DILabIntern

2020.12~2021.02 Internship @ Yonsei Univ. DILab

---

Research Assistant

Dataset

HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering

Download

Glove: Stanford's GloVe 100d word embeddings

Download

Model

AIR-Retriever: Alignment-based Iterative Evidence Retrieval for Multi-hop Question Answering

Download

How to Run?

1. Download Dataset and Model and put it into the same directory
2. Prepare requirement for AIR-retriever. (Check readme.md)
3. Change directory of Air-Retreiver files. Check readme.txt Input data name is 'hotpot_modded_train.json' and 'hotpot_modded_dev' as we first convert HotPotQA to MultiRC to fit into AIR-Retriever
4. Run Dataset-Builder.ipynb
5. Use hotpot_dev_distractor_made.json to other QA models!

---

Papers

Check paper

PDF file with PPTX means it is reviewed.