Leveraging BERT-Enhanced MLP Classifier for Automated Stress Detection in Social Media Articles

The surge in stress-related content on social media platforms has sparked concerns over the mental well- being of online community members. This study addresses the detection of stress from social media articles, focusing on data gathered from platforms such as Reddit and Twitter. Through the application of machine learning techniques, our objective is to create automated tools capable of identifying stress-inducing content and tracking mental health trends in online discourse. To achieve this, we compiled a dataset comprising over 8000 social media articles from Reddit and Twitter, each labeled as stress-inducing or non-stress-inducing. Our analysis began with exploratory data analysis to uncover insights into the dataset's characteristics, followed by preprocessing steps to clean and prepare the text data for modeling. Experiments using various machine learning models, including Logistic Regression, Gaussian Processes Classifier, and BERT- Enhanced MLP Classifier, were then conducted. The latter leverages BERT embeddings to capture contextualized representations of the text data, resulting in superior classification performance. The results demonstrate the BERT- Enhanced MLP Classifier's effectiveness, achieving an accuracy of 0.95 and surpassing baseline models. This study contributes to the advancement of automated tools for mental health monitoring and support in online communities, potentially aiding in timely interventions for individuals at risk.