Sentiment Analysis for Depression Detection Using Machine Learning Models

Abstract

Based on the abundance of sentiment and emotional data on social networks, this study introduces a machine learning (ML) model that can predict depression at its initial stages, which is a significant issue of the population. As the mental health (MH) symptoms are complex and challenging to diagnose, the project concentrates on text-based analysis of digital footprint that can be scaled. It presents a new, hybrid deep learning (DL) architecture, which serves to produce better performance, as is the case with the Fast text Convolution Neural Network with Long Short-Term Memory (FCL) model . This FCL network aims to provide better text representation by incorporating Fast text word embeddings with a Convolutional Neural Network (CNN) to learn features of a global context and a Long Short-Term Memory (LSTM) component to learn features of local features and sequential dependencies. At the same time, the research confirms the classical approaches, including the Decision Tree (DT), which uses different psycholinguistic characteristics, and which proves its effectiveness and scalability in the classification of depressive texts. The overall aim is to develop and come up with high-quality and robust screening tools in mental health with the subsequent future directions in work towards Multimodal Machine Learning to process and integrate different type of data (e.g., image and text) by tackling fundamental problems in feature representation, alignment, and fusion through the use of developed neural network architectures, and thus, enhancing the computational power to analyze human communicative behaviours to predict mental health among others in an automated manner.