AI Enabled User-Specific Cyberbullying Severity Detection with Explainability

User context matters for cyberbullying detection: The model achieved 98% accuracy by incorporating victim-specific factors like self-esteem, anxiety, depression, and demographic attributes alongside comment content—suggesting that the same comment may cause different harm depending on who receives it and their psychological state.

Demographic vulnerabilities are measurable: The explainability analysis (using SHAP and LIME) revealed that "victims belonging to specific racial and gender groups are more frequently targeted and exhibit higher incidence" of severe cyberbullying, highlighting systemic disparities in who experiences the most harmful attacks.

Psychological factors should inform platform responses: By integrating mental health indicators (self-esteem, anxiety, depression) into severity assessment, the research suggests that platforms could prioritize interventions for vulnerable users rather than treating all cyberbullying incidents equally.

Explainability is critical for fairness: The use of SHAP and LIME techniques to interpret model decisions helps identify which factors drive severity classifications, enabling stakeholders to audit whether the system is making fair distinctions or reinforcing biases based on protected characteristics like race and gender.

One-size-fits-all approaches are insufficient: Traditional models that only analyze comment content miss crucial context about victim vulnerability, suggesting that responsible AI systems for content moderation must account for individual differences rather than applying uniform standards.