Enhancing conversational sentimental analysis for psychological depression prediction with Bi-LSTM

Abstract

Human mental health (HMH) is a pervasive and impactful condition that profoundly affects an individual’s cognitive, emotional, and behavioural aspects in a negative manner. Among various mental health disorders, depression is particularly prevalent, with approximately 20% of women experiencing at least one depressive episode during their lifetime. Identifying depression early on is crucial for timely intervention and support. This study examines user-generated content from major social platforms like Twitter, Facebook, and Instagram, aiming to detect potential signs of depression through behavioural symptoms such as mood changes, loss of interest, altered sleep patterns, focus difficulties, and impaired decision-making. Leveraging natural language processing and machine learning, sentiment analysis deciphers emotional context in posts and comments. A new efficient methodology utilizing Bidirectional Encoder Representations from Transformers (BERT) is proposed for efficient analysis of the posts and comments. Knowledge distillation transfers insights from a large BERT model to a smaller one, enhancing accuracy. Integrating word2vec and BERT with bidirectional long short-term memory (Bi-LSTM), the approach effectively analyses depression and anxiety indicators in social media data. Comparative assessments highlight the system’s excellence, achieving a remarkable 98.5% accuracy through knowledge distillation. The proposed methodology marks a substantial stride in identifying mental health signals from social media, facilitating better early intervention and support for those facing depression and anxiety-related challenges.