Using the Poly-encoder for a COVID-19 Question Answering System

Seolhwa Lee, João Sedoc

Keywords: Poly-Encoder, COVID19, Question Answering System

TL;DR: Poly-encoder based question and answering system on COVID19.

Abstract: In this unprecedented era of COVID-19, thereis a huge information need for truthful answersto questions relating to all aspects of the Coro-navirus.In this paper, we adapt the Poly-Encoder (Humeau et al., 2020) informationalretrieval from FAQs. We show that after fine-tuning the Poly-encoder we achieve higher re-call at the same level of precision. We compareseveral models on the task of answer or ques-tion selection on evaluation sets and demon-strate superior performance.We make ourcode publicly available for other researchers touse.