This repository implements models from the following two papers:

BERT4Rec: Sequential Recommendation with BERT (Sun et al.)

KeBERT4Rec: Integrating keywords into BERT4Rec for Sequential Recommendation

and lets you train them on MovieLens-1m and MovieLens-20m.

The implementation of BERT4Rec is based on:

https://github.com/jaywonchung/BERT4Rec-VAE-Pytorch (66f0853) from Jaewon Chung

Run main.py with arguments to train and/or test you model. There is a template file 'templates.py' available.

python main.py --template train_bert

1. Train BERT4Rec on ML-20m and run test set inference after training

   python main.py --dataset_code=ml-20m --template=train_bert --experiment_description=ml-20m_baseline --train_batch_size=64"

2. Train BERT4Rec on ML-20m with merge embedding

   python main.py --dataset_code=ml-20m --template=train_content_bert 
   --experiment_description=ml-20m_merge_embedding 
   --train_batch_size=64      ```
   

3. Train BERT4Rec on ML-20m with multi-hot embedding

   python main.py --dataset_code=ml-20m --template=train_content_bert 
   --content_encoding_type=simple_embedding --experiment_description=ml-20m_simple_embedding 
   --train_batch_size=64      ```