Transforming Fake News: Robust Generalisable News Classification Using Transformers

Abstract

As online news has become increasingly popular and fake news increasingly prevalent, the ability to audit the veracity of online news content has become more important than ever. Such a task represents a binary classification challenge, for which transformers have achieved state-of-the-art results. Using the publicly available ISOT and Combined Corpus datasets, this study explores transformers' abilities to identify fake news, with particular attention given to investigating generalisation to unseen datasets with varying styles, topics and class distributions. Moreover, we explore the idea that opinion-based news articles cannot be classified as real or fake due to their subjective nature and often sensationalised language, and propose a novel two-step classification pipeline to remove such articles from both model training and the final deployed inference system. Experiments over the ISOT and Combined Corpus datasets show that transformers achieve an increase in F1 scores of up to 4.9% for out of distribution generalisation compared to baseline approaches, with a further increase of 10.1% following the implementation of our two-step classification pipeline. To the best of our knowledge, this study is the first to investigate generalisation of transformers in this context.

Exploring the Generalisability of Fake News Detection Models

“…This finding motivates the need for more robust models. It also adds to the body of evidence that suggests that current publicly available datasets are simply too small to train generalisable models [4]. This may explain why models trained on the ISOT dataset performed marginally better in terms of generalisability compared to models trained on the smaller datasets, and it further demonstrates the need for larger, well-labelled datasets, such as those hosted by Facebook, to be made more readily available.…”

“…This finding is also replicated by a study by [5] which also performed a similar test on similar, if not identical, datasets of different topics and found similar drops in accuracy between models trained on political news and tested on celebrity news and vice versa. Similarly, a study by [4] also explores how well models generalise across different topics by testing across two datasets: the ISOT [1], [2] dataset and the Combined Corpus (CC) [25] dataset. Most of the data contained in the ISOT dataset is political in nature whereas the Combined Corpus covers additional topics such as healthcare, sports and entertainment.…”

“…https://www.kaggle.com/c/fake-news4 https://www.kaggle.com/datasets/mdepak/fakenewsnet This article has been accepted for publication in a future issue of this conference proceedings, but has not been fully edited. Content may change prior to final publication.…”

Using of Transformers Models for Text Classification to Mobile Educational Applications

“…Las investigaciones relacionadas con modelos de clasificación multiclase con Transformers demuestran que los modelos BERT y DistilBERT superan alternativas de aprendizaje profundo en clasificación de noticias falsas [36]. Los estudios comparativos en un clasificador de reseñas falsas implementado con RoBERTa superó a los modelos ALBERT y DistilBERT [37].…”

Advanced Misinformation Detection: A Bi-LSTM Model Optimized by Genetic Algorithms

“…This comparison is based on a set of performance metrics: accuracy, precision, recall, and F1 score. Linear SVM [34] ---92.00% FakeBERT [35] 98.90% ---deBERTa [36] 97.70% 97.70% 97.70% 98.90%…”