Abstract: Collaborative filtering recommendation system suffers from series data sparsity problem. To solve the problem, this paper proposes a collaborative filtering recommendation method by combining rating preference and dual prediction. In the stage of calculating the nearest neighbors, to improve the calculation method of similarity, rating preference is introduced firstly. Then, in the stage of generating recommendation, a dual prediction method is proposed which is based on the user and the item nearest neighbors to predict the user preference more accurately. The experimental results on the MovieLens-1M data set indicate that the proposed method can relieve the influence of rating data sparsity on recommended results, significantly reduce the mean absolute error and effectively improve the recommendation precision.

Key words: recommendation system, collaborative filtering, user preference, rating prediction

摘要： 协同过滤推荐算法面临着严重的数据稀疏性问题，提出一种融合评分倾向度和双重预测的协同过滤推荐算法以解决该问题。在选择最近邻阶段，引入评分倾向度来改进相似性度量方法，更加准确地得到最近邻居集；在推荐生成阶段，利用基于用户最近邻和基于项目最近邻的双重预测方法来进行评分预测，提高预测的准确度。通过在MovieLens-1M数据集上的实验结果表明：该算法能够缓解数据稀疏性对推荐结果的影响，有效降低平均绝对误差，提高推荐准确率。

关键词: 推荐系统, 协同过滤, 用户偏好, 评分预测