Musica ex machina: Towards AI-supported research into mechanisms of music-based therapies and interventions

Le résumé fourni par la source

Background: The expanding literature on music-based therapies and interventions makes it increasingly difficult for clinicians and researchers to identify relevant evidence on treatment mechanisms, moderators, mediators, and outcomes. Objective: This study explored whether artificial intelligence (AI) and natural language processing (NLP) methods could support the organisation and retrieval of information from this literature. Methods: An automated PubMed search using the terms “music therapy”, “music intervention”, and “sound therapy” retrieved 4,396 articles. Four approaches were evaluated: (i) expert-defined terminology, (ii) KeyBERT-based keyword extraction with k-means clustering, (iii) Word2Vec-based abstract representations with k-means clustering, and (iv) a BERT-based question-answering (QA) system for retrieving specific answers and source passages. Results: The expert-defined approach identified a limited set of frequently occurring terms, whereas KeyBERT generated a broader vocabulary and 10 clusters of related abstracts. Word2Vec produced 50 clusters, but the mean silhouette coefficient was 0.20, indicating weak cluster separation. The QA system retrieved potentially relevant passages from source articles. However, its answers and scores changed when the amount of input text or wording of the query was modified. Conclusions: AI-supported clustering and QA methods may help clinicians and researchers navigate the growing literature on music-based therapies and interventions. However, these findings are preliminary. Further work should focus on improving domain-specific model performance and representativeness. Such systems should support literature retrieval and professional judgment rather than replace clinical expertise.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.