On Unifying Misinformation Detection

Abstract

In this paper, we introduce UNIFIEDM2, a general-purpose misinformation model that jointly models multiple domains of misinformation with a single, unified setup. The model is trained to handle four tasks: detecting news bias, clickbait, fake news and verifying rumors. By grouping these tasks together, UNIFIEDM2 learns a richer representation of misinformation, which leads to stateof-the-art or comparable performance across all tasks. Furthermore, we demonstrate that UNIFIEDM2's learned representation is helpful for few-shot learning of unseen misinformation tasks/datasets and model's generalizability to unseen events. * Work partially done while interning at Facebook AI. † Work partially done while working at Facebook AI.

The Battlefront of Combating Misinformation and Coping with Media Bias

“…Stylistic approaches may be simple but yet effective for detecting low-quality human-written fake news, but not so good for machine-generated misinformation, which is stylistically consistent regardless of the underlying motives (Schuster et al, 2020). We then cover recent approaches (Lee et al, 2021b; that leverage a combination of these elements for greater representation power and robustness. Importantly, we also cover works that explore the diachronic bias of fake news detection and portability across data in different time and language settings (Murayama et al, 2021;Gereme et al, 2021).…”

Accurate and Explainable Misinformation Detection: Too Good to be True?

“…We review the related work on misinformation concerning our main contributions to (i) the characterization of content spread by this phenomenon and (ii) the models developed to detect such content. In this work, we adopt the definition of misinformation presented in [46], or rather, an umbrella term to include all false or inaccurate information that is spread online, such as rumor, clickbait or fake news, among others [5,23,46], intentionally or unintentionally propagated. Moreover, we only consider content-based classification models, which rely exclusively on textual data from various misleading online sources, such as web articles or social media posts, supporting content pre-bunking [12].…”

“…A work that extends the use of EmoCred with the transformers [42] is presented in [21]. The exploration of model generalizability across various misinformation data sources has been pursued by employing transformers in multi-task learning [29] and transfer learning [23] contexts.…”

“…We select datasets from [23], which consolidates various forms of misinformation, domains, and text structures, providing a comprehensive representation of the phenomenon. This list employs 9 datasets, divided into 2 main groups, that have undergone manual annotation in contrast to distantly supervised data used in other research [5,35].…”

“…We employ RoBERTa large for our misinformation content-based classifier, as in [23]. We use adapters fine-tuning with the Pfeiffer architecture [31] and adapter drop method [37] to efficiently manage multiple dataset training, avoiding full model fine-tuning.…”