understanding-reviews

this is the official implementation of "Understanding the Effectiveness of Reviews in E-commerce Top-N Recommendation" in proceedings of ICTIR 2021. If you use our code, please cite our paper:

@inproceedings{xu2021understanding,
  title={Understanding the Effectiveness of Reviews in E-commerce Top-N Recommendation},
  author={Xu, Zhichao and Zeng, Hansi and Ai, Qingyao},
  booktitle={Proceedings of the 2021 ACM SIGIR International Conference on Theory of Information Retrieval},
  pages={149--155},
  year={2021}
}

if you need code for rating prediction, please refer to AHN official implementation: https://github.com/Moonet/AHN, ZARM official implementation https://github.com/HansiZeng/ZARM

Requirements:
python 3.6+ PyTorch 1.4.0
Scikit-learn 0.23.2

Dataset:

download subcategory files from http://jmcauley.ucsd.edu/data/amazon/links.html

Pretrained Word2vec:

download from https://drive.google.com/file/d/0B7XkCwpI5KDYNlNUTTlSS21pQmM/edit?resourcekey=0-wjGZdNAUop6WykTtMip30g

Run matrix factorization:

python preprocess.py
python train.py
python rerank.py to create ranklist_with_gt.json
python train.py to rerank
python evaluate.py to calculate hit rate and ndcg

Run deepconn:

python divide_and_create_example_doc.py
python train.py
python evaluate.py