The present disclosure relates to systems, methods, and computer-readable media that generate a platform-specific representative image (e.g., a thumbnail image) for a digital video that is predicted to effectively engage users of a communication system (e.g., a social media system). For example, systems described herein include identifying keywords and associated engagement scores for the content sharing platform. The systems described herein further include identifying keywords associated with segments of the digital video. The systems can further determine a segment of interest based on the engagement scores for the content sharing platform and a semantic relationship between keywords for the content sharing platform and one or more keywords associated respective segments of the digital video. The system can further determine a representative image that effectively engages users of the sharing platform by determining a representative image from the identified segment of interest.