Evaluating the Influence of Text Representation techniques on Diverse Machine Learning Algorithms for Stress Detection in Social media users

Over the past few years, there has been a notable surge in social media usage like Twitter, Facebook, Instagram. Due to the anonymity of the internet, people's mental health is frequently reflected in their social media activity. Through such activities, psychiatric problems are frequently discovered and can be treated in their early stages, potentially preventing the effects of untreated mental disorders like depression and anxiety. The present study aims to examine the potential consequences of utilising two commonly used methods for text representation, namely Term Frequency-Inverse Document Frequency(TF-IDF) and Bag-of-Words(BoW). The specific focus of this investigation is on the application of these techniques to the task of stress detection. The aim is to comprehend how these methods influence the effectiveness of different classifiers in identifying stress patterns within textual data. By assessing the impact of TF-IDF and BoW, this research seeks to enhance our understanding of their roles in shaping the performance of various algorithms. This study encompasses various machine learning algorithms, including Support Vector Machines(SVM), Random Forest, Logistic Regression, and Naive Bayes. In the case of deep learning techniques, Recurrent Neural Network(RNN), Long Short Term Memory(LSTM), RoBerta, and implementation of BERT embeddings in Recurrent Neural Networks was studied. The experimentation involves extensive datasets taken from two social media platform, Reditt and Twitter to ensure the robustness and generalizability of findings. The results shed light on the varying effects of TF-IDF and BoW, offering insights into which techniques are better suited for stress detection and contributing to the optimization of future stress detection systems.