Leveraging Rule-Based Filtering and Explainable Al for Improved Detection of Cyber Bullying Comments
V. Sheeba, D. Raghu Raman, G. Abarna, M. Akash · 2025
In the era of pervasive social media use, cyberbullying has emerged as a significant threat to online communities, leading to severe psychological impacts on victims. This project presents a novel framework that integrates rule-based filtering with Explainable AI (XAI) to detect and mitigate bullying comments on social media platforms. The proposed system leverages a dynamic rule-based engine that uses keyword matching and pattern recognition to identify potentially harmful content. To enhance detection accuracy, the system incorporates advanced Natural Language Processing (NLP) techniques, such as sentiment analysis and contextual text classification, which work in tandem with the rule-based approach. To address the challenge of interpretability in AI-driven decisions, the framework employs Explainable AI methodologies, specifically LIME (Local Interpretable Model-Agnostic Explanations) and SHAP (SHapley Additive exPlanations). These tools provide transparency in the decision-making process by explaining why a particular comment is flagged as bullying. The combination of rule-based filtering and XAI not only improves the system’s accuracy but also builds user trust by making the detection process understandable to end-users and administrators. The effectiveness of the proposed framework is validated through extensive experimentation on real-world social media datasets, showing a significant reduction in false positives and an overall improvement in detection rates.