An emotion-recognition corpus for HIV digital-health communication: 7,177 messages combining the GoEmotions corpus (projected to seven target emotions) with 1,395 messages authored for this study (995 HIV-contextualized, 400 general-domain). Built for the HIVRoBERTa + LLM-refinement pipeline described in the accompanying paper.

	Composition and splits

Split Rows Content

train 5,041 GoEmotions train (4,621, multi-label) + authored… See the full description on the dataset page: https://huggingface.co/datasets/SebastianVla/HIVEmotions.