融合多元信任机制的协同过滤算法

A Collaborative Filtering Algorithm by Fusing Multiple Trust Mechanism

针对协同过滤算法的数据稀疏性问题,融合信任网络,提出了一种融合多元信任机制的协同过滤算法。首先,依据社会学中信任的定义,结合推荐系统中可利用的数据信息,对影响信任的主要因素(基本信任度、可靠度、影响力与自我取向)进行合理的量化,构建多元信任模型;其次,利用信任的可传递性,提出间接信任度计算公式,对用户信任矩阵进行有效扩充;最后,以信任度取代协同过滤算法中的相似度产生目标用户最近邻,进行推荐。最终,以MAE值为评价标准,通过与其他相关推荐算法的对比实验,结果表明该算法能够有效避免数据稀疏带来的推荐效果不佳问题,推荐精度得到了明显提高。

In order to solve the problem of data sparsity in collaborative filtering algorithm,we put forward an improved collaborative filtering algorithm by fusing multi trust mechanism.Firstly,we build multiple trust models in recommender system according to the definition of trust in sociology by quantifying the main factors that affect trust including basic trust degree,reliability,influence,and self orientation.Secondly,we propose the formula for calculating the degree of indirect trust and extend user trust matrix by using trust transitivity.Lastly,the user nearest neighbors are generated by trust degree instead of similarity,so as to get recommendation results.Finally,with MAE value as the evaluation criterion,through the comparison experiment with other relevant recommendation algorithms,it shows that this algorithm can effectively avoid the problem of poor recommendation effect caused by data sparsity,and the recommendation accuracy has been significantly improved.