Early Detection of Fake News with Multi-source Weak Social Supervision

Social media has greatly enabled people to participate in online activities at an unprecedented rate. However, this unrestricted access also exacerbates the spread of misinformation and fake news which cause confusion and chaos if not detected in a timely manner. Given the rapidly evolving nature of news events and the limited amount of annotated data, state-of-the-art systems on fake news detection face challenges for early detection. In this work, we exploit multiple weak signals from different sources from user engagements with contents (referred to as weak social supervision), and their complementary utilities to detect fake news. We jointly leverage limited amount of clean data along with weak signals from social engagements to train a fake news detector in a meta-learning framework which estimates the quality of different weak instances. Experiments on real-world datasets demonstrate that the proposed framework outperforms state-of-the-art baselines for early detection of fake news without using any user engagements at prediction time.