Codes for our paper published in TOIS'22: RESUS: Warm-Up Cold Users via Meta-Learning Residual User Preferences in CTR Prediction
This repo includes an example for training RESUS upon a DeepFM model on the MovieLens-1M dataset. The dataset is preprocessed and splitted already.
Requirements: Python 3 and PyTorch.
Please run pretrain.py to train a DeepFM first. Then run main_nn.py or main_rr.py to train our RESUS. The default value of overfit_patience is 2 with less time cost, but a higher one may produce better results.