Collaborative filtering recommendation based on conditional probability and weight adjusting. (26th January 2015)

• Record Type: Journal Article
• Authors: Wu, Haitao; Chou, Wen–Kuang; Hao, Ningbo; Wang, Duan; Li, Jingfu
• Is Part Of: International journal of computational science and engineering. Volume 10:Number 1/2(2015)
• Page Start: 164
• Page End: 170
• Publication Date: 2015-01-26
• Subjects: collaborative filtering recommendation -- item–based filtering -- conditional probability -- weight adjusting -- mean absolute error -- MAE -- recommender systems -- similarity weights; Computer science -- Mathematics -- Periodicals; Computer simulation -- Mathematical aspects -- Periodicals; Computational intelligence -- Periodicals; 004.015105
• Languages: English
• ISSNs: 1742-7185

Abstract

Collaborative filtering recommendation algorithm is one of the most successful technologies for building recommender systems. However, a user–based collaborative filtering method has its limits related to similarity and ratings. To avoid those limits, we propose a new item–based collaborative filtering algorithm based on conditional probability and weight adjusting in this paper. At first, any two items are selected to compute the similarity from common user ratings, and only the items with the similarity greater than preset thresholds are chosen as the set of supporting items. Then an integral parameter, that is the frequency of two items present simultaneously, is used to adjusted similarity weights. Finally, a new algorithm combining conditional probability and weight adjusting is proposed to predict ratings. The experimental results show the proposed algorithm is feasible and effective in practice.