基于评论数据中隐式邻居关系的推荐系统研究

Research on Recommendation System Based on Implicit Neighbor Relations in Comment Data

传统推荐系统大多使用基于协同过滤的方法进行推荐,然而在现实场景中,大多数用户只对很少的项目进行了评分,因为缺少历史评分数据造成了冷启动问题,导致协同过滤方法的推荐质量不佳。本文使用丰富的评论数据挖掘用户之间和项目之间的隐式邻居关系,并联合项目信誉问题建立基于评论数据的社交矩阵分解模型ReTOMF。实验表明,与对应的其他推荐模型相比,ReTOMF展现了更好的推荐性能。

Traditional recommendation systems mostly use collaborative filtering-based methods for recommendation. However, in real-life scenarios, most users only score very few items, because the lack of historical score data causes a cold start problem. This paper uses rich comment data to mine the implicit neighbor relationship,and combines project reputation to establish a social matrix decomposition model ReTOMF. Experiments show that Re TOMF exhibits better recommendation performance than the corresponding other recommended models.