SafeCRS: Personalized Safety Alignment for LLM-Based Conversational Recommender Systems

Researchers introduce SafeCRS, a safety-aware training framework for LLM-based conversational recommender systems that addresses personalized safety vulnerabilities. The system reduces safety violation rates by up to 96.5% while maintaining recommendation quality by respecting individual user constraints like trauma triggers and phobias.

• →Current LLM-based conversational recommender systems lack personalized safety protections that could harm users with specific sensitivities.
• →SafeRec benchmark dataset was created to systematically evaluate safety risks in LLM-based recommendation systems.
• →SafeCRS framework integrates Safe Supervised Fine-Tuning with Safe Group reward-Decoupled Normalization Policy Optimization.
• →The system achieved 96.5% reduction in safety violations compared to baseline recommendation systems.
• →The framework maintains competitive recommendation quality while prioritizing user-specific safety constraints.