DynFinRAG-Agent: A Dynamic Risk Segmentation and Graph-Enhanced Retrieval-Augmented Multi-Agent Framework for Multimodal Financial Risk Analysis and Automated Investment Decision-Making
DOI:
https://doi.org/10.66238/fsrma145Keywords:
Dynamic Risk Segmentation, Graph Neural Networks, Retrieval-Augmented Generation, Multi-Agent Large Language Models, Multimodal Financial Analysis, Financial Risk Management, Automated Investment Decision-MakingAbstract
The increasing heterogeneity of financial data and rapidly evolving user behaviors pose significant challenges to financial risk monitoring and automated investment decision-making, particularly when static risk Aprofiles and isolated models fail to capture temporal risk transitions, relational anomalies, and knowledge-intensive financial contexts. This study proposes DynFinRAG-Agent, a Dynamic Risk Segmentation and Graph-Enhanced Retrieval-Augmented Multi-Agent framework for multimodal financial risk analysis and automated investment decision-making. The framework integrates a Multimodal Financial Encoder (MFE) to jointly represent market time-series, corporate financial indicators, user behavioral records, and financial texts, while Dynamic Risk Segmentation (DRS) continuously identifies evolving user risk states. A Dynamic Risk Graph Network (DRG-Net) further models interactions among users, accounts, firms, assets, and financial events, and its risk representations are incorporated into Graph-Enhanced Retrieval-Augmented Generation (GF-RAG) to retrieve relevant regulatory rules, historical cases, and market knowledge. Specialized Multi-Agent Large Language Models (LLMs) then perform risk assessment, compliance analysis, market intelligence extraction, portfolio evaluation, and investment decision generation. Experiments on a temporally and entity-aligned benchmark constructed from multiple public financial datasets demonstrate that DynFinRAG-Agent achieves 95.1% accuracy and an Area Under the Curve (AUC) of 0.978 for financial risk prediction, while reaching 93.5% decision consistency in automated investment decision-making. These results indicate that dynamic risk segmentation, relational graph reasoning, and evidence-grounded multi-agent collaboration jointly improve the reliability and effectiveness of multimodal financial risk analysis and investment decisions.
References
[1] Hilal, Waleed, S. Andrew Gadsden, and John Yawney. "Financial fraud: a review of anomaly detection techniques and recent advances." Expert systems With applications 193 (2022): 116429.
[2] Zhu, Xiaoqian, et al. "Intelligent financial fraud detection practices in post-pandemic era." The Innovation 2.4 (2021).
[3] Hamilton, Will, Zhitao Ying, and Jure Leskovec. "Inductive representation learning on large graphs." Advances in neural information processing systems 30 (2017).
[4] Veličković, Petar, et al. "Graph attention networks." arXiv preprint arXiv:1710.10903 (2017).
[5] Rossi, Emanuele, et al. "Temporal graph networks for deep learning on dynamic graphs." arXiv preprint arXiv:2006.10637 (2020).
[6] Cheng, Dawei, et al. "Graph neural network for fraud detection via spatial-temporal attention." IEEE Transactions on Knowledge and Data Engineering 34.8 (2020): 3800-3813.
[7] Motie, Soroor, and Bijan Raahemi. "Financial fraud detection using graph neural networks: A systematic review." Expert Systems with Applications 240 (2024): 122156.
[8] Chen, Zhiyu, et al. "Finqa: A dataset of numerical reasoning over financial data." Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing. 2021.
[9] Huang, Allen H., Hui Wang, and Yi Yang. "FinBERT: A large language model for extracting information from financial text." Contemporary Accounting Research 40.2 (2023): 806-841.
[10] Wu, Shijie, et al. "Bloomberggpt: A large language model for finance." arXiv preprint arXiv:2303.17564 (2023).
[11] Yang, Hongyang, Xiao-Yang Liu, and Christina Dan Wang. "Fingpt: Open-source financial large language models." arXiv preprint arXiv:2306.06031 (2023).
[12] Wang, Ziao, et al. "Finvis-gpt: A multimodal large language model for financial chart analysis." arXiv preprint arXiv:2308.01430 (2023).
[13] Lewis, Patrick, et al. "Retrieval-augmented generation for knowledge-intensive nlp tasks." Advances in neural information processing systems 33 (2020): 9459-9474.
[14] Peng, Boci, et al. "Graph retrieval-augmented generation: A survey." ACM Transactions on Information Systems 44.2 (2025): 1-52.
[15] Hu, Yuntong, et al. "Grag: Graph retrieval-augmented generation." Findings of the association for computational linguistics: NAACL 2025. 2025.
[16] Yu, Yangyang, et al. "Fincon: A synthesized llm multi-agent system with conceptual verbal reinforcement for enhanced financial decision making." Advances in Neural Information Processing Systems 37 (2024): 137010-137045.
[17] Liu, Xiao-Yang, et al. "FinRL: Deep reinforcement learning framework to automate trading in quantitative finance." Proceedings of the second ACM international conference on AI in finance. 2021.
[18] Zhong, Ziwen, and Yuheng Huang. "CRE-CNN-LSTM-XAI: An Explainable Deep Learning Framework for Risk Assessment of Real Estate Collateralized Financial Products." Journal of Computer Science and Frontier Technologies 3.1 (2026): 127-140.
[19] Li, Zheyu, Yue Hao, and Peifan Zeng. "CLASNet: A Cognitive Load–Aware CNN-LSTM-Attention Framework for Supply Chain Demand Forecasting and Adaptive Human–Computer Interaction." Journal of Computer Science and Frontier Technologies 3.2 (2026): 44-57.
[20] Yao, Yilin, Wenchao Zhang, and Mengtong Li. "Cloud-Edge Federated Incremental Learning Framework for PMSM Efficiency Optimization with Lightweight CNN-LSTM Models and OTA Differential Deployment." Journal of Computer Science and Frontier Technologies 3.1 (2026): 98-111.
[21] Gao, Ziyang, Yuanliang Qu, and Yi Han. "Cross-Lingual Sponsored Search via Dual-Encoder and Graph Neural Networks for Context-Aware Query Translation in Advertising Platforms." arXiv preprint arXiv:2510.22957 (2025).
[22] Zhang, Yangyang, et al. "Cognitive Digital Twin Framework: Modeling and Real-Time Decision Making." ICLR 2026 the 2nd Workshop on World Models: Understanding, Modelling and Scaling. 2026.
[23] Xu, Chengde, Chongye Wang, and Zeyu Li. "H2-LBM: A Hierarchical Hybrid Deep Reinforcement Learning Framework for L7 Load Balancing and Global Traffic Scheduling in Multi-Cloud LLM Serving." Fundamental Scientific Reports in Multidisciplinary Areas 2.02 (2026): 109-120.
[24] Zhang, Jingsi, and Jixiang Ding. "CA-PPO-AXI: Coverage-Aware Deep Reinforcement Learning for Adaptive AXI Bus Protocol Verification." Fundamental Scientific Reports in Multidisciplinary Areas 2.02 (2026): 159-169.
[25] Hua, Zonghui, Yue Luo, and Yunting Ling. "LPG-SAC-TCO: A Large Language Model--Policy-Guided Soft Actor-Critic Framework for Total Cost of Ownership--Optimal Energy Management of Hydrogen Fuel Cell Heavy-Duty Trucks." European Journal of AI, Computing & Informatics 2.3 (2026): 1-11.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Yang Shi, Yujie Lu, Shaoli Qian

This work is licensed under a Creative Commons Attribution 4.0 International License.
License Terms (Recommended – CC BY 4.0)
All articles published in Fundamental Scientific Reports in Multidisciplinary Areas are licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0).
This license permits unrestricted use, distribution, and reproduction in any medium, provided the original author(s) and source are properly cited. Users are allowed to copy, redistribute, remix, transform, and build upon the material for any purpose, including commercial use, as long as appropriate credit is given.
Authors retain copyright of their work while granting the journal the right of first publication. The journal provides immediate open access to its content on the principle that making research freely available supports a greater global exchange of knowledge.
For more information about this license, visit:
https://creativecommons.org/licenses/by/4.0/