社交关系在基于模型社会化推荐系统中的影响

Impact of Social Relationship on Model-Based Social Recommender Systems

目前社会化推荐系统方面的研究主要集中于构建性能更优的基于模型的推荐算法,然而模型算法中分解得到的隐式特征和社交信息的变化会给推荐性能带来不确定性。为了消除不确定性,探究了在基于模型的社会化推荐系统中社交关系的变化对推荐性能的影响。实验首先按比例移除关系网络中的连边或节点,再对推荐质量进行评估,结果表明,社交关系的数量增多将对推荐质量带来明显提升,同时关系网络中心节点对推荐质量的影响巨大。因此,在构建基于模型的社会化推荐系统的过程中应尽可能多地获取社交关系,并提升中心节点的关系在推荐中的权重,降低非中心节点(潜在噪声)的影响。

At present, the research of social recommender systems mainly focuses on the construction of models based recommendation algorithms. However, the latent factors and the social information involved in the algorithms may bring uncertainty to the recommendation quality. To explore this uncertainty, this paper explores the impact of the social relation in the model-based social recommender systems. The experiments remove edges or nodes in the relation network proportionally, and then evaluate the recommendations. The results show that, with the increase of the relation density, the performance of the recommendation algorithms is boosted, and the central nodes make huge difference. Hence the social relation information should be collected much more during building the model-based social recommender systems. Moreover, processing data according to the importance of the nodes in the recommendation can help to enhance the recommendation performance and reduce the impact of noncentral nodes.