一种利用非对称相似度强化信任用户关系的推荐算法

Enhancing Trust Networks Using Asymmetric Similarity for Recommendation

为提高推荐算法的准确率,针对Social MF中用户将其信任用户同等对待的问题,提出一种在评分数据稀疏情况用于计算信任用户相似度的方法—非对称相似度方法(AC-Sim),通过AC-Sim来判别存在信任关系用户间是否有共同偏好,并将此偏好信息融合到已有的用户关系网中,达到强化信任网络的目的;其次将强化后的信任网络应用到PMF算法中,评分矩阵在分解过程中,用户特征向量受信任用户影响的同时,也受到与其有共同偏好用户的影响.实验结果表明,与目前较为流行的算法相比,新算法在RMSE和MAE上均取得更好的推荐效果.

In order to improve the accuracy of recommendation algorithms,aiming at,this paper firstly presents a novel method—Asymmetric Similarity（ AC-Sim）,w hich can be applied to calculate the asymmetric similarity of users. The information of common preference betw een users w hich can be found through the AC-Sim,is added into the existing trust netw ork. Secondly,Asymmetric Similarity is applied to the algorithm of Social M F to improve its performance. In the process of factorization of the user-item rating matrix,the user latent feature vectors are effected by relationship betw een trust users,and are also depended on their common preference Experimental results on real w orld product rating data set of Epinions show that the proposed approaches outperform the state-of-the-art algorithms in terms of RM SE and M AE and the trust enhanced can further improve traditional recommendation methods.