utility and debias effect at the same time,and a debias model with mixed rating bias and popularity bias was proposed.The fuzzy logic was introduced,and the K-order parabolic fuzzy distribution was used to redistribute the rating by integrating the age of users,so as to reduce the rating bias.The loss function was optimized by using the continuously increasing traffic and popularity of the item to reduce the rating bias and popularity bias.User emotion and item popularity were combined to construct user psychological tendency and improve the recommendation utility.Experiments were conducted on actual datasets to authenticate the model’s performance.The findings demonstrate that the proposed model not only significantly mitigates recommendation bias but also ensures the usefulness of the recommendations.