Abstract: Collaborative filtering has been widely used in the recommendation system, but the traditional  algorithm has some limitations, such as inability to adapt to the sparsity of user-item rating matrix data sets well, failure to consider the classification of item, users-scores, interest change over time and other factors when calculating the similarity of the items. Regarding these limitations, it proposed a big data set hybrid dynamic collaborative filtering algorithm, based on the traditional collaborative filtering algorithm. When calculating the similarity of items, time decay functions were introduced in the algorithm, which considered both the similarity of items, scores and items classified. The weights of project integrated similarity could be adjusted automatically. In the algorithm, some improvements have also been made in similarity computing and the selection of the neighboring items. To verify the effectiveness of the algorithm, experiments were done on movielens data sets. Experimental results show that the algorithm is better than the traditional recommendation algorithms.

Key words: collaborative filtering, recommendation system, item classification, time weight