FinDialogLens: Revolution Event Extraction for Missed-Trade Identification in Multi-Party Financial Chatrooms

Summary:FinDialogLens is a novel hybrid LLM pipeline designed for event extraction and missed-trade identification in multi-party financial chatrooms. It employs compact fine-tuned classifiers to detect RFQ triggers and price/trade outcome metadata, an RFQ-Level Module to segment per-event RFQ windows, and a Trade Engine to fill argument roles. With GPT-4o, FinDialogLens achieves 92.1% and 94.3% accuracy on final price and trade outcome, respectively, outperforming full-chatroom Chain-of-Thought prompti

---

Technical Highlights

1. Hybrid LLM Pipeline Design

   • FinDialogLens integrates fine-grained classifiers, an RFQ-Level Module, and a Trade Engine, achieving efficient event extraction and trade identification through a modular design.

2. GPT-4o's Superior Performance

   • With GPT-4o, the system achieves 92.1% and 94.3% accuracy on final price and trade outcome, respectively, outperforming traditional methods.

3. Difficulty-Aware Router

   • The introduction of a difficulty-aware router allows the system to balance cost and accuracy, reducing LLM calls by 85% while maintaining high accuracy.

4. Potential of Open-Source LLMs

   • The study shows that fine-tuned open-source LLMs with as few as 3B parameters can achieve comparable performance with modest in-domain data, providing a feasible alternative for resource-constrained applications.

Industry Impact and Developer Recommendations

• Transformation in the Financial Sector: FinDialogLens offers a highly efficient and reliable tool for transaction identification in the financial industry, significantly improving transaction efficiency and reducing human errors.
• Promising Applications in Multi-Party Dialogues: The system demonstrates the potential of event extraction in complex multi-party dialogue environments, providing a reference for similar applications in other fields.
• Opportunities for Open-Source LLMs: Developers can leverage open-source LLMs for fine-tuning to meet specific application needs, thereby reducing reliance on high-end LLMs.

Conclusion

The advent of FinDialogLens marks a significant advancement in event extraction and trade identification technology for multi-party financial dialogues. Its modular design and efficient LLM calling strategy make it well-suited for large-scale applications.

---

Source: ArXiv NLP/LLM (cs.CL) (2026-10-05)

— END —

Tags: #FinDialogLens #Event Extraction #Multi-Party Dialogue #Fintech #LLMs & Foundation Models