Research On The Key Technologies Of Collaborative Filtering Recommendation System

Among current various recommendation algorithms,collaborative filtering recommendation algorithm is one of the most popular and most widely used recommendation algorithms.The traditional collaborative filtering recommendation algorithm has the problems of data sparsity,new user cold start problems,user interest drift,and inaccurate scoring prediction caused by different user rating scales.In recent years,there have been many clustering-based recommendation algorithms.The application of clustering in the recommendation system has greatly improved the performance of the recommendation system.This paper does some theoretical research on the recommendation system and clustering,and proposes some specific problems in the traditional collaborative filtering recommendation algorithm,such as data sparsity problem,scalability problem,and inaccurate score prediction caused by different user rating scales.improve proposals.The main research contents of this article are:(1)Aiming at the high sparsity and high dimensionality of the Movienlens data set,a new missing value filling method and dimension reduction method are proposed.First calculate the average Euclidean distance of the vector to get two similar vectors,then combine the mean of the two to fill in the target missing value and achieve the initial dimension reduction;(2)In order to further increase the scalability of the system and reduce the time lost in the implementation of the algorithm,CFSFDF clustering is introduced into the processing of data sets.At the same time,some improvements are made to the traditional CFSFDP clustering to eliminate the clustering process.Possible outliers in the cluster to achieve clustering for all users;(3)In order to reduce the influence of different user rating scales on similarity calculation results in the process of selecting neighbor user sets,this paper proposes a user similarity calculation method based on cosine distance and Euclidean distance;(4)In order to further improve the accuracy of the target project scoring prediction,this paper proposes a scoring prediction method that integrates the target users and the averagescore of neighboring users.This paper carries out corresponding experimental verification in MovieLens dataset,and proves the effectiveness of this algorithm.