Stress Detection In Social Media Posts Using Hybrid CNN And Feedforward Neural Networks

Abstract

In the recent few years, the unsparing use of social media platforms has provided researchers with a novel opportunity to analyze the behavior of users and patterns concerning mental health. This paper proposes a deep learning method for psychological stress detection in social media posts, via a hybrid architecture combining convolutional neural networks (CNNs) and feedforward neural networks (FNNs). The Dreaddit data set, which contains stress-labeled entries of Reddit posts, was used to preprocess and tokenize textual data, which was then padded so that all instances had the same sequence length. The proposed model extracts local semantic features by CNN layers followed by dense feedforward layers for classification. The training and evaluation of the model were carried out with an 80-20 train-validation split. The model reached an accuracy of 99% on the training dataset, with a precision of 99%, a recall of 99%, and an F1 score of 99%, suggesting a good fit of the model on training data. Although overfitting is an issue, the high assessment on train data provides confidence in the ability of this model to learn meaningful representations from text inputs indicating stress. This hybrid deep learning framework sets a good starting point for the development of automated mental health diagnostics from social media-derived data.