Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FinToolSyn: A forward synthesis Framework for Financial Tool-Use Dialogue Data with Dynamic Tool Retrieval

Caishuang Huang, Yang Qiao, Rongyu Zhang, Junjie Ye, Pu Lu, Wenxi Wu, Meng Zhou, Xiku Du, Tao Gui, Qi Zhang, Xuanjing Huang

Abstract

Tool-use capabilities are vital for Large Language Models (LLMs) in finance, a domain characterized by massive investment targets and data-intensive inquiries. However, existing data synthesis methods typically rely on a reverse synthesis paradigm, generating user queries from pre-sampled tools. This approach inevitably introduces artificial explicitness, yielding queries that fail to capture the implicit, event-driven nature of real-world needs. Moreover, its reliance on static tool sets overlooks the dynamic retrieval process required to navigate massive tool spaces. To address these challenges, we introduce \textit{FinToolSyn}, a forward synthesis framework designed to generate high-quality financial dialogues. Progressing from persona instruction and atomic tool synthesis to dynamic retrieval dialogue generation, our pipeline constructs a repository of 43,066 tools and synthesizes over 148k dialogue instances, incorporating dynamic retrieval to emulate the noisy candidate sets typical of massive tool spaces. We also establish a dedicated benchmark to evaluate tool-calling capabilities in realistic financial scenarios. Extensive experiments demonstrate that models trained on FinToolSyn achieve a 21.06\% improvement, providing a robust foundation for tool learning in financial scenarios.

FinToolSyn: A forward synthesis Framework for Financial Tool-Use Dialogue Data with Dynamic Tool Retrieval

Abstract

Tool-use capabilities are vital for Large Language Models (LLMs) in finance, a domain characterized by massive investment targets and data-intensive inquiries. However, existing data synthesis methods typically rely on a reverse synthesis paradigm, generating user queries from pre-sampled tools. This approach inevitably introduces artificial explicitness, yielding queries that fail to capture the implicit, event-driven nature of real-world needs. Moreover, its reliance on static tool sets overlooks the dynamic retrieval process required to navigate massive tool spaces. To address these challenges, we introduce \textit{FinToolSyn}, a forward synthesis framework designed to generate high-quality financial dialogues. Progressing from persona instruction and atomic tool synthesis to dynamic retrieval dialogue generation, our pipeline constructs a repository of 43,066 tools and synthesizes over 148k dialogue instances, incorporating dynamic retrieval to emulate the noisy candidate sets typical of massive tool spaces. We also establish a dedicated benchmark to evaluate tool-calling capabilities in realistic financial scenarios. Extensive experiments demonstrate that models trained on FinToolSyn achieve a 21.06\% improvement, providing a robust foundation for tool learning in financial scenarios.
Paper Structure (96 sections, 6 equations, 8 figures, 14 tables)

This paper contains 96 sections, 6 equations, 8 figures, 14 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Tool Calling for LLMs
  4. Synthetic Data for Tool Use
  5. Candidate Tool Retrieval
  6. LLMs in Finance
  7. FinToolSyn
  8. Persona-Grounded Financial Instruction Synthesis
  9. Open-Ended User Persona Inference
  10. Corpus-Based Financial Instruction Synthesis
  11. Task-Decomposed Atomic Tools Forward Synthesis
  12. Task Decomposition and Tool Synthesis
  13. Dual-Layer Verification and Evolution
  14. Vector Index and Tool Graph Construction
  15. Globally-Controlled Dynamic Retrieval Dialogue Generation
  16. ...and 81 more sections

Figures (8)

  • Figure 1: Comparison between reverse synthesis and authentic interaction dynamics.
  • Figure 2: The architecture of FinToolSyn. Our framework employs a forward synthesis paradigm that progresses from persona-based intent generation to dynamic tool retrieval, effectively capturing the implicit and event-driven nature of real-world financial inquiries.
  • Figure 3: Performance comparison of various general capabilities before and after training different models.
  • Figure 4: FC Mode data pipeline: (1) original MCP tool schema, (2) converted input schema passed to the model API, and (3) expected model output format verified by the Format Compliance check in CB-HWS Phase 1.
  • Figure 5: System Prompt for Prompt Mode
  • ...and 3 more figures