Latent Preference Modeling for Cross-Session Personalized Tool Calling

2026-04-20

AI Summary

This work addresses the challenge of incomplete tool-call parameters arising from users’ omission of critical details during interactions with large language model (LLM) agents. To study cross-session personalized tool calling, the authors introduce MPT, a benchmark comprising 265 multi-turn dialogues, and propose PRefine—a test-time method that leverages a memory-augmented mechanism to model user preferences as dynamically evolving hypotheses. PRefine employs a generate–verify–refine loop to extract reusable constraints from historical dialogues, innovatively focusing on memorizing the rationale behind user choices rather than merely recording the choices themselves. Using only 1.24% of historical prompt tokens, this approach significantly improves tool-call accuracy, demonstrating that reasoning-based preference memory is crucial for robust personalization in LLM-agent interactions.

Technology Category

• Knowledge Representation and Reasoning: Preferences
• Multiagent Systems: Mechanism Design
• Machine Learning: Learning Preferences or Rankings

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Users often omit essential details in their requests to LLM-based agents, resulting in under-specified inputs for tool use. This poses a fundamental challenge for tool-augmented agents, as API execution typically requires complete arguments, highlighting the need for personalized tool calling. To study this problem, we introduce MPT, a benchmark comprising 265 multi-session dialogues that cover three challenges: Preference Recall, Preference Induction, and Preference Transfer. We also propose PRefine, a test-time memory-augmented method that represents user preferences as evolving hypotheses. Through a generate--verify--refine loop, it extracts reusable constraints from history and improves tool-calling accuracy while using only 1.24% of the tokens required by full-history prompting. These results indicate that robust personalization in agentic systems depends on memory that captures the reasons behind user choices, not just the choices themselves.

Problem

• personalized tool calling
• latent preference modeling
• cross-session
• under-specified inputs
• tool-augmented agents

Innovation

• Latent Preference Modeling
• Memory-Augmented Inference
• Tool Calling
• Cross-Session Personalization
• Generate-Verify-Refine Loop