A method, device, and computer-readable storage medium storing instructions are provided for detecting controversial events that are reflected in user-generated content items. In a single-step approach, user-generated content items are received and analyzed by a controversial event detection module, which determines the likelihood that sets of content items reflect controversial events. In one example, public posts by users of a social networking service are grouped into snapshots of posts that are associated with an entity and were generated during a window of time. An event detection module may determine the likelihood that snapshots reflect events. In a two-step approach, event snapshots are provided to a controversy detection module, which determines the likelihood that event snapshots are controversial. In a blended approach, snapshots are provided to a controversy detection module, which determines the likelihood that snapshots are controversial events based in part on the event score.