Research On Personalized Recommendation Based On Item Popularity

Recommendation systems can effectively alleviate the information overload problem existing in today’s Internet world,and the merits of the recommendation algorithm will directly affect the alleviation effect of the recommendation system,so the selection of the recommendation algorithm is crucial.The earliest and widely used collaborative filtering algorithm has the advantages of discovering users’ potential preferences,scalability,and wide range of applications,but it also has certain shortcomings: sparsity problem of user rating data,popularity bias problem,group bias problem,etc.In this paper,we mainly combine item popularity,i.e.how much users pay attention to the item,to make recommendations.The recommendation algorithm is oriented to a certain user or some users,emphasizing the personalization of user preferences and the precision of recommendation results,ignoring the importance of social uniformity;the popularity is oriented to all users,reflecting social uniformity and universality,ignoring the individual needs of users.If we focus only on personalization and precision,it is easy to form an information cocoon,which may cause serious derailment between users and society;if we focus only on uniformity and universality,we may not be able to locate the actual needs of users and reduce social activism,therefore,it is necessary to integrate the popularity of items for relevant recommendations.The specific research contents of this paper are as follows:(1)To alleviate the popularity bias problem and user rating data sparsity problem,a new improved collaborative filtering recommendation method with fused item popularity penalty factors is proposed.The method is mainly divided into two parts: the first part is the pre-population part of the rating data,using the pre-rating method based on the dichotomous strategy of user preferences and average rating fusion to pre-populate the user rating data;the second part is the fusion item penalty factor for recommendation part,introducing item popularity,dichotomizing items through the popularity threshold,i.e.,popular items and non-popular items,and optimizing the collaborative filtering recommendation method by fusing item popularity penalty factor for popular items.The method is experimentally validated on the MovieLens dataset,and the experimental results show that the proposed method can effectively reduce the error of rating prediction and improve the recommendation accuracy.(2)To address the problem of group bias in item popularity and item quality,a recommendation method that fuses item local popularity and local quality is proposed.First,the user-item type preference matrix is constructed using item data and user rating data;then,the user-item type preference matrix is used to cluster users,form different user groups,and calculate the local popularity and local quality corresponding to each item in each user group;finally,the local popularity and local quality are fused to optimize the collaborative filtering method.Experimental results on publicly available datasets show that the proposed method outperforms the comparison algorithm in terms of precision and recall,and can improve the accuracy of item local popularity and local quality metrics to a certain extent.