Dirichlet multinomial modelling approaches in analysing anxiety therapy messages/ Teh Faradilla Abdul Rahman ... [et al.]

Despite the effectiveness of anxiety therapy through text messages, limited re-search was found to analyze the topics included in the therapy session. It is also unclear of which topic modelling approaches is the best in extracting anxiety therapy topics from text messages. Thus, this study aims to compare the performance of four topic modelling methods, namely Latent Feature Dirichlet Multinomial Mixture (LFDMM), Gibbs Sampling Dirichlet Multinomial Mixture, Generalized Polya-urn Dirichlet Multinomial Mixture and Poisson-based Dirichlet Multinomial Mixture Model on 28 text messages of anxiety-therapy. Four combinations of parameter settings were applied in the experiments to compare and decide the most suitable ones for future analysis. The performance of the topic modelling was evaluated using classification accuracy, clustering, and coherence scores. LFDMM has the best accuracy (34.10%) and clustering scores (0.5000, 0.4808) with combinations of hyperparameters α = 0.1 and β = 0.01 to infer more relevant topics. This study contributes to the growing body of research on topics in anxiety therapy, offering insights into the role of topic modelling in shaping valuable content. The findings highlight the potential of topic modelling for therapy con-tent exploration, aiding in text messages strategies for anxiety intervention.