# Latent Preference Modeling for Cross-Session Personalized Tool Calling

Source: [Latent Preference Modeling for Cross-Session Personalized Tool Calling](<https://www.sota2.com/research/paper/69e7050b7d0172140ec7f237>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-04-24T06:00:02.177Z

Authors: Yejin Yoon, Minseo Kim, Taeuk Kim

Published: 2026-04-20T06:57:50.000Z

## 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.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Latent Preference Modeling on MPT Context-Free, Preference Recall](<https://www.sota2.com/research/sota/latent-preference-modeling-on-mpt-context-free-preference-recall.md>): Latent Preference Modeling; MPT Context-Free, Preference Recall; Precision: 76.1.

- [Latent Preference Modeling on MPT Context-Free Preference Induction](<https://www.sota2.com/research/sota/latent-preference-modeling-on-mpt-context-free-preference-induction.md>): Latent Preference Modeling; MPT Context-Free Preference Induction; Precision: 0.5487.

- [Latent Preference Modeling on MPT Context-Free, Preference Transfer](<https://www.sota2.com/research/sota/latent-preference-modeling-on-mpt-context-free-preference-transfer.md>): Latent Preference Modeling; MPT Context-Free, Preference Transfer; Precision: 30.92.

- [Latent Preference Modeling on MPT Context-Free Average](<https://www.sota2.com/research/sota/latent-preference-modeling-on-mpt-context-free-average.md>): Latent Preference Modeling; MPT Context-Free Average; F1 Score: 58.5.

- [Preference-driven Tool Calling on MPT Context-Guided, Preference Recall](<https://www.sota2.com/research/sota/preference-driven-tool-calling-on-mpt-context-guided-preference-recall.md>): Preference-driven Tool Calling; MPT Context-Guided, Preference Recall; P-EM: 64.88.

- [Preference-driven Tool Calling on MPT Context-Guided, Preference Induction](<https://www.sota2.com/research/sota/preference-driven-tool-calling-on-mpt-context-guided-preference-induction.md>): Preference-driven Tool Calling; MPT Context-Guided, Preference Induction; P-EM: 37.95.

- [Preference-driven Tool Calling on MPT Context-Guided, Preference Transfer](<https://www.sota2.com/research/sota/preference-driven-tool-calling-on-mpt-context-guided-preference-transfer.md>): Preference-driven Tool Calling; MPT Context-Guided, Preference Transfer; P-EM: 26.19.

- [Preference-driven Tool Calling on MPT Context-Guided Average](<https://www.sota2.com/research/sota/preference-driven-tool-calling-on-mpt-context-guided-average.md>): Preference-driven Tool Calling; MPT Context-Guided Average; OA-F1: 67.18.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2604.17886>)

[PDF](<https://arxiv.org/pdf/2604.17886.pdf>)