FORTUNE: Financial ORiented Tool Utilizing llms for Novel rEports
Keywords: Artificial intelligence system, Financial research report generation, Large language model, Data mining
Abstract: In recent years, the development of Large Language Models (LLMs) such as GPT-4 has markedly influenced various sectors, especially in text generation. Financial reports, crucial for economic decision-making, can be enhanced through LLMs, potentially revolutionizing the creation of financial research reports. However, LLMs encounter challenges in generating comprehensive financial reports due to limitations in real-time knowledge acquisition, context length, and data visualization capabilities. Existing methodologies have only partially addressed these issues, leading us to propose a groundbreaking framework, FORTUNE (Financial ORiented Tool Utilizing LLMs for Novel rEports), which employs a multi-tiered agent collaboration for financial report generation. FORTUNE utilizes an `agent-tree' architecture, wherein a hierarchical network of agents operates under a parent-child directive system, effectively managing real-time data collection, analysis, and visualization within token constraints. To address the issue of LLMs being unable to obtain real-time data, we developed FinKit, a toolkit comprising a series of tools capable of acquiring various types of real-time financial data. This toolkit was generated by proposed Multi-stage Cross Generation (MSCG) strategy and employs a Coarse-to-Fine Retrieval (CFR) strategy. Our evaluations show that FORTUNE enables efficient data integration, resulting in the production of user-tailored, comprehensive financial reports. This innovative method provides a new perspective for financial analysts and institutions seeking to leverage AI-driven insights.
Submission Number: 2
Loading