An improved collaborative filtering recommendation algorithm based on reliability
Shiping Fan, Hao Yu, Haihui Huang · 2018
The similarity of the user-based collaborative filtering algorithm is based on the user-item rating matrix. However, the accuracy of predicting the users' rating was affected by the data sparseness problem. To alleviate the data sparseness problem of the collaborative filtering and the variant ratings on the common items. In this paper, we propose a similarity calculation measure that based on reliability. Firstly, the measure makes use of the users' ratings on the common items to obtain the credibility of rating between users, and then introduces the credibility into the adjusted cosine similarity to alleviate the affect of variant ratings on common items. Secondly, the punishment function was introduced to the adjusted cosine similarity, and alleviate the influence of the popular items on the similarity calculation. Finally, the former two kinds of similarity were measured comprehensively to predict the user's rating more accurate and to improve the reliability of the similarity calculation. Experimental results show that compared with the Pearson similarity and the adjusted cosine similarity, the improved method proposed in this paper could get a lower value of MAE, which means that it could improve the accuracy of predicting users' rating and the personalized recommendation efficiency.