A Framework for Mental Health Detection In Teenagers From Online Social Networks

Abstract

This report introduces a framework for detecting mental health indicators from teenagers' social media posts sourced from various mental health-related subreddits on Reddit. The framework utilizes natural language processing (NLP) techniques, specifically employing the BERT (Bidirectional Encoder Representations from Transformers) model for sequence classification. The dataset used comprises posts from categories including addiction, ADHD, anxiety, autism, bipolar disorder, loneliness, and schizophrenia. Prior to model development, the dataset undergoes preprocessing steps such as text cleaning and tokenization. The methodology includes label encoding and the development of a BERT-based sequence classification model. Evaluation of the model's performance is conducted using stratified k-fold cross validation, with metrics including accuracy, precision, recall, and F1-score employed to assess the effectiveness of the detection framework. This framework distinguishes itself from existing models by focusing specifically on detecting mental health indicators from teenagers' social media posts, leveraging advanced NLP techniques and a diverse dataset from Reddit. The outcomes underscore the potential of this approach in enhancing early detection and support for at-risk individuals within the realm of adolescent mental health. The findings contribute to advancing mental health informatics by showcasing the efficacy of NLP-driven approaches for surveillance and intervention using social media data. Future research directions include refining model performance and exploring broader applications of the detection framework in public health initiatives