Research On Collaborative Filtering Algorithm Based On Frequent Itemsets Mining And User Clustering

Collaborative filtering technology has data sparsity and scalability problems,which can easily lead to low recommendation accuracy and recommendation efficiency,and severely limit the development of recommendation technology.In response to the above problems,this paper uses frequent pattern mining technology and clustering technology to improve the collaborative filtering algorithm.The main research contents are as follows:First,this paper combines the association rule algorithm to propose a scoring filling matrix method based on frequent itemsets mining.By mining the potential correlations between items,predicting the scoring value of unscored items is used to reduce the sparsity and filling error of the scoring matrix.After experimental evaluation,the sparsity of the scoring matrix after filling is reduced by about 7% compared with the traditional collaborative algorithm,and the scoring error value is reduced by about2%.Second,this paper proposes a recommendation algorithm based on TF-IDF(term frequency-inverse text frequency index)and user clustering.Using the TF-IDF algorithm,the user-item-characteristic TF value matrix and TF-IDF value of the item are obtained.Combine the above matrix with user identity attribute information,use a clustering algorithm to divide the user set,and use the characteristic TF-IDF value to improve the similarity calculation formula to generate a recommendation list.After experimental evaluation,compared with the traditional collaborative filtering algorithm,the time required to calculate the nearest neighbor set is reduced by about half,and the accuracy of the recommendation result is increased by about 3%.Third,combining the above two improved algorithms,this paper proposes a collaborative filtering algorithm based on frequent itemset mining and user clustering,which is used to alleviate the data sparsity and scalability issues of collaborative filtering at the same time.The nearest neighbor sets obtained by the above two algorithms are merged,and the nearest neighbor of the user with higher similarity is selected for recommendation.The proposed algorithm is compared with the above two algorithms,the traditional collaborative filtering algorithm,and the algorithm proposed in related documents.In different situations,the algorithm proposed in this paper is better than the above algorithm.Finally,a movie recommendation system is designed by using the algorithm given in this article.The system can realize movie recommendation,movie search,movie details query,movie scoring and other functions,which reflects the practical application value of the algorithm.