关键词: 推荐系统, 数据稀疏性, 标签系统, 因子分析, 评分预测

Abstract:

Item (or information) recommendation is one of hot research topics currently. However the issue of sparseness in dataset challenges all traditional recommendation algorithms. Limitations of knowledge representation in traditional recommendation algorithms were studied. The tag-system-based knowledge to represent information of each user's behavior was proposed. That it the account on user's behavior on items is transferred to an account on a user's behavior on tags. To decrease the computation complexity on high dimensional tag-based datasets, a factor analysis method was taken to extract those most important latent factors to represent users' behaviors. Based on each user's representing vector of latent factors, a new way was given to compute similarities among users. By incorporating this similarity measure, a new collaborative recommendation method with low sensitivity to sparseness was built to meet the need of practical and dynamic datasets. Experiments were carried on real-world datasets to compare the proposed method with other state-of-the-art collaborative filtering and matrix factorization based recommendation methods. It is shown the proposed method can achieve better prediction accuracy while keeps a lower sensitivity to sparseness.

Key words: recommendation system, dataset sparseness, tag system, factor analysis, rating prediction