Abstract: In order to alleviate the sparsity of the user project rating matrix, the characteristics of the project are analyzed and the project characteristic matrix is introduced based on the traditional scoring matrix of collaborative filtering, then the similarity of the users is calculated based on the similarity of the cosine and the similarity of the user preference to the project attributes, and the importance of both is calculated by a weight. A collaborative filtering algorithm based on item attribute preference is proposed. The results show that when the similarity of the cosine and the similarity of the user preference to the project attributes are equal, the recommendation quality of the recommendation system is the best. When the scoring matrix is ??sparse, the greater the proportion of the similarity of the user preference to the project attributes, the better the quality of recommendation can be achieved. Meanwhile, the collaborative filtering algorithm based on user preferences for project attributes in [MAE] is smaller than the other two traditional collaborative filtering algorithms.

Key words: e-commerce, recommendation system, item attribute, collaborative filtering, recommendation accuracy

摘要： 为了缓解用户项目评分矩阵数据的稀疏性，在传统的协同过滤项目评分矩阵的基础上，对项目的特征进行分析，引入项目特征矩阵，然后结合余弦相似性和基于用户对项目属性偏好相似性综合计算用户的相似性，并通过一个权值来控制两者的重要程度，提出了一种基于用户对项目属性偏好的协同过滤算法。研究结果表明余弦相似性和用户对项目属性偏好的用户相似性比重相等时，推荐系统的推荐质量最好；而且当评分矩阵越稀疏的时候，用户对项目属性偏好的用户相似性的比重越大越可以提高推荐质量；同时提出的基于用户对项目属性偏好的协同过滤算法在[MAE]值都要小于两种传统的协同过滤算法。

关键词: 电子商务, 推荐系统, 项目属性, 协同过滤, 推荐精度