Research On Personalized Recommendation Algorithm Based On Rating System

Recommendation system is an important part of each e-commerce platform,and the accurate recommendations,which without doubt plays a significant role on the promotion of product marketing.As the key and core technology of recommendation system,recommendation algorithm is highly emphasized and widely concerned by scholars and engineers in the field of network science and e-commerce,and its research has important theoretical significance and economic value.Since the emergence of recommendation systems,a number of algorithmic techniques have emerged,however,there are some limitations to the various existing algorithms.For example,the most widely used Collaborative Filtering(CF)has a strong dependence on the interaction information between users and items,and its accuracy of recommendations is greatly reduced when the data is sparse.Moreover,in order to better analyze users’ preferences and needs,many online platforms have a rating system,such as users’ ratings of movies in movie platforms,and users’ ratings of songs in music platforms.How to use this rating information to make personalized recommendations is an important issue in the research of recommendation algorithms.There is a relative lack of research on recommendation algorithms specifically for rating systems,and most of the relevant literature only makes use of high-rating information,while ignoring the intrinsic information implied by low-rating data,which may reduce the accuracy of recommendations.Therefore,this paper focuses on the scoring system and proposes several improved personalized recommendation algorithms for the above problems by combining K-means clustering method,vector Euclidean distance thresholding method,and complex network node similarity method.Firstly,a personalized recommendation algorithm(ICF algorithm)based on reduced item space on the scoring system is proposed.On the one hand,ICF algorithm improves the data sparsity problem to some extent by reducing the item space and projecting the sparse information of individual items onto a finite class of item space.On the other hand,K-means clustering method is used to extract the similarity information among users twice,and a weighted similarity metric is proposed,which can extract and portray the similarity of interests among users more accurately,and thus ICF algorithm can obtain more accurate recommendation results.With the original user-item data is obtained,the recommendation list of target users would be generated automatically while the computer executes the process of ICF,which would improve the implementation efficiency of the recommendation system.Secondly,the personalized recommendation algorithm based on user interest association network(UNP algorithm)and personalized recommendation algorithm based on object interest association network(ONP algorithm)are proposed for the scoring system in this paper.In order to achieve personalized recommendations for targeted users,UNP algorithm and ONP algorithms filter out strong correlations between objects by calculating the Euclidean distance of vectors and setting thresholds on the basis of making full use of user rating information.After that,user interest association networks and item interest association networks are constructed by the two algorithms respectively,and then the structural information of the bipartite networks is combined to calculate the ranking of items of interest to the target users.Finally,in the purpose of examining the recommendation effectiveness,the ICF algorithm,UNP algorithm,and ONP algorithm are tested on Netflix and Movie Lens datasets with corresponding performance.From the experimental comparison results with several classical recommendation algorithms,the experimental results show that the recommendation quality of the ICF algorithm and UNP algorithm has been significantly improved.