中文摘要：

个性化推荐系统中使用最广泛的算法是协同过滤算法,针对该算法存在的数据稀疏和扩展性差问题,提出了一种基于用户兴趣和社交信任的聚类推荐算法。该算法首先基于聚类技术根据用户评分信息将具有相同兴趣的用户聚为一类,并建立基于用户兴趣相近的邻居集合。为了提高兴趣相似度计算的准确性,采用了修正余弦计算公式来消除评分标准的差异问题。然后,引入信任机制,通过定义直接信任、间接信任、传递路径和计算方法来度量社交网络用户之间隐含的信任值,将社交网络转换为信任网络,依据信任程度来创建基于社交信任的邻居集合。通过加权的方式将基于两种邻居集合的预测值融合起来为用户产生项目的推荐。在Douban数据集上进行仿真实验,确定了最优的协调因子值和分类数值,并与基于用户的协同过滤算法和基于信任的推荐算法进行对比,实验结果表明,所提算法的平均绝对误差（MAE）减少了6.7%,准确率（precision）、覆盖（recall）和F1值分别增加了25%、40%和37%,有效提高了推荐系统的推荐质量。

英文摘要：

Collaborative filtering algorithm is the most widely used algorithm in personalized recommendation system.Focusing on the problem of date sparseness and poor scalability,a new clustering recommendation algorithm based on user interest and social trust was proposed. Firstly,according to user rating information,the algorithm divided users into different categories by clustering technology,and set up a user neighbor set based on interest. In order to improve the accuracy of the calculation of interest similarity,the modified cosine formula was used to eliminate the difference of user scoring criteria.Then,the trust mechanism is introduced to measure implicit trust value among users by defining the direct trust calculation method and indirect trust calculation method,converted a social network to a trust network,and set up a user neighbor set based on trust. Finally,this algorithm combined the predictive value of two neighbor sets to generate recommendations for users by weighting method. The simulation experiment was carried out to test the performance on Douban dataset,found suitable value of α and k. Compared with collaborative filtering algorithm based on users and recommendation algorithm based on trust,the Mean Absolute Error（ MAE） decreased by 6. 7%,precision,recall and F1 increased by 25%,40% and 37%.The proposed algorithm can effectively improve the quality of recommendation system.