An improved Collaborative Filtering combined with confidence function and user rating preference
Jihong Li, Qing Li, Chong Shao, Mengke Yao · 2013
Collaborative Filtering(CF) is one of the most significant and successful algorithms in the field of personal recommender system. The key aspect of this algorithm is the calculation of similarity. Due to the problem of data sparsity, traditional similarity metrics (such as cosine distance, pearson correlation coefficient) fail to measure the similarity between two users exactly when the number of items co-rated by the two users. In this paper, by analyzing the original data sets, a confidence function has been introduced to mitigate the shortage of traditional similarity metrics. Meanwhile, the user-type average ratings are used to characterize rating preference of users and substitute average ratings of users in the phase of predicting ratings. The experimental results show that both the new similarity metrics and predict ratings strategy are effective to improve the recommender results.