Testing the Generalization of Neural Language Models for COVID-19 Misinformation Detection

TL;DR

This paper addresses whether Transformer-based language models trained for COVID-19 misinformation detection generalize across diverse content sources. It conducts a comprehensive cross-dataset evaluation of 15 models, spanning general-purpose and domain-adapted variants, across five datasets derived from social media, news, and scientific articles, with selective pre-training on the CORD-19 corpus. The key finding is that domain-specific pre-training and vocabularies do not consistently improve performance; in many cases, general-purpose models perform as well or better, underscoring the importance of cross-domain generalization over source-specific tuning. The work contributes a broad benchmark framework and highlights the need for diverse data and standardized evaluation to inform the development of robust misinformation-detection systems.

Abstract

A drastic rise in potentially life-threatening misinformation has been a by-product of the COVID-19 pandemic. Computational support to identify false information within the massive body of data on the topic is crucial to prevent harm. Researchers proposed many methods for flagging online misinformation related to COVID-19. However, these methods predominantly target specific content types (e.g., news) or platforms (e.g., Twitter). The methods' capabilities to generalize were largely unclear so far. We evaluate fifteen Transformer-based models on five COVID-19 misinformation datasets that include social media posts, news articles, and scientific papers to fill this gap. We show tokenizers and models tailored to COVID-19 data do not provide a significant advantage over general-purpose ones. Our study provides a realistic assessment of models for detecting COVID-19 misinformation. We expect that evaluating a broad spectrum of datasets and models will benefit future research in developing misinformation detection systems.