Automated Emotion Detection for Mental Health Applications Using Bidirectional Encoder Representations From Transformers (BERT)

Abstract

Mental health has become a major concern worldwide, especially amid the COVID-19 pandemic. This research aims to apply the RoBERTa algorithm in emotion analysis, particularly in the context of mental health. The data used consisted of 37,685 rows of commentar text and was processed through pre-processing, data labeling, and model evaluation stages. The pre-processing stage involved cleaning the data from unnecessary symbols and converting letters to lowercase. The data was then labeled based on emotion categories such as happy, sad, surprise, and neutral. The models used are RoBERTa and DistilBERT, with evaluation results showing that the RoBERTa model achieved an accuracy of 98.66%, higher than DistilBERT, which reached 98.17%. This research not only contributes to the development of technology in language analysis but also has a significant social impact in enhancing understanding and addressing mental health issues in society. It is hoped that this research can serve as a foundation for further application development in the field of mental health