Fig. 2From: Recommend what to cache: a simple self-supervised graph-based recommendation framework for edge caching networksThe overall framework of SimSGR. Following the scheme of CL, one view of the contrastive positive pair \(Z\) is encoded from the original graph \(G\), and the other \(Z^{\prime}\) is generated from the Mixing layer by mixing both the neighboring nodes and similar nodes. The two contrastive views are then converted into the rating matrices \(R\) and \(R^{\prime}\) in the Conversion layer, SimSGR aims to maintain the invariance of \(R\) and \(R^{\prime}\), and the covariance criterion is used to prevent the model from collapsingBack to article page