JIITA, Vol.10 no.2 pp.1308-1320 (2026), DOI: 10.22664/ISITA.2026.10.2.1308
C. Kalpana, Manoj Devare , Bhagya P Bijay Kumar
Abstract. The platforms for online communication have expanded rapidly. These are
creating new opportunities for interaction but also increasing the chances of offensive
and harmful content. These types of messages affect the mental health of the user, which
also affects the social well-being of the individual user. This shows the importance of
strategies for moderating such issues. In this work, we present a real-time framework
for detecting and blocking harmful text using Natural Language Processing (NLP) and
machine learning techniques. The text is represented with Term Frequency–Inverse
Document Frequency (TF–IDF) features and classified using Random Forest, Support
Vector Machine, and XGBoost algorithms. From our experiments and comparison the
tuned XGBoost model performed the best, reaching 90% accuracy with an F1-score of
0.75. The per-sample prediction latency was of only 0.004 seconds.
Keywords; AI based message detection, social well-being, Machine Learning, Natural
Language Processing, TF-IDF