Requirements

• python3.6
• tensorflow-gpu 1.11
  • GPU: pip install tensorflow-gpu==1.11.0
• tensorflow_hub 0.1.1
• nltk,numpy, pandas, json, sklearn

(Note: our code is developed based on original Bert's implementation original)

Annotated_Data

Data

1. The original dataset are from dataset link, including Amazon QA and reviews data.
2. In our work, we have chosen 7 categories, namely Tools_and_Home_Improvement, Patio_Lawn_and_Garden, Automotive, Cell_Phones_and_Accessories, Health_and_Personal_Care, Sports_and_Outdoors, Home_and_Kitchen.
3. In order to replicate our results, you need to download QA (such as QA_Tools_and_Home_Improvement.json.gz etc) and review dataset (such as reviews_Tools_and_Home_Improvement.json.gz etc) in all of these categories.
4. Pleae keep downloaded files in folder "data".

Dataset preprocessing

python data_preprocess.py

Training and Prediction

1. Cross-domain training FLTR: python FLTR_1st.py
2. Fine-tuning FLTR for each category, for example: python FLTR_2rd.py Home_and_Kitchen
3. Cross-domian training BERTQA: python BertQA.py ALL
4. Fine-tunning BERTQA for each category, for example: python BertQA.py Home_and_Kitchen

Citation

Please cite our paper if you use our model or data in your work:

@inproceedings{zhang2019discovering,
  title={Discovering relevant reviews for answering product-related queries},
  author={Zhang, Shiwei and Lau, Jey Han and Zhang, Xiuzhen and Chan, Jeffrey and Paris, Cecile},
  booktitle={2019 IEEE International Conference on Data Mining (ICDM)},
  pages={1468--1473},
  year={2019},
  organization={IEEE}
}