Open leon-cas opened 2 years ago
Sorry for the late reply. The time cost of training on a graph comes from the aggregation process of GNN. A way to accelerate the sampling process of MixGCF is re-using the inner product between user-item and updating the scores during the training. In this way, many computations can be saved but it may introduce bias.
Hi! thank you for the great work. I was wondering if there is a kind of solution to make this work adaptive to large user-item graph?