用于组推荐的超图卷积网络笔记

1 Title

        Hypergraph Convolutional Network for Group Recommendation（Renqi Jia; Xiaofei Zhou; Linhua Dong; Shirui Pan）【2021 ICDM】

2 Conclusion        

        This paper proposes a novel dual channel Hypergraph Convolutional network for group Recommendation (HCR), which consists of member-level preference network and group-level preference network. In the member-level preference network, in order to capture cross-group collaborative connections among users and items, we devise a member-level hypergraph convolutional network to learn group members’ personal preferences. In the group-level preference network, the group’s general preference is captured by a group-level graph convolutional network based on group similarity. 

3 Good Sentences

        1、These predifined strategies are data-independent, short of the ability to model the preferences of group members and adjust their weights dynamically. This ability is significant for a group to make decisions on different items.(The reason why Group Recommendaition need new methods to improve)
        2、it is of significance to capture cross-group collaborative relation for better group preference modeling.  As a group may pursue some targets distinct from each user’s interests, the above aggregation methods are not sufficient to characterize the group’s general preference.(The reason why modeling the group’s general preference is a valuable research issue.) 
        3、In contrast, our setting is conservative and does not include extra side information: we know only user and item ids, and item implicit feedbacks. We capture the cross-group collaborative information through a novel dual channel hypergraph convolutional network.(The Input of this model that decrease the necessary of large dataset)
        4、Because of the data sparsity of group interactions, the learned group representation is not sufficiently accurate. To further accelerate and enhance the group preference learning, we propose to incorporate the user-item interaction data to optimize the group-item and user-item recommendation tasks simultaneously.（The way to slove the problems of data sparsoty）

---

本文提出了一种新的双通道超图卷积网络用于群体推荐，该网络提取协作信息和群体相似性，分别拟合群组成员个人偏好和群体一般偏好。

如图所示，除了超图外本文还定义了一个重叠图，重叠图是对应于超图而存在的，每个节点都是超图的一条超边，如果超图的超边中有共享的节点，那么重叠图的节点也相连。

超图构建部分：

        将每个小组表示为一个超边。每个超边都包含小组成员以及他们的交互项目。每个超边上具有一个权重，权重的大小等于共享节点/全部节点。

个人小组成员偏好包括：Member-level Hypergraph Convolutional Network和Member Preference Aggregation Network两个部分，在这部分内容构建的超边中，两个超边共享的集合用户和项目表示组之间的协作兴趣。一个就是用来捕捉普通成员兴趣，另一个则是“会员级成员具有更多权重”，直观上就是如果用户对某个项目有更多的专业知识，他应该对小组对该项目的选择有更大的影响。
        上面的方法简单地将群体成员的个人兴趣汇总为群体的偏好，会忽略内在的群体层面的偏好，这些偏好可能与群体内所有个人的偏好不同且独立，针对这个问题，本文提出了Group-level Preference Network，在超图的重叠图上从图卷积网络中捕获组级偏好。然后聚合成员级别的首选项和组级别的首选项，以获得组的最终表示。