Abstract:

In order to further improve the accuracy of similarity calculation, we propose a collaborative filtering recommendation algorithm to optimize the combination of similarity. Firstly, we establish a matrix of time. According to the time sequence that users score for the items, it can calculate the influence between users. Secondly, according to score differences on the common items, we can calculate the weighted information entropy of score difference. Finally, the influence of temporal behavior is incorporated into the similarity based on weighted information entropy. Fusion parameters therein are selected by the stochastic particle swarm optimization algorithm. In comparison with several other similarity calculation methods, the proposed algorithm reduces the normalized mean absolute error and the popularity. To some extent, it reduces the effect of data sparsity and is more accurate in similarity calculation, thus improving the quality of recommendation.

Key words: collaborative filtering recommendation algorithm, temporal behavior influence, information entropy, particle swarm optimization algorithm