Algorithms of emotion: A hybrid NLP analysis of neurodivergent Reddit communities.

Résumé

Online communities such as Reddit offer neurodivergent individuals a unique space to express emotions, seek psychosocial support, and negotiate identity outside conventional social constraints. Understanding how these communities articulate and structure emotional discourse is essential for inclusive technology design. This study employed a hybrid natural language processing (NLP) framework that integrates lexicon-based sentiment analysis (VADER) with transformer-based topic modeling (BERTopic). A dataset comprising 238,100 posts and comments from subreddits focused on autism, ADHD, neurodiversity, and mental health (January 2023-February 2024) was analyzed to detect thematic and affective patterns in digital neurodivergent discourse. The analysis revealed ten semantically coherent and emotionally salient topics, including emotional overload, masking, diagnostic barriers, coping strategies, and AI-mediated interaction. Sentiment-topic alignment highlighted how affective tone is embedded within distinct thematic clusters, underscoring the structured emotional dynamics of neurodivergent communities. This study advances methodological rigor in affective computing and human computer interaction by combining scalable NLP tools with visual analytics to map digital neurodivergent discourse. It contributes a reproducible framework for analyzing emotion-rich online content and provides empirical grounding for the inclusive design of emotionally responsive digital environments tailored to neurodivergent users.