Depression Detection on Malay Dialects Using GPT-3

• Published in: 2022 IEEE-EMBS Conference on Biomedical Engineering and Sciences (IECBES) pp. 360 - 364
• Main Authors: Hayati, Mohamad Farid Mohd, Ali, Mohd Adli Md, Rosli, Ahmad Nabil Md
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.01.2022
• Subjects: Biological system modeling; Data collection; Depression; Medical services; Natural language processing; Task analysis; Transformers
• DOI: 10.1109/IECBES54088.2022.10079554

The increasing number of depression cases and lack of manpower in mental healthcare services call for alternatives use of Artificial Intelligence. Natural language processing (NLP) can offer help in the form of early detection and frequent assessments. However, such technology is still limited in the local Malaysian context. In this paper we aimed to deploy Malay NLP in the local mental healthcare services. We performed depression detection on dialectal Malay speeches, specifically Kuala Lumpur, Pahang, and Terengganu dialects. Generative Pre-Trained Transformer-3 (GPT-3), a large language model, was used to perform the task. We experimented with different hyperparameters to test the capability of GPT-3 in few-shot learning. The results obtained are promising considering the size of our dataset that is very limited. We hope to see more studies in the field for the better development of Malay NLP in future. Clinical Relevance- This research tests the possibility of using NLP technology in the local setting of Malaysia. The advancement of this technology will be beneficial in mental healthcare especially in term of the availability of services such as early detection and frequent assessment.