Research Of Collaborative Filtering Algorithm Based On Clustering

The cold start-up problem of collaborative filtering algorithm makes it difficult to make personalized recommendations based on user preferences,resulting in user loss.The existing data sparsity problem leads to the inaccuracy in calculating the similarity between users or projects,which reduces the recommendation accuracy.The problem of scalability results in large search space and time consuming when calculating the nearest neighbor.In order to alleviate the impact caused by the existing problems of collaborative filtering,the following work is done in this paper for the existing problems of data sparsity and scalability:In view of the problem of data sparsity in the algorithm,this paper adopts weighted Slope One algorithm to fill the gap of scoring matrix based on the idea of pre-estimated filling.The item similarity in the item type matrix and the item similarity in the item scoring matrix were calculated,and the final item similarity was obtained through linear combination of the two parameters.The first M items with high similarity were taken as neighbor items,and the grading items were estimated by weighted Slope One algorithm,and the user scoring matrix was filled in.The introduction of project type similarity can avoid the interference of projects with low similarity and improve the accuracy of the estimation.This paper introduces the idea of clustering to solve the scalability problem of the algorithm.According to the scoring matrix,the number and score value of the rated items are obtained.Then the graded project types are obtained according to the project type matrix;The user preference model was extracted by calculating the degree of interest,and the k-means algorithm for optimizing the clustering center was used for user clustering.Calculate the distance between the target user and the clustering center point,find the cluster,and the points in the cluster serve as neighbor users.Combined with the filled scoring matrix,the improved formula is used to calculate the similarity between the target user and the neighbor user and predict the score.Clustering can reduce the search space of target users and reduce the time complexity of the algorithm,so as to improve the scalability of the system.In this paper,based on MovieLens data set,the effectiveness of prediction module based on item attribute similarity weighted Slope One algorithm in collaborative filtering and the comparison of recommendation accuracy between this algorithm and other algorithms were experimentally verified.The experimental results show that the algorithm proposed in this paper is effective and feasible,which improves the accuracy of prediction and the quality of recommendation.