Abstract: The low accuracy problem for recommendation is a great challenge in traditional Collaborative Filtering（CF） algorithm. And there is difficulty to recommend new items for this method. So this paper introduces Kansei tags which is different from the existing tags in other platforms as a bridge between users and items and proposes a double-matrix model based on CF algorithm. Then using the praise pointing activity of users to complete this method. As a result, the experimental result shows that these two methods can significantly expand the recommending range and the completed method which contains the praise pointing behaviors can effectively improve the accuracy of recommendation.

Key words: Collaborative Filtering（CF）, recommendation algorithm, Kansei tags, double-matrix model, praise pointing

摘要： 传统的协同过滤算法主要通过已有项目的评分信息确定用户邻近集再进行评分预测，并以此进行推荐，这种推荐方法的推荐精度并不高。引入一个新的项目属性-意象标签作为用户与项目之间的联系纽带，在协同过滤算法的基础上提出一种双矩阵模型，并利用平台用户对于意象标签的点赞行为再次改进方法。实验结果证明，两种方法均大幅度提高推荐范围，且引入用户支持度的方法能够有效地提高推荐精度。

关键词: 协同过滤, 推荐算法, 意象标签, 双矩阵模型, 点赞行为