Detecting Propaganda Techniques in English News Articles using Pre-trained Transformers

The rapid evolution of technology and the Internet has enabled massive data to reach many audiences. People search for news, provide opinions, and discuss decisions online. Online news has become available in various news outlets, including websites and social networks. Consequently, several types of fake news have been raised, for instance, propaganda and rumors. Fake news leads to significant societal risks and has emerged as a major societal problem. Propaganda is spread over the Internet and appears in news articles that aim to manipulate people’s public opinion and influence their attitudes by using psychological and rhetorical methods that appeal to people’s feelings. This paper resolves one of the state-of-the-art NLP research; propaganda techniques classification. We apply the state-of-the-art pre-trained language model, RoBERTa, to detect propaganda techniques from online news articles. The model has been evaluated using a reference dataset for the SemEval-2020 Task 11. The results show that the word embedding pre-trained model (RoBERTa) can detect complex propaganda techniques and outperform the baseline by achieving an F1 score of 60.2%.