Abstract: The structure between user and item is only considered in the network-based inference algorithm regardless of personalized preferences, collaborative tags contain rich information about personalized preferences and item contents, and then a hybrid recommendation method is proposed based on network and tag. In this paper, personalized preferences is constructed according to the method of TF-IDF and the tag support, and a linear combination of recommendation model is presented by merging network-based inference and personalized preferences. The benchmark data set, MovieLens, is used to evaluate the algorithm. Experimental results demonstrate that the usage of tag information can significantly improve accuracy, diversification and personalized of recommendations.

Key words: personalized recommendation, tag, TF-IDF, network-based inference

摘要： 基于网络结构的推荐算法利用用户与项目间的结构关系进行推荐，忽略了用户偏好，而项目的标签隐含了项目的内容及用户的偏好，提出一种基于网络结构和标签的混合推荐方法。算法根据用户选择项目的标签统计信息，分别采用TF-IDF和用户对标签的支持度两种方法构建用户偏好模型，与基于网络的推荐模型进行线性组合推荐。通过在基准数据集MovieLens上测试证明，该算法在推荐结果命中率、个性化程度、多样性等方面均优于基于网络的推荐算法。

关键词: 个性化推荐, 标签, TF-IDF, 基于网络的推荐