MindTS-MMD: A Chinese Multimodal Dataset for Emotion Expression Analysis in Children with Tourette Syndrome

The MindTS-MMD dataset was developed to support multimodal analysis of emotional expression in children with Tourette Syndrome (TS), a clinically underrepresented population in affective computing research. Based on an emotion-induction paradigm, data were collected from 60 children aged 6–12 years through semi-structured interactive tasks designed to elicit observable emotional responses. To enable privacy-aware data sharing and standardized multimodal analysis, MindTS-MMD is released in the form of instance-level symbolic multimodal representations. Each representation provides a de-identified, abstracted summary of task-relevant information derived from synchronized video, audio, and text modalities, without exposing identifiable raw signals. In total, the dataset comprises 10,034 symbolic instances corresponding to emotional task segments, with associated availability of visual, acoustic, and semantic content. Each instance is annotated with one of seven elicited emotion categories—anxious, calm, focused, irritable, relaxed, shy, and tense—together with dual valence–arousal ratings obtained from child self-reports and experimenter observations, as well as clinically relevant tic-behavior labels. All annotations were independently performed by multiple trained raters and reviewed by domain experts to ensure reliability and consistency. Baseline unimodal and multimodal classification experiments are provided for technical validation. By offering a large-scale, well-documented collection of symbolic multimodal representations from a pediatric TS population, MindTS-MMD serves as a comprehensive benchmark for multimodal emotion analysis and supports future research on emotion–tic interactions in both clinical and computational settings.