Behavioural Ambivalence/Hesitancy

The BAH dataset addresses ambivalence and hesitancy recognition in videos for digital behaviour change research, with expert annotations, transcripts, and participant metadata.

BAH was created to support machine-learning models that recognize subtle ambivalence and hesitancy cues across facial, vocal, and body-language signals.

The source page describes benchmark work for frame-level and video-level recognition, mono-modal and multi-modal setups, zero-shot prediction, and personalization through unsupervised domain adaptation.