AI-Driven Cyberbullying Detection Network using Google Perspective API and Keyword-based Hybrid Moderation System
[S1]AKSHAY S NAVALAGUNDA, [S2]DARSHAN BALRAJ ILIGER,[S3]MADAN KUMAR E S and [S4]PAVAN KUMAR K, STUDENT, AMCEC
[G1]PRASHANT KUMAR MISHRA, PROFESSOR, AMCEC
Department of Artificial Intelligence and Machine Learning, AMCEC, VTU, India
E-mail : [S1] akshaynavalagund@gmail.com , [S2] darshubi9686@gmail.com , [S3] madankumares023@gmail.com , [S4] pkumark027@gmail.com , [G1] mishraaprashant21@gmail.com
Abstract— Cyberbullying has emerged as one of the most critical challenges in online communities and digital communication networks. With the exponential growth of social media usage, users—especially young individuals—are frequently exposed to abusive language, hate speech, harassment and toxic behaviour. Traditional moderation approaches rely on manual review or static keyword filters which fail to scale in high engagementenvironments.
This research presents a web-based cyberbullying prevention system that integrates Google Perspective API, a pre-trained NLP model capable of returning real-time toxicity scores, along with a custom keyword-based fallback module to detect slang terms that may bypass the API. The system is implemented as a mini social network platform that supports posts, comments, likes, following mechanism and instant notifications. Unlike popular research articles which involve creating or training ML models, this work focuses on practical enforcement, real-time moderation, and hybrid decision logic for safer user interaction.
The model has not been trained locally, instead utilizes Google's deep learning models from the cloud, making the system faster to develop, resource-friendly and deployable for academic institutions. The platform can be extended into a full-scale AI moderation ecosystem with multilingual support, deeper contextual detection and dataset-driven training.
Keywords: Cyberbullying Detection, Natural Language Processing, Toxicity Filtering, Perspective API, Flask, Real-Time Moderation, Online Safety
