Abstract: To solve the problem of not considering the ratio of co-rated items and the user preference in some traditional similarity measures, two weighting factors, asymmetry factor and preference factor, are added to improve the accuracy of similarity computation. The experimental results on two public datasets, Movielens and Yahoo Music, show that the proposed model significantly lowers the prediction error and has better performance than other similarities. The model with these two factors can better reflect the rating difference between users, and effectively deal with the user preference effectively. Thus, it improves the recommendation quality.

Key words: user preference, asymmetry, collaborative filtering, recommendation algorithm

摘要： 为解决常见的相似性方法存在未考虑用户间共同评分项在目标用户所评项目中的比例以及用户评分偏好的问题。提出了非对称因子和偏好因子，用于提高用户相似性计算的准确性。在公开的MovieLens和Yahoo Music数据集上的实验表明，引入这两个因子后，相似性模型的预测误差下降显著,优于其他相似性方法。非对称因子和偏好因子的引入更合理地体现出用户间的评分差异性，有效地处理了用户偏好问题，提高了推荐质量。

关键词: 用户偏好, 非对称, 协同过滤, 推荐算法