AWARE-US: Preference-Aware Infeasibility Resolution in Tool-Calling Agents

Researchers developed AWARE-US, a system to improve AI agents' ability to handle failed database queries by intelligently relaxing the least important user constraints rather than simply returning 'no results'. The system uses three LLM-based methods to infer constraint importance from dialogue, achieving up to 56% accuracy in correct constraint relaxation.

• →New framework addresses common failures in tool-calling AI agents when database queries return empty results
• →Three methods developed to rank constraint importance: local weighting, global one-shot weighting, and pairwise ranking
• →Global weighting method achieved highest correct-relaxation accuracy at 56% in car recommendation experiments
• →AWARE-US benchmark introduces 120+ persona-grounded queries for testing agent disambiguation and infeasibility resolution
• →System outperforms existing baselines by preserving user preferences when relaxing query constraints