SEER

This is the code for the paper:

Synthesizing Aspect-Driven Recommendation Explanations from Reviews
Trung-Hoang Le and Hady W. Lauw
Presented at IJCAI-PRICAI-20

We provide:

• Code to run SEER framework
• Data we used for our experiments

If you find the code and data useful in your research, please cite:

@inproceedings{ijcai2020-336,
  title     = {Synthesizing Aspect-Driven Recommendation Explanations from Reviews},
  author    = {Le, Trung-Hoang and Lauw, Hady W.},
  booktitle = {Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence, {IJCAI-20}},
  publisher = {International Joint Conferences on Artificial Intelligence Organization},
  editor    = {Christian Bessiere},
  pages     = {2427--2434},
  year      = {2020},
  month     = {7},
  note      = {Main track}
  doi       = {10.24963/ijcai.2020/336},
  url       = {https://doi.org/10.24963/ijcai.2020/336},
}

How to run

pip install -r requirements.txt

The following execution scripts are constructed for toy category as in the paper.

Prepare data

python prepare_data.py --input data/toy/profile.csv --out data/toy --ratio_validation 0.2 --ratio_test 0.2

Train aspect-level sentiments model (EFM or MTER)

python train_efm.py --indir data/toy --epoch 1000 --out data/toy/efm

python train_mter.py --indir data/toy --epoch 100000 --out data/toy/mter

Train Aspect-Sentiment Context2Vec (ASC2V) model for opinion completion task

Execute the following command to prepare training data for this task

python prepare_opinion_contextualization_data.py --indir data/toy --efm_dir data/toy/efm --mter_dir data/toy/mter --out data/toy

Train ASC2V by the following command:

python train_asc2v.py --indir data/toy --gpu 0 --out data/toy/asc2v --context asc2v

For ASC2V with sentiment score from MTER model, executing the following command:

python train_asc2v.py --indir data/toy --gpu 0 --out data/toy/asc2v-mter --context asc2v-mter

Synthesizing explanation

python seer.py --input A10L9NQO44OLOU,B0044T2KBU,toy,toy --corpus_path data/toy/train.csv --strategy greedy-efm --preference_dir data/toy/efm --contextualizer_path data/toy/asc2v/model.params

or if you want to use the sentiments produced by MTER

python seer.py --input A10L9NQO44OLOU,B0044T2KBU,toy,toy --corpus_path data/toy/train.csv --strategy greedy-mter --preference_dir data/toy/mter --contextualizer_path data/toy/asc2v-mter/model.params

The above command generates an explanation for user A10L9NQO44OLOU, item B0044T2KBU, and the demanded aspects are toy,toy (2 sentences about aspect toy). This example is taken from the test dataset.

By default, we run SEER with greedy algorithm. SEER-ILP is provided and IBM ILOG CPLEX Optimization Studio must be ready to be able to proceed further. After setting up CPLEX, you can try SEER-ILP with argument --strategy ilp-efm (or --strategy ilp-mter).

To run the framework with other data, please modify the arguments accordingly.

Contact

Questions and discussion are welcome: lthoang.com