一种对数据集稀疏度不敏感的协同推荐新方法-南京大学pasa大数据.pdf

一种对数据集稀疏度不敏感的协同推荐新方法 蔡国永 吕瑞 (桂林电子科技大学广西可信软件重点实验室 桂林 541004) (ccgycai@ ) A Novel Collaborative Recommendation Method with Low Sensitivity to Data Sparseness Cai Guoyong and Lv Rui (Guangxi Key Lab of Trusted Software, Guilin University of Electronic Technology Guilin, 541004) Abstract In the field of recommendation systems, most studies on the sparseness are based on static datasets. However, the datasets in practical application are dynamic and there exist at least two features: one is the increasing scale of User-Item matrix for users and items join into the datasets continuously; another is the increasing degree of sparseness. As a result, the accuracy of traditional recommendation methods will reduce gradually with the datasets becoming sparser. Based on the above consideration, a novel collaborative recommendation method with low sensitivity to different degrees of sparse datasets is proposed to meet the need of practical and dynamic datasets. This method incorporates tag information and factor analysis method to discover the most similar top-N users based on the similarity of users’ inner idiosyncrasies. Based on the most similar top-N users discovered, an improved collaborative filtering method is designed. Extensive experiments are conducted to compare the proposed method with other state-of-the-art collaborative filtering and the matrix factorization methods. The results demonstrate that our proposed method can achieve better accuracy and has a low sensitivity to different degrees of sparse datasets. Key words recommendation system；dataset sparseness ；tag system ；factor analysis ； rating prediction 摘 要 在推荐系统领域中，针对数据集稀疏性问题的研究大都建立在静态数据集的基础上，而实际工业 应用中的数据集则往往是动态的并且具有以下两个明显的特征：1) User-Item 矩阵维度逐渐增大；2) 稀疏 程度不断增加．因此，传统