Abstract:

Collaborative filtering algorithms are one of the most successful techniques which is widely used in the field of recommendation system. However, with the rapid growth in the number of users and the lack of corresponding rating data, the problem of data sparsity in recommender systems is becoming more and more obvious, which affects recommendation quality and efficiency seriously. To solve the sparse data problem in traditional collaborative filtering algorithms, we propose a method based on gray correlation degree, combination of tag overlap factors and user ratings, which is used for the standardization of the data processing of the user rating matrix, thus obtaining the user correlation degree and the association degree matrix. Clustering based on gray association degree on the correlation matrix of users can reduce computational complexity. Tag overlap factors are also introduced to improve the accuracy of the similarity between users in traditional collaborative filtering algorithms. Experimental results show that the proposed algorithm can greatly improve the prediction accuracy.

Key words: collaborative filtering;gray correlation degree;factors of tag overlap