Abstract: Collaborative filtering algorithm suffers from sparsity and cold-start problems which affect the performance of recommendation system badly，which can not take advantage of the collaborative filtering.A new collaborative filtering optimization algorithm is introduced.The measurement of correlation-based similarity has higher quality of recommendation than that of cosine-based similarity.Deviation from mean remedies the disadvantage of new-item in cold-start problem，increases the quality and precision of recommendation.

Key words: personalized recommendation, collaborative filtering, similarity, deviation of mean, recommendation system

摘要： 基于用户的协作过滤推荐技术中存在的稀疏问题和冷开始问题，影响推荐系统的性能，使协作过滤的效果得不到充分的发挥，提出了一种优化的协作过滤推荐算法，对用户-项目评价矩阵进行降维预处理，对原始评价矩阵降低了噪音，有效地解决了稀疏问题，验证了使用相关相似度的度量标准所得到的算法要比余弦相似度算法的推荐质量要高，验证了采用中心加权求和的方法能很好地弥补协作过滤算法在冷开始中的新项目问题上的不足，可以提高推荐系统的推荐质量与推荐精度。

关键词: 个性化推荐, 协作过滤, 相似度, 中心加权, 推荐系统