这篇文章的主要贡献点在于通过user-item interactions建立interactive graph,通过social network建立social links,以及textual reviews建立semantic links.
interactive graph的建立没有讲,一笔带过;
social links S is a user-by-user square matrix initially
filled by any existing social links between users;
textual link embedding 用SBERT训练得到。
最后的heterogeneous graph H表示为:
![](https://img.haomeiwen.com/i9211006/a8d695185e561672.png)
embedding的生成——the graph Laplacian method:
![](https://img.haomeiwen.com/i9211006/e74de6dcb1df766b.png)
![](https://img.haomeiwen.com/i9211006/cc72b4c4d6d0fa8e.png)
做实验的时候将cold-start user(观看记录小于10)和regular user的结果分开写,看出该方法对cold-start user有明显提升。
网友评论