Abstract: To overcome the data sparseness and ignoring items, user feature of traditional collaborative filtering algorithm which can lead to the decrease of efficiency of recommendation, this paper proposes a collaborative filtering recommendation algorithm based on safe and high confident semi-supervised method S4VM to predict the rating of unrated items effectively. Through predicting the rating of unrated items, the sparseness of data has been alleviated and the accuracy of searching nearest neighbor item has been improved simultaneously. The experimental results show that the algorithm can improve the recommended quality of the system.

Key words: collaborative filtering, recommendation system, Safe Semi-Supervised Support Vector Machine（S4VM）, semi-supervised learning, confidence

摘要： 为解决传统协同过滤算法中用户评分数据稀疏性，忽视物品及用户特征，所带来的推荐质量下降的问题，提出了一种基于安全的、高置信度的半监督方法的协同过滤推荐算法，采用安全的，高置信度的半监督方法S4VM对没有评分的数据进行有效预测，同时考虑用户的行为信息以及物品及用户特征。通过对未评分数据进行预测，能够有效地缓解数据的稀疏性，从而提高寻找最近邻的准确度。实验结果表明，该算法能够有效地提高系统的推荐质量。

关键词: 协同过滤, 推荐系统, 安全的半监督支持向量机（S4VM）, 半监督学习, 置信度