Comparing qualitative thematic analysis and machine-based topic modelling in the analysis of autistic and ADHD young people's accounts of emotions.

Systematic analysis of interview data can provide important insights into how young people experience and interpret their emotions. Both human-led qualitative (e.g., thematic analysis) and machine-driven quantitative (e.g., natural language processing [NLP]) analytical approaches are available, but their solutions are rarely compared. Interview responses by 57 neurodivergent adolescents to questions about their emotions, previously analysed using reflexive thematic analysis (RTA), were submitted to Topic Modelling (TM). Topic labels were developed in collaboration with neurodivergent co-researchers to ground their meaning in the lived experience of neurodivergent communities. Topics were mapped to RTA themes or subthemes to examine their proximities. Topic-based cluster analysis was used to identify participant groupings with similar topic distributions. TM revealed 10 interpretable and meaningful emotion-related topics - some closely overlapping with and others differing from the RTA themes. TM topics differentiated the young people's emotional experiences at school from those in other settings. TM and RTA resulted in overlapping and different insights into the meaning of neurodivergent young people's accounts of their emotions. Our findings demonstrate the potential use of TM in interview analysis and might suggest a potential complementarity between the TM topics and RTA themes, to be further explored using more advanced algorithms and a more sophisticated NLP implementation.