Pytorch Version QA4IE Code (Journal Version)

Here is the original implementation for the following series of publications on QA4IE.

• ISWC 2018: QA4IE: A Question Answering based Framework for Information Extraction
• IEEE Access 2020: QA4IE: A Question Answering based System for Document-Level General Information Extraction
• SIGIR DEMO 2020: QuAChIE: Question Answering based Chinese Information Extraction System

This branch mantains the code for the journal version QA4IE. The ISWC conference version is maintained in iswc branch. The chinese benchmark and the corresponding code will also be released soon.

Requirements

• torch = 1.7.1
• wandb

Pretrained Models

Download preprocessed data and pretrained models here.

Uncompress them in ./data and ./out, respectively.

Training & Evaluation Scripts

Training and evaluation scripts are provided in ./scripts. Note that you need to execute the dump_SS.sh script before training QA module, and execute the dump_QA.sh before training the AT module.

IE-setting Evaluation

Evaluate in IE-setting with different types of scorer:

• mean: average probability of answer sequence
• prod: product of answer sequence probabilities
• AT: use the output of the AT module as the score

python3 eval_ie.py --scorer <mean|prod|AT>

Issues on Experimental Results

• Note that the results obtained by running the code in this repo will be slightly better than the results reported in the paper. The main reasons are the usage of a more proper optimizer, a larger batch size, and a learning rate scheduler in the new implementation.

Cite Us

@inproceedings{qiu2018qa4ie,
  title={QA4IE: A question answering based framework for information extraction},
  author={Qiu, Lin and Zhou, Hao and Qu, Yanru and Zhang, Weinan and Li, Suoheng and Rong, Shu and Ru, Dongyu and Qian, Lihua and Tu, Kewei and Yu, Yong},
  booktitle={International Semantic Web Conference},
  pages={198--216},
  year={2018},
  organization={Springer}
}

@article{qiu2020qa4ie,
  title={Qa4ie: A question answering based system for document-level general information extraction},
  author={Qiu, Lin and Ru, Dongyu and Long, Quanyu and Zhang, Weinan and Yu, Yong},
  journal={IEEE Access},
  volume={8},
  pages={29677--29689},
  year={2020},
  publisher={IEEE}
}

@inproceedings{ru2020quachie,
  title={QuAChIE: Question Answering based Chinese Information Extraction System},
  author={Ru, Dongyu and Wang, Zhenghui and Qiu, Lin and Zhou, Hao and Li, Lei and Zhang, Weinan and Yu, Yong},
  booktitle={Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval},
  pages={2177--2180},
  year={2020}
}