Detecting early signs of depression in twitter posts through LSTM-based sentiment analysis

Department of Computer Engineering, Dr. D. Y. Patil Institute of Technology, Pune, Maharashtra, India

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

Depression is one of the most common mental health disorders, yet a large proportion of the cases are left undiagnosed due to associated social stigma, limited awareness, and lack of timely clinical intervention. The unprecedented growth of social media has encouraged users to frequently express their emotions on the social Web, thus opening an opportunity for the early detection of psychological distress from their digital footprints. Most of the available research works target single-language sentiment analysis or few classifier benchmarking, and thus there is a lack in multilingual and comparison-based deep learning-based studies. This work attempts to fill such a research gap by proposing a supervised machine learning framework for the identification of depression indicators from both English and Marathi tweets. The proposed approach integrates the standard NLP preprocessing—tokenization, stop-word removal, stemming, and TF-IDF and Word2Vec-based feature extraction—with training multiple classifiers, namely Logistic Regression, SVM, Random Forest, and LSTM. The experimental results demonstrate that the LSTM model yields state-of-the-art performance compared to other traditional models due to its natural ability to capture contextual dependencies of texts. The findings have important implications for the social media-driven method for early mental health screening and public health monitoring and digital psychological intervention systems.