Misinformation, disinformation, and propaganda undermine trust in institutions, spread falsehoods, and can incite violence. Despite their significant impact, the research community largely lacks automated and programmatic approaches for tracking misinformation across different online platforms.
In his talk, Hans W. A. Hanley demonstrates how recent advancements in transformer-based models can help combat the global proliferation of misinformation. Specifically, he proposes and develops a system leveraging these models to help scalably identify, track, and analyze the spread of misinformation across thousands of international news websites. By employing novel multilingual Matryoshka embeddings and hierarchical level-wise clustering, his proposed system identifies individual news stories, topics, and overarching themes across news outlets. Utilizing multilingual stance detection, natural language inference, and network analysis, his system further assesses biases and factual inconsistencies in news articles, enabling the identification of websites and networks that disseminate propaganda or misinformation.
His approach illustrates how narrative-based tracking and deep natural language understanding can track multilingual and international news stories, support reporting and fact-checking, and ultimately help mitigate the global spread of misinformation, disinformation, and propaganda.
