Misinformation in Motion: A Continuously Updated Dataset and Benchmark for Real-Time Misinformation Detection

Misinformation in Motion is a continuously updated dataset designed to support research on real-time misinformation detection. Spanning content from 2024\u20132025, the dataset contains 2,402 manually and semi-automatically labeled samples across nine domains, including Politics, Health, Science, and International Affairs. Data sources include social media platforms, online news aggregators, and verified fact-checking organizations. Each entry includes metadata such as timestamp, source, domain label, and a binary classification (real or fake). This dataset is intended for evaluating the performance of misinformation detection models, especially retrieval-enabled large language models. It is updated quarterly to reflect emerging misinformation narratives and trends.