Bangla Social Media Comments Analysis Using Machine Learning And Deep Learning Approaches

Abstract

Depression as a mental health problem is emerging as a serious problem in the world, while people take social networks as the only place where they can share their feelings and experiences. Identifying depressive comments in SNS can help find those people in need primary intervention to prevent the disease(s). As there is scarce literature on mental health prediction using Bangla text, this study aims to build a Bangla depressive comment detection system using machine learning and deep learning algorithm. This system works on a dataset of 3420 Bangla social media comments classified into Depressive and Non-Depressive. Categorized into main categories that are the classic methods including: SVM, Logistic Regression and Decision Trees, the deep learning methods like LSTM networks and CNNs. The proposed models are designed to predict depressive comments, keeping in mind the dispersed features of regular Bangla text. The assessment of the system is done comprehensively where parameters such as accuracy, precision, recall and F1-score are used in measuring the efficiency of several models.