Collaborative filtering methods based on user relevance degree and weights of recommend-items
Suhuan Sun, Gongsheng Kong, Changwei Zhao · 2011
The way to measure similarity among users is a key factor affecting accuracy of recommendation algorithm. A similarity measuring algorithm based on degree of correlation among users and weighting of recommended items is proposed to improve less marking aspects between users and lower prediction accuracy of traditional similarity measuring algorithms. The method can solve the non-convex problem generated by non-exclusive parameters in the objective function by introducing an adjusting parameter of similarity and alternatively optimizing the relative parameters using least squares. Experimental results show that the proposed method is effective in improving the accuracy of whole system and the prediction accuracy of recommendation items.