DC-FGLLM: Financial Data Risk Prediction Based on Dual-Channel Network
Year: 2025
Type: conference-paper
Abstract: Amid global financial volatility and universities’ complex, diversified finances, traditional risk prediction methods—limited by linear assumptions, strict data distribution, and poor handling of high-dimensional long-term data—fail to meet dynamic early warning needs. To address this, this paper proposes DC-FGLLM, integrating graph structure analysis and LLM. Its dual-channel framework uses an improved graph attention network (adaptive adjacency matrix, multi-parallel attention) to convert sequential financial data to graph data and capture spatial couplings, while the GPT-based LLM aligns text-process data to extract semantics and long-distance temporal dependencies. Experiments show DC-FGLLM outperforms mainstream models, offering an efficient university financial risk solution and a time-series prediction reference with theoretical and practical value.
Authors: Haiyang Liu, Zeyu Yang, Hua Li, Ran Wang
Institutions: Qufu Normal University, Binzhou Polytechnic, Jining University
Language: English
Subfield: Accounting
Domain: Social Sciences
Open Access status: hybrid
OpenAlex ID: W7124754841