IRGPR_CIKM_2020

This is a pytorch re-implementation of the paper Personalized Re-ranking with Item Relationships for E-commerce, built upon PyG library.

Citation

@inproceedings{liu2020personalized,
  title={Personalized Re-ranking with Item Relationships for E-commerce},
  author={Liu, Weiwen and Liu, Qing and Tang, Ruiming and Chen, Junyang and He, Xiuqiang and Heng, Pheng Ann},
  booktitle={Proceedings of the 29th ACM International Conference on Information \& Knowledge Management},
  pages={925--934},
  year={2020}
}

Dependecies

• Python3.7
• PyTorch
• PyG
• networkx
• pandas
• gensim

Experiment Data

• Amazon Review Data
• To process a dataset from raw files
  • Please get the following files from Amazon Review Data and put them at the raw/ directory.

  meta_Video_Games.json.gz
  ratings_Video_Games.csv
  reviews_Video_Games.json.gz
  

  • Obtain node features from reviews by gensim.models.doc2vec, and put the .d2v file at raw/.
  • As well as the initial ranked lists, examples as data/Amazon/raw/train_ratings_Video_Games.txt and data/Amazon/raw/test_ratings_Video_Games.txt in the format of [uid] [iid] [label] [initial score].
• We provided a sample processed heterogenous graph Amazon_Video_Games.pt from Amazon Video Games raw data, so that you can directly load the processed data and train the model.

Experiment

• Before running, please modify the corresponding Amazon data category in amazon_rerank_loader.py.

python run_irgpr.py --lr [lr] --node_emb [node embedding dim]