Misinformation is often edited and repeated across multiple platforms, websites, languages, and formats (audio, video, image, text). In this talk, Dr Scott A. Hale will detail initiatives to crowdsource misleading content through "tiplines" on messaging platforms like WhatsApp and Telegram as well as state-of-the-art natural language processing approaches to group messages making the same claims.
While much effort focuses on large, high-resource languages and unencrypted platforms. Tiplines offer the opportunity for users of end-to-end encrypted platforms to share potentially misleading content with fact-checking organizations and check whether that content has been fact-checked. This talk will show how knowledge distillation can be used to create machine learning models for claim matching that perform well on low-resource as well as high-resource languages.
The research detailed in this talk has been integrated into our product Check and is used by over a dozen fact-checking organizations across the globe. We are currently working to create better infrastructure to allow academics, practitioners, and community leaders to collaborate more easily on misinformation response.