关键词: 关键词: 项目相似性, 标签, k近邻, 协同过滤推荐算法

Abstract: Abstract: With the application of personalized recommendation system, a more accurate and stable recommendation algorithm is needed. A new recommendation algorithm called M2_KSP is designed in this paper. The algorithm first defines a new M2 similarity calculation method combined with item tag, and solves the problem that the traditional similarity calculation method ignores the item attribute, then refers to Slope One weighted algorithm theory to define the new rating prediction method combined with this similarity. Firstly, according to the M2 similarity defined by the number of important tag of item, the M2_KSP algorithm finds the k nearest items. Then, based on the user rating of the k nearest items, the new rating prediction method is used to predict user ratings. In order to verify the accuracy and stability of the algorithm, the comparison experiments with other recommended algorithms were carried out by using the MovieLens data set. The other recommended algorithms include KNN algorithm based on Cosine distance and so on. The results show that the algorithm is more accurate, better and more stable than the other recommended algorithms.

Key words: Keywords: item similarity, label, k nearest neighbor, collaborative filtering recommendation algorithm