SURGE: A New Benchmark for Social Media Sentiment Time Series with Interaction Structure

Researchers have unveiled SURGE, a benchmark for social media that integrates event-level time series with corresponding text and interaction structures that connect posts related to specific events. This dataset encompasses 67 events and over 800,000 posts spanning five categories, created through an automated system that generates calendar-aligned time series at three different temporal resolutions. Unlike existing datasets, which often limit their scope to a few events within a single category and overlook the interaction dynamics between posts, SURGE overcomes these challenges by organizing scattered posts into cohesive event-level time series. This allows for the analysis of how collective dynamics progress throughout an event's duration, making the benchmark useful for opinion forecasting and crisis management, capitalizing on the extensive discussions that public events generate on social media.