Deep Learning Approaches for Predicting Mental States through Tweet Analysis

Depression, a globally pervasive mental health concern, significantly impacts individuals' cognitive processes, emotional well-being, and behaviors, giving rise to the alarming trend of depression-related suicides in recent years. This critical issue necessitates urgent attention and innovative approaches to address it. The usage of social media platforms has provided a unique and valuable opportunity. These platforms have evolved into channels for individuals, including those battling depression, to express their innermost thoughts, feelings, and emotions. The central objective of our study is to harness this wealth of information and explore the possibility of predicting a user's mental state by effectively distinguishing between depressive and non-depressive tweets on the Twitter platform. Leveraging the textual content of these tweets, we employ advanced deep learning models to analyze the semantic context woven within the textual narratives meticulously. This paper dives into a comprehensive analysis of detecting depression using deep learning models, including Simple Recurrent Neural Networks, Long Short-Term Memory (LSTM), Bidirectional LSTM, and Bidirectional Encoder Representations from Transformers (BERT) in combination with a Multilayer Perceptron (MLP), to study and classify the tweets. When examining the complete dataset, the LSTM model performed better with 99.27% accuracy. In contrast, when working with a balanced dataset, the BERT model coupled with a Multilayer Perceptron excelled, achieving an accuracy of 98.49%.