Topic modelling analysis of depression therapy text: a preliminary study / Teh Faradilla Abdul Rahman ... [et al.]

• Main Authors: Abdul Rahman, Teh Faradilla, Mohd Said, Raudzatul Fathiyah, Buja, Alya Geogiana, Mat Nayan, Norshita
• Format: Article
• Language: English
• Published: UiTM Cawangan Perlis 2024
• Subjects: Algorithms
• Online Access: https://ir.uitm.edu.my/id/eprint/94363/1/94363.pdf; https://ir.uitm.edu.my/id/eprint/94363/
• Institution: Universiti Teknologi Mara

The coronavirus disease 2019 (COVID-19) that has plagued the world since 2019 has initiated several issues and challenges in the mental health services field. World Health Organisation (WHO) recommended implementing remote mental health services such as telehealth to reach out to patients. One of telehealth services is text messaging therapy. Despite the challenges in treating depression via text messaging, the text messages for depression therapy that were built with different content renders this situation as a captivating subject for study. Nonetheless, the topics included in depression mobile therapy are scarce, particularly from the short text perspective. Fortunately, a machine learning technique known as topic modelling (TM) can be used to extracts topics from a set of documents without manually reading individual documents. It is very useful in searching for topics contained in short texts. This study aims to determine the topics in the text messages sent by mental health practitioners for depression therapy. In this study, three topic modelling techniques, i.e., Biterm Topic Model (BTM), Word Network Topic Model (WNTM), and Latent Feature Dirichlet Multinomial Mixture (LFDMM), were evaluated on 258 text messages of depression therapy. The performance of the TM techniques was evaluated using classification accuracy, clustering, and coherence scores. The findings indicate that the set of text messages comprises five topics. BTM performed better than the other techniques in classification accuracy and clustering in some cases based on the performance measures. Consequently, not much significant difference was found in the coherence score between the three topic modelling.