Topic-Awared Contrastive Learning for Incoming Fake News Detection in News Streams

The prevalence of social media streams has transformed them into essential sources of real-time information on various topics. However, the simultaneous spread of fake news poses a significant challenge to the authenticity of information dissemination. Traditional fake news detection methods, which focus on analyzing individual articles in news streams, often neglect the topic-based correlations crucial for identifying misinformation patterns. This oversight hinders their ability to effectively detect fake news, especially when it relates to similar topics with a tendency for falsehoods. To mitigate this limitation, we propose the Topic-Awared Contrastive Learning Framework (TCLF), an efficient mechanism for analyzing topic-based correlations. The TCLF comprises two key components: the dynamic correlation mapping module, which employs incremental clustering to categorize news articles into familiar and unfamiliar topic domains, and the contrastive correlation learning module, which applies contrastive learning techniques to identify fake news across diverse topics. Our framework better detects fake news in news streams, outperforming existing state-of-the-art methods.