A Comprehensive Survey on Datasets for Affective Computing and Mental Disorder

The automated recognition of emotions, moods, and mental disorders is a critical area of research with profound implications for advancing human-computer interaction and digital healthcare. However, the development of robust models is fundamentally dependent on high-quality, well-documented datasets. Researchers often face a complex and fragmented domain, making the selection of appropriate data resources problematic. To address this challenge, this article presents a comprehensive survey of key datasets for the recognition of emotions and mental disorders. We introduce a novel structured taxonomy, which classifies these resources based on crucial dimensions: label structure, signal and modality, collection methodology, domain and purpose, and cultural and idiomatic sensitivity. This framework provides a systematic guide for researchers and practitioners to make more informed decisions in dataset selection. By clarifying the strengths, limitations, and provenance of existing datasets, this work facilitates the development of more effective computational systems, fostering impactful innovation in the field.