User-Oriented Multi-Turn Dialogue Generation with Tool Use at scale

TL;DR

This paper introduces a scalable, user-oriented multi-turn dialogue generation framework that enhances tool use in large reasoning models by simulating authentic, extended human-agent interactions for complex, open-ended tasks.

Contribution

It proposes a novel user-oriented simulation paradigm and a flexible generation pipeline to produce high-quality, multi-turn dialogues with dynamic tool use at scale.

Findings

Generated high-density, multi-task dialogue datasets

Enabled realistic, extended human-agent interaction simulations

Improved multi-turn dialogue quality with user behavior modeling

Abstract

The recent paradigm shift toward large reasoning models (LRMs) as autonomous agents has intensified the demand for sophisticated, multi-turn tool-use capabilities. Yet, existing datasets and data-generation approaches are limited by static, predefined toolsets that cannot scale to the complexity of open-ended human-agent collaboration. To address this, we initially developed a framework for automated task-oriented multi-turn dialogue generation at scale, utilizing an LRM-based simulator to dynamically generate high-value, domain-specific tools to solve specified tasks. However, we observe that a purely task-oriented design often results in "solely task-solving" trajectories, where the agent completes the objective with minimal interaction, failing to generate the high turn-count conversations seen in realistic scenarios. To bridge this gap, we shift toward a user-oriented simulation paradigm. By decoupling task generation from a dedicated user simulator that mimics human behavioral rules - such as incremental request-making and turn-by-turn feedback - we facilitate more authentic, extended multi-turn dialogues that reflect the iterative nature of real-world problem solving. Our generation pipeline operates as a versatile, plug-and-play module capable of initiating generation from any state, ensuring high scalability in producing extended tool-use data. Furthermore, by facilitating multiple task completions within a single trajectory, it yields a high-density dataset that reflects the multifaceted demands of real-world human-agent interaction.