Open Zachary-ZS opened 3 years ago
Actually, I haven't tried BD on GNN recommender systems (LIghtGCN or NGCF). However, Since they use the BPR loss function for the training, you could use our implementation (BD on BPR) as well with a minor model architecture change.
Got it. Thx
I've watched your presentation on Youtube and noticed that no graph-based recommender systems are tested. So have you ever tried the BD framework on GNN recommender systems? If so, could you tell me the results? Thank U!