propagate-selector

This repository contains the source code & data corpus used in the following paper,

Propagate-Selector: Detecting Supporting Sentences for Question Answering via Graph Neural Networks, LREC-20, [paper]

---

[requirements]

tensorflow==1.10 (tested on cuda-9.2, cudnn-7.6.5)
python==3.5
tqdm==4.31.1
nltk==3.3
h5py==2.8.0
ujson==1.35

[download dataset & preprocessing]

• download "HotpotQA" to ./data/raw/hotpot/
  
• clone ELMo repository and download pretrained models

sh init_make_dataset.sh

• processed file (train/dev.pkl)
• [#samples, 4]
  • 0: question [#token]
  • 1: list (sentences) [#sentence, #token]
  • 2: index of first sentence of each passage [#sentence]
  • 3: label [#sentence]

---

[training Phase]

• run reference script in "./model" folder
• results will be displayed in console
  
• results will be saved to "./model/TEST_run_result.txt"
  

sh reference_script_train.sh

[hyper parameters]

• major parameters : edit from "./model/reference_script_train.sh"
  
• other parameters : edit from "./model/params.py"
  

---

[cite]

• Please cite our paper, when you use our code | dataset | model

  @inproceedings{yoon2020propagate,
  title={Propagate-Selector: Detecting Supporting Sentences for Question Answering via Graph Neural Networks},
  author={Yoon, Seunghyun and Dernoncourt, Franck and Kim, Doo Soon and Bui, Trung and Jung, Kyomin},
  booktitle={Proceedings of The 12th Language Resources and Evaluation Conference},
  pages={5400--5407},
  year={2020}
  }