JOURNAL OF COMMUNICATION, LANGUAGE AND CULTURE

From Conversations to Insights: Analysing Social Networks for Early Mental Health Detection - A Systematic Review of Causal Inference and Deep Learning

Early detection of mental health issues is crucial for timely intervention, reducing the severity of conditions, and improving overall well-being. Social networks have emerged as valuable platforms for identifying mental health issues, thanks to user-generated content and social interactions. Artificial intelligence (AI), especially causal inference and deep learning, has great potential for analysing large-scale social media data. It helps identify patterns and relationships that enable earlier and more accurate prediction of mental health issues. The paper aims to systematically review academic articles on the applications of AI, including deep learning and causal inference in social networks for early mental health detection. The systematic review initially considered 1,018 academic articles retrieved from major scholarly databases, including IEEE Xplore, Scopus, and ScienceDirect. The review was conducted in accordance with the PRISMA framework to ensure a transparent and rigorous screening and selection process. After careful review, the articles were filtered down to 90 for full analysis to present a classification framework based on four dimensions: Applications in mental health, methods/techniques, datasets used, and challenges. It was identified that AI continues to significantly outperform humans in terms of accuracy, efficiency, and early intervention for mental health detection. Methods and techniques map directly to relevant keywords such as AI applications, causal inference, deep learning, and social networks. Although AI shows great promise in mental health detection, challenges such as data bias, privacy risks, and a lack of model interpretability remain. Combining causal inference with deep learning can create personalised mental health interventions. Still, future research must prioritise explainable AI, privacy-preserving methods, and ethical data collection to ensure responsible, transparent AI applications in mental health care.

Mental Health Disorder Detection Beyond Social Media: A Systematic Review of Available Datasets

Detecting mental health disorders in a timely manner is an important societal challenge. NLP and machine learning (ML) methods used to assist with detection rely on data collected primarily from social media. However, such datasets often have sampling biases and inherent ethical and privacy issues. One avenue to overcome these limitations is non-social media data. We present the first comprehensive review of non-social media, free-text datasets for mental health research. We use the PRISMA methodology to conduct our survey and we review datasets available in multiple languages. We find that non-social media free-text based datasets are predominantly focused on English and on detecting depression. These datasets also vary in demographics, platforms, data types, annotation techniques, and methodologies. This systematic review also reveals key gaps and highlights opportunities to develop more diverse, reliable and clinically-relevant resources.

Detection of Anxiety and Depression from Social Media Text Using Natural Language Processing

: The increasing global burden of anxiety and depression necessitates innovative computational approaches for early, non-intrusive detection. Social media platforms offer naturalistic data where individuals voluntarily express psychological states through text. This paper presents a comprehensive Natural Language Processing (NLP) pipeline for detecting linguistic correlates of anxiety and depression from social media text, a task distinct from clinical diagnosis. We systematically compare traditional machine learning models (Logistic Regression, SVM, Random Forest) and deep learning architectures (BiLSTM with Attention, fine-tuned BERT) across three benchmark datasets: SMHD, Dreaddit, and CLPsych 2015. Fine-tuned BERT achieves the highest macro-averaged F1 scores of 0.918, 0.909, and 0.862 respectively, outperforming traditional models by 7.1–17.3%. SHAP analysis identifies negative emotion, past-focused language, and somatic markers as the most predictive features. Cross-dataset validation reveals 15–25% performance degradation, highlighting the need for domain adaptation. Ethical considerations including false-positive risks and the gap between linguistic detection and clinical validity are discussed.

The African Journal of Information Systems The African

AI-Based Assessment of Mental Health Status in College Students through Social Network Data Analysis

This study aims to develop and validate an artificial intelligence (AI) algorithm for assessing college students' mental health status by analyzing their social network data, with the goal of establishing an effective early warning system for psychological issues. The research involved 1,500 college students from three universities. Social network data including posting frequency, content themes, interaction patterns, and behavioral characteristics were collected over six months. Machine learning algorithms, particularly natural language processing (NLP) and deep learning models, were employed to analyze the data. The participants also completed standardized psychological assessments (SCL-90, PHQ-9, GAD-7) to validate the AI algorithm's accuracy. The AI model achieved an overall accuracy of 86.4% in identifying students with potential mental health issues. Specific indicators such as changes in posting frequency (p < 0.01), negative emotional vocabulary usage (p < 0.001), and social withdrawal patterns (p < 0.05) were significant predictors. The algorithm demonstrated strong performance in detecting depression (AUC = 0.89) and anxiety (AUC = 0.87) symptoms, with early detection capability averaging 4-6 weeks before clinical manifestation. AI-based analysis of social network data provides an effective, non-invasive method for assessing college students' mental health status. This approach enables timely intervention and support, offering significant potential for improving mental health monitoring and prevention strategies in educational settings. This work was supported by the NIT Science and Technology Innovation Fund under Grants No.XYJS202501; the National Science Foundation of China under Grant No. 62172229,62471258; The Natural Science Fund of Jiangsu Province under Grants No. BK20211295; Jiangsu Provincial Natural Science Research General Project for Colleges and Universities in 2024 (24KJB520034); Key Cultivation Project for Scientific Research of Nantong Institute of Technology (XKYPY202304). Yingya Zhi, School of Information Engineering, Nantong Institute of Technology, Nantong, 226000, China.

A Web-Based Mental Health Assessment Framework using Natural Language Processing Techniques

Mental illnesses like depression, anxiety, stress, and so on have become more widespread, and this has necessitated the availability of assessment tools that are readily available, scalable, and automated. The paper will offer a web-based mental health risk assessment system that utilizes the latest technology of Natural Language Processing (NLP) in real-time to analyze textual information provided by users. The offered system incorporates a hybrid deep learning framework with the relation of DeBERTa, BiLSTM, and XGBoost to promote the contextual comprehension, sequential emotional pattern identification, and effective classification of the performance. First, text input by the users is received with a secure web interface and processed with general NLP preprocessing methods, such as tokenization, lemmatization, and sentiment normalization. To extract deep semantic relationships in the text, DeBERTa is used to extract contextual embeddings. Such embeddings are also trained in the form of a Bidirectional Long Short-Term Memory (BiLSTM) network in order to capture emotional dynamics and linguistic reinforcing relations. Fused feature representation, sentiments, and linguistic indicators are input into an XGBoost classifier to predict mental health in multi-class. There is a weighted risk scoring system used to measure the level of severity and provide tailored feedback. It is experimentally tested on standard mental health text data sets that the proposed hybrid framework is more effective than the traditional machine learning and standalone transformer models in terms of accuracy, precision, recall, and F1-score. The architecture is scalable to the deployment of a web system, which is guaranteed to perform in real-time, secure data, and privacy of users. The suggested framework offers a solid and smart instrument to identify the risk of mental health early and help intervene in time and to promote the development of digital health care.