摘要
传统的协同过滤方法假设相似的用户有着相似的偏好,然而在不同的消费领域用户往往表现出不同的特征.此外,由于用户评价矩阵的稀疏性,使得相似用户的寻找极为困难.针对上述问题,该文提出了基于社团主题的领域相关推荐算法.首先,提出了一种包含社会网络,用户对商品的评价记录和项的分类三类信息的推荐框架.然后,分别提出了专家指导的主题模型和社会网络约束的主题模型.最后,对这两种模型进行综合,提出了统一推荐模型.实验表明,该文提出的方法具有较好的预测准确性,其性能明显优于其他相关算法.
Traditional collaborative filtering methods assume that similar users have similar taste in buying commodities, but users usually present different features in different consumer sectors. In addition, the user rating matrix is sparse, which makes it difficult for finding similar users. This paper proposes a domain specific recommendation algorithm based on community topic in social network. Firstly, we propose a rec- ommendation framework including social network, rating records of users for items, and item category. Secondly, we propose an expert guided topic model and a network constrained topic model. Finally, we combine the above models into a unified recommendation model. The experiments show that, the proposed method has higher prediction accuracy, and obviously performs better than related works.
出处
《湘潭大学自然科学学报》
CAS
北大核心
2015年第4期92-97,共6页
Natural Science Journal of Xiangtan University
基金
内蒙古自治区教育厅课题(NJSC14338)
关键词
推荐算法
社团
聚类
社会网络
recommendation algorithm
community
cluster
social network