Research On Collaborative Filtering Recommendation Algorithm Based On Revised Score

While the rapid development of the Internet brings convenience to people,it also generates a lot of data redundancy.How to retrieve the data required by users and accurately recommend them to users is a problem that is currently facing.The recommendation system,as an effective means of information filtering,has gradually become a popular research object in the field of computer science and applications,and is widely used in Internet systems.Traditional recommendation algorithms are divided into user-based recommendation algorithms and item-based recommendation algorithms,while collaborative filtering recommendation algorithms are widely used as classic userbased recommendation algorithms,but still face problems such as low recommendation efficiency,sparse matrix,and user anomalies.Value,transaction multi-classification and other issues.Therefore,this paper conducts an in-depth study of the traditional collaborative filtering algorithm,and improves it from two aspects: the revision of the scoring matrix and the multi-dimensional attributes.The revision of the scoring matrix is used to solve the difficulty of searching for nearest neighbors,user outliers,and multidimensional transaction recommendations in traditional algorithms.The work of this paper is as follows:(first)Based on the traditional algorithm,this paper proposes a collaborative filtering algorithm with revised scores for the problems of low recommendation efficiency of collaborative filtering algorithm,sparse matrix,and user outliers.The algorithm firstly improves the search efficiency of the user's nearest neighbors through clustering.In view of the fact that weighted scoring in traditional algorithms cannot solve the problem of outliers,this paper proposes a modified score as a measurement index to solve the interference of user outliers on the accuracy of the recommended algorithm.Since the traditional algorithm only considers the correlation between users and users,and does not consider the correlation between transactions,the recommendation results are distorted when recommending to users.In order to solve this problem,this paper proposes a collaborative filtering algorithm with revised scores based on the confidence in association rules and weighted scores.The algorithm improves the recommendation accuracy by solving the problem of user outliers.(second)Because in the face of transactions with multi-dimensional attributes,the accuracy of the collaborative filtering algorithm with modified scores is low.This paper improves the collaborative filtering algorithm of revised scoring,and proposes a recommendation algorithm for multidimensional division of transactions.The algorithm gives the user's preference coefficient for the attribute category through the transaction multidimensional division strategy,and makes top-N recommendation to the user in combination with the modified scoring index.Experiments prove that the algorithm solves the problem of multi-dimensional attribute classification of transactions well,and also guarantees the recommended accuracy of the algorithm.Experiments show that the collaborative filtering algorithm and the multi-division recommendation algorithm proposed in this paper can effectively provide personalized recommendations for users.