Research On Collaborative Filtering Algorithm Based On Clustering And User Preference

The recommendation algorithm is the core of the personalized recommendation system,and the accuracy and scalability of the algorithm directly affect whether the recommendation system can be put into use.After years of research and development,various recommendation algorithms have been derived from various circles,among which the collaborative filtering algorithm is the most widely used.However,with the change of the application scene,the problems of the original data matrix sparse,cold start and expansion of the algorithm are gradually emerging.In view of the above problems,this paper studies the recommendation algorithms,the main research contents are as follows.Firstly,for the problem that the original data involves a wide range of recommendation algorithm and affects accuracy and efficiency,a collaborative filtering algorithm based on fuzzy C-Means Clustering is proposed.The algorithm optimizes the clustering results by optimizing the process of distance calculation and cluster center selection in fuzzy clustering.The recommendation algorithm of the fusion clustering process can not only narrow the calculation range,but also reduce the interference of redundant data on the recommendation results,and improve the accuracy and execution efficiency of the recommendation results.Secondly,for the problem of sparseness,cold start and actual factor interference of the original data matrix faced by the traditional recommendation algorithm,a collaborative filtering algorithm based on matrix filling and user preference is proposed.The algorithm improves the existing algorithms by decomposing the low rank matrix to fill the sparse matrix,improving the similarity calculation formula,and optimizing the prediction scoring formula.Then,based on the above two improved algorithms,a collaborative filtering algorithm based on clustering and user preference is proposed.At the same time,the spark big data computing platform is proposed for the scalability of the algorithm,and the improved algorithm is parallelized on the Spark platform.Finally,using the time-stamped music dataset as the data source,the improved algorithm proposed above was tested,Through experimental results to show the algorithm's accuracy and efficiency.