Outer Product-based Neural Collaborative Filtering(ONCF)一文发表在IJCAI 2018上,作者使用外积对嵌入空间中各个维度之间的相互关系进行显式建模。
The key to design a CF model is in 1) how to represent a user and an item, and 2) how to model their interaction based on the representation.
模型结构
目标函数
作者使用了BPR损失函数:
$$
L = \sum_{(u,i,j) \in \mathcal{D}} - \ln \sigma(\hat y_{ui} - \hat y_{uj}) + \lambda_{\Delta} \left \Vert \Delta \right \Vert^2
$$