Mental Health on Twitter in Turkey: Sentiment Analysis with Transformers

Social media are regarded as excellent mediums for capturing individuals’ everyday routines, interests, and ideologies. Also, these platforms are now used to exchange health-related information. By collecting and analyzing data from social media, hidden patterns of people’s activities, status, etc., can be gathered. The goal of this study is to employ machine learning to conduct sentiment analysis and assess the mental health of Turkish Twitter users. Using terms relating to common mental health issues such as anxiety, stress, depression, suicide, and eating disorders, we collected over 25,000 tweets. The data was then analyzed, and automated sentiment scoring for the Turkish language was applied using a transformer-based machine learning model. By utilizing BERT, our final deep-learning classifier showed 82.6% accuracy in predicting sentiment from tweets. This study shows how effective deep learning models and transformers are for Turkish natural language processing tasks. The findings may help to improve mental health services by providing a better understanding of the sentiment expressed in Turkish tweets about mental health.