Real-time NLP-based stress detection in social media for digital mental health intervention
This study presents an NLP-based machine learning system for detecting stress in social media posts, enabling timely digital mental health intervention. A dataset of 45,792 posts from Reddit and Twitter was compiled, cleaned, tokenised, lemmatised, and balanced using Random Oversampling, with sentim...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Proceeding Paper |
| Language: | en en |
| Published: |
IEEE
2025
|
| Subjects: | |
| Online Access: | http://irep.iium.edu.my/127113/7/127113_Real-time%20NLP-based%20stress.pdf http://irep.iium.edu.my/127113/8/127113_Real-time%20NLP-based%20stress_Scopus.pdf http://irep.iium.edu.my/127113/ https://ieeexplore.ieee.org/document/11233519 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | This study presents an NLP-based machine learning system for detecting stress in social media posts, enabling timely digital mental health intervention. A dataset of 45,792 posts from Reddit and Twitter was compiled, cleaned, tokenised, lemmatised, and balanced using Random Oversampling, with sentiment features extracted via VADER. TF-IDF and sentiment scores were used to train four classifiers—Logistic Regression, LinearSVC, Random Forest, and XGBoost—evaluated on accuracy, precision, recall, F1-score, and inference time. LinearSVC achieved the highest F1-score (0.898) and fastest GUI inference (2.44 s), demonstrating strong performance and sensitivity to subtle stress cues. A Gradio-based GUI enables instant, accessible predictions, validating the system’s practicality. The results confirm the feasibility of combining linguistic and sentiment analysis for scalable, real-time stress detection, laying a foundation for future integration with cyberincivility monitoring in digital health tools. |
|---|