A Collaborative Filtering Algorithm for Improving Similarity Based on Probability Weighted Association Rules

The sparsity of data and the accuracy of similarity calculation have always been the focus of collaborative filtering algorithms. To address this issue, a collaborative filtering algorithm PWAR-CF (The combination of probability weighted association rules in collaborative filtering) which combines probability weighted association rules is proposed. The algorithm employs probability-based weighted association rules to extract the potential correlation among users, and obtains the weighted confidence between users. According to the weighted confidence, the predictive score matrix is used to fill the original score matrix to alleviate the problem of sparse data. The weighted confidence is fused with the Person correlation coefficient, and enhanced similarity formula is utilized to compute the updated similarity between users. Finally, according to the prediction scoring formula, recommendations are generated for users. The Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) indexes are used on MovieLens data sets to prove that the proposed algorithm effectively alleviates the problems of data sparsity and similarity calculation accuracy, and has better recommendation quality than traditional algorithms.