一种基于领域信任及不信任的奇异值分解推荐算法

Recommendation Algorithm with Field Trust and Distrust Based on SVD

传统协同过滤算法存在数据稀疏与冷启动问题,社会化推荐算法虽然能在一定程度上缓解这些问题,但大多数的算法都只从单一的角度来衡量信任关系的影响。为了更准确地度量社交关系对推荐预测的影响,提出了一种基于领域信任及不信任的社会化奇异值分解(Field Trust and Distrust based Singular Value Decomposition,FTDSVD)推荐算法。该算法在SVD推荐算法的基础上加入了用户的信任关系与不信任关系,利用不信任关系对社交关系进行修正,并且充分考虑用户的信任领域相关性和全局影响力。在Epinions数据集上将FTDSVD算法与相关算法进行了对比,结果证实了该算法在提高推荐质量和缓解冷启动问题上效果显著。

The collaborative filtering algorithms in recommender systems usually suffer from data sparsity or cold-start problems.Although most of the existing social recommendation algorithms can alleviate these problems to a certain extent,they only measure the influence of trust relationship from a single aspect.In order to measure the influence of the social relationship on recommendation prediction more accurately,this paper proposed a novel social recommendation algorithm with field trust and distrust based on singular value decomposition (SVD),named FTDSVD.Based on the SVD algorithm,the trust relationship and distrust relationship information of users is added in order to correct the social relationship,and the global influence of users and the field relevance of trust are considered.Finally,it is compared with the state-of-the-art methods on the Epinions dataset .Experiment results show that the FTDSVD algorithm has obvious effects in improving the recommendation quality and alleviating the cold start problem.