This repo contains the PyTorch implementation of the model Sentiment Aligned Explainable Recommendation (SAER) proposed in our WSDM 2021 paper "Explanation as a Defense of Recommendation". Please refer to the paper for the details of the algorithm.

All the depencies are included in the file environment.yml. If you use conda, installation can be done with the following command:

conda env create --name saer_env --file=environments.yml

Please refer to the data preprocessing README and follow the steps to prepare the data.

Check the configurations under the folder /config. Create a new or update existing model configuration.

Train the model specified by the given configuration file and optionally resume from an existing checkpoint

python train.py -m=saer --checkpoint=10

Specify a text decoding strategy to generate recommendation explanations for the testing data with a trained model of a checkpoint, and save the output to a file for later evaluation

python decode.py -m=saer --checkpoint=20 --search=greedy --output=sear_decode.txt

Evaluate the model on some metrics on-the-fly

python eval.py -m=saer --checkpoint=20 rmse mae ...

Directly evaluate the previously decoded output

python eval_decode.py -f=saer_decode.txt rmse bleu ...

Please cite our paper if you use this code in your research:

@inproceedings{yang2021explanation,
  title={Explanation as a Defense of Recommendation},
  author={Yang, Aobo and Wang, Nan and Deng, Hongbo and Wang, Hongning},
  booktitle={In Proceedings of the Fourteenth ACM International Conference on Web Search and Data Mining (WSDM ’21)},
  year={2021}
}