Semantic web-based propaganda text detection from social media using meta-learning

In recent years, due to the rapid development of social media, there have been many propaganda texts and propaganda activities on the internet. While previous studies have primarily concentrated on linguistic characteristics for detecting propaganda, there has been a lack of systematic investigation into the role of semantic features in the dissemination of propaganda. It may cause harm to society, so it is necessary to study automated methods to identify the semantic features of propaganda. To address these issues, we proposed a meta-learning-based sentence-level text classification method to detect semantic web-based propaganda in news stories automatically. Our method uses a multi-task learning technique, which is utilized to capture the semantic contradictions in news content, combines conditional random fields (CRF), bidirectional long short-term memory (BiLSTM) networks, and a language model that has already been trained. Its efficacy is demonstrated by the evaluation of a news article dataset, where it obtained an F1 score of 0.61 on multi-lingual data and 0.688 on mono-lingual data. Moreover, our model outperforms current approaches in mono-lingual and multi-lingual scenarios, demonstrating the effectiveness of our multi-task learning methodology in detecting disinformation tactics in news stories.