Identification of depressing tweets using natural language processing and machine learning: application of grey relational grades

Depression is a global public health concern that affects millions of people worldwide. Social media platforms, where individuals connect and share personal data, have emerged as potential sources for mental health detection. This study explored the use of computational models to identify individuals with depression based on Twitter posts. We retrieved and cleaned 1.6 million tweets using Natural Language Processing (NLP) techniques for feature extraction. The Grey Relational Grade (GRG) technique was applied to investigate the association between likes and shares of Twitter posts. Furthermore, the significant values of GRG in both cases, when data is limited and when data is large, represent that GRG provides better results at large data sets. The equal distri bution and selection approach (EDSA) can extract a small sample to describe the large data set and apply the GRG technique. Subsequently, we applied various machine learning models to classify user tweets into "stressed" or "not stressed" categories. These models achieved promising results, demonstrating high accuracy, precision, recall, and F1-score. Specifically, Logistic Regression, Support Vector Machine, XGBoost Classifier, and Random Forest Classifier yielded accuracies of 96, 95, 96, and 97%, respectively. These findings suggest the potential of social media data and computational models for mental health detection, thus opening avenues for further research and development.