Open MochizukiShinichi opened 1 year ago
@MochizukiShinichi Hello, thanks for your attention to RecBole. In fact, the models of context-aware recommendation in RecBole are optimized and can be used to make CTR predictions, as you can see from their loss function (BCELoss). However, you can also customize the model, by optimizing it with your own loss function. Some module used in context-aware recommendation can also be reused. Please refer to customized model and customized metric in our docs for details.
Reading from the docs, it looks like that context-aware model type only works for 0/1 label and even for rating-like labels I'll still have to provide a threshold to transform it into 0/1 label. Is that the case? Can I customize my own loss/metric using Customized module?
Thank you very much for your great work!!