Code of paper “EAGER: Two-Stream Generative Recommender with Behavior-Semantic Collaboration” (pdf) accepted by KDD2024.
- raw_data_file
The raw 5-core data can be downloaded from official website.
- preprocess
Set data_set_nome (e.g. 'Amazon_Beauty') and have_processed_data to False. Run:
python train_rec.py
After it, set have_processed_data to True to avoid repeated preprocessing.
- DIN
To constuct the behavior code, you need to train a DIN model. Run:
python train_din.py
- EAGER
With the pre-trained DIN model, set DIN_Model_path to its checkpoint path. Run:
python train_rec.py
We greatly appreciate the official RecForest repository. Our code is built on their framework.