Abstract：Traditional similarity calculation methods draw correlation information between users through scoring information, these methods only consider the similarity between users from user evaluation information, and the sparse data sets can influence correctness of the recommendations in the recommendation system. Aiming to the problem of data sparsity in the general collaborative filtering recommendation algorithm, an improved similarity collaborative filtering algorithm was proposed. In the scheme, the user similarity weight coefficient was introduced, and the Pearson correlation coefficient was weighted and combined with the Jaccard similarity method.In this way, the key performance of the similarity metric algorithm in the collaborative filtering algorithm was improved by the scheme. The experiments in the public data sets ( MovieLens and Book-Crossing set) showed that the improved algorithm could reduce the mean absolute error by 5. 2% to the maximum. The improved algorithm could also reduce the influence of sparse data set on the recommendation accuracy greatly, and improved the accuracy of the recommendation system significantly.

Keyword：recommendation system;collaborative filtering;similarity;mean absolute error;