Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Refining Financial Consumer Complaints through Multi-Scale Model Interaction

Bo-Wei Chen, An-Zi Yen, Chung-Chi Chen

TL;DR

This work tackles the problem of converting informal financial dispute statements into formally persuasive legal text by introducing the FinDR dataset and the Multi-Scale Model Interaction (MSMI) framework. MSMI combines a lightweight discriminating model with iterative, counterfactual refinement by large language models to improve output quality while avoiding self-bias and model reliance issues. Empirical results show MSMI outperforms single-pass prompting, enhances robustness across diverse LLMs, and generalizes to short-text benchmarks, all while preserving semantic content. The approach promises practical benefits for affordable, accessible, and reliable automated legal writing in high-stakes contexts.

Abstract

Legal writing demands clarity, formality, and domain-specific precision-qualities often lacking in documents authored by individuals without legal training. To bridge this gap, this paper explores the task of legal text refinement that transforms informal, conversational inputs into persuasive legal arguments. We introduce FinDR, a Chinese dataset of financial dispute records, annotated with official judgments on claim reasonableness. Our proposed method, Multi-Scale Model Interaction (MSMI), leverages a lightweight classifier to evaluate outputs and guide iterative refinement by Large Language Models (LLMs). Experimental results demonstrate that MSMI significantly outperforms single-pass prompting strategies. Additionally, we validate the generalizability of MSMI on several short-text benchmarks, showing improved adversarial robustness. Our findings reveal the potential of multi-model collaboration for enhancing legal document generation and broader text refinement tasks.

Refining Financial Consumer Complaints through Multi-Scale Model Interaction

TL;DR

This work tackles the problem of converting informal financial dispute statements into formally persuasive legal text by introducing the FinDR dataset and the Multi-Scale Model Interaction (MSMI) framework. MSMI combines a lightweight discriminating model with iterative, counterfactual refinement by large language models to improve output quality while avoiding self-bias and model reliance issues. Empirical results show MSMI outperforms single-pass prompting, enhances robustness across diverse LLMs, and generalizes to short-text benchmarks, all while preserving semantic content. The approach promises practical benefits for affordable, accessible, and reliable automated legal writing in high-stakes contexts.

Abstract

Legal writing demands clarity, formality, and domain-specific precision-qualities often lacking in documents authored by individuals without legal training. To bridge this gap, this paper explores the task of legal text refinement that transforms informal, conversational inputs into persuasive legal arguments. We introduce FinDR, a Chinese dataset of financial dispute records, annotated with official judgments on claim reasonableness. Our proposed method, Multi-Scale Model Interaction (MSMI), leverages a lightweight classifier to evaluate outputs and guide iterative refinement by Large Language Models (LLMs). Experimental results demonstrate that MSMI significantly outperforms single-pass prompting strategies. Additionally, we validate the generalizability of MSMI on several short-text benchmarks, showing improved adversarial robustness. Our findings reveal the potential of multi-model collaboration for enhancing legal document generation and broader text refinement tasks.

Paper Structure

This paper contains 10 sections, 1 figure, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Dataset
  4. Method
  5. Discriminating Model for Evaluation
  6. LLM-based Refinement via Counterfactual Generation
  7. Iterative Loop and Early Stopping
  8. Experiments
  9. Further Exploration
  10. Conclusion

Figures (1)

  • Figure 1: Comparison of generated counterfactual texts between our purposed task and the other classification tasks. The example of the legal text refinement task is a plaintiff’s claim in an insurance dispute, sampled from our dataset.