EdinburghNLP at WNUT-2020 Task 2: Leveraging Transformers with Generalized Augmentation for Identifying Informativeness in COVID-19 Tweets

TL;DR

This paper presents an ensemble of transformer models trained with semi-supervised learning to identify informative COVID-19 tweets, significantly improving detection accuracy in social media emergency communication.

Contribution

It introduces a novel ensemble approach combining RoBERTa, XLNet, and BERTweet with semi-supervised training for tweet informativeness classification.

Findings

Achieved an F1 score of 0.9011 on the test set

Outperformed baseline FastText embedding system

Ranked 7th on the WNUT-2020 leaderboard

Abstract

Twitter and, in general, social media has become an indispensable communication channel in times of emergency. The ubiquitousness of smartphone gadgets enables people to declare an emergency observed in real-time. As a result, more agencies are interested in programmatically monitoring Twitter (disaster relief organizations and news agencies). Therefore, recognizing the informativeness of a Tweet can help filter noise from the large volumes of Tweets. In this paper, we present our submission for WNUT-2020 Task 2: Identification of informative COVID-19 English Tweets. Our most successful model is an ensemble of transformers, including RoBERTa, XLNet, and BERTweet trained in a Semi-Supervised Learning (SSL) setting. The proposed system achieves an F1 score of 0.9011 on the test set (ranking 7th on the leaderboard) and shows significant gains in performance compared to a baseline system using FastText embeddings.