MER2026-Prefer

Organizer: coffey; 437 submissions; MER-Prefer is a newly introduced track that predicts which of two emotion descriptions is preferred by human annotators for a given video. The concept of emotion preference was first proposed in EmoPrefer, playing a crucial role in training reward models capable of understanding human emotions. We employ the weighted F1-score (WAF) as the primary metric and accuracy (ACC) as the secondary metric. This is part of the MER2026 Challenge at ACM Multimedia 2026.