Exploring Mental Health Status Correlations Through Topic Modeling on Social Media Data

Mental health, once a stigmatized and overlooked topic, is now widely discussed. As concerns related to mental health began to grow, people began to share their personal experiences on various platforms and also began to seek the help of professionals. With this growth, social networks have become a prominent space where people share what they are going through. The main goal of this study is to analyze social media data to uncover and understand key topics linked to different mental health disorders using topic modeling techniques. This could help mental health professionals gain deeper insights into each disorder and it can also help them understand the correlation between usage of language and various mental health disorders. As early identification and intervention are crucial in providing effective treatment and improved long-term results, this work could help identify those topics that contribute to these issues in social media texts. The social media platform is where people express their emotions in the most open and unfiltered way due to anonymity and the benefit of not being under the pressure of in person judgment. This work analyzed over 1,00,000 statements classified into seven mental health disorders such as personality disorders, suicide, stress, anxiety, bipolar, depression, and normal to identify prominent topics related to each. Topic modeling models such as LDA, NMF, Top2Vec, and BERTopic are used to implement this work and identify topics that could reflect a particular mental disorder and also differentiate between disorders. The evaluation is performed based on the factors of coherence score, silhouette score, Calinski-Harabasz score, and human interpretability.