Introduction

This is the repository of our accepted CIKM 2021 paper "Continuous-Time Sequential Recommendation with Temporal Graph Collaborative Transformer" and the proposed model is TGSRec. Paper is available on arxiv. This work focuses on multi-steps continuous-time recommendation, where user and item embeddings are generated in any unseen future timestamps. Different from existing sequential recommendation methods, which are optimized for next-item prediction, this work is learned for recommendation in any timestamps.

Citation

Please cite our paper if using this code.

@inproceedings{fan2021continuous,
  title={Continuous-Time Sequential Recommendation with Temporal Graph Collaborative Transformer},
  author={Fan, Ziwei and Liu, Zhiwei and Zhang, Jiawei and Xiong, Yun and Zheng, Lei and Yu, Philip S.},
  booktitle={Proceedings of the 30th ACM International Conference on Information and Knowledge Management},
  year={2021},
  organization={ACM}
}

Implementation

The code is implemented based on TGAT.

Environment Setup

The code is tested under a Linux desktop (w/ GTX 1080 Ti GPU) with TensorFlow 1.12 and Python 3.6. Create the requirement with the requirements.txt

ML-100K Dataset Execution

Sample code to run

mkdir log
mkdir saved_checkpoints/ml-100k
mkdir saved_models/ml-100k
mkdir processed

python run_TGREC.py -d ml-100k --uniform --bs 600 --lr 0.001 --n_degree 30 --agg_method attn --attn_mode prod --gpu 2 --n_head 2 --n_layer 2 --prefix Video_Games_bce --node_dim 32 --time_dim 32 --drop_out 0.3 --reg 0.3 --negsampleeval 1000

ytchx1999 / TGSRec