网络安全与数据治理2026，Vol.45Issue(8)：37-44,8.DOI:10.19358/j.issn.2097-1788.2026.08.005
混合计算架构与知识库增强的资源受限财务报表分析方法
Research on enterprise financial statement analysis method based on hybrid computing architecture and knowledge base augmentation under resource constraints
周少磊 1张薷倩 2王信洁 3吴志成 4龙向阳3
作者信息
- 1. 国家国防科技工业局信息中心,北京 100039||北京理工大学 计算机学院,北京 100081
- 2. 澳门大学 协同创新研究院,澳门 999078
- 3. 国家国防科技工业局信息中心,北京 100039
- 4. 北京京航计算通讯研究所,北京 100074
- 折叠
摘要
Abstract
Enterprise financial statement analysis is traditionally characterized by a heavy reliance on manual labor and low efficiency.Further-more,small-parameter Large Language Models(LLMs)deployed in resource-constrained environments often suffer from numerical hallucina-tions,knowledge deficits,and cross-chapter semantic confusion.To address these challenges,this paper proposes a novel method for enterprise financial statement analysis based on hybrid computing architecture and knowledge base augmentation.This approach decouples numerical indi-cator calculation from LLM reasoning,employing structured query language to execute deterministic computations.Furthermore,this paper constructs an"indicator-industry-rule"knowledge base,parameterizing threshold,trend,and comparison rules into anomaly detection func-tions.Additionally,a"chapter-wise reasoning with content isolation"mechanism is designed,establishing exclusive prompts for individual sub-chapters to maintain context integrity.Experimental results demonstrate that this method effectively resolves issues regarding numerical ac-curacy and professional reliability in financial analysis under resource constraints,which offers a viable solution for lightweight financial state-ment analysis.关键词
大语言模型/知识库/财务报表/混合计算架构Key words
large language model/knowledge base/financial statement/hybrid computing architecture分类
信息技术与安全科学引用本文复制引用
周少磊,张薷倩,王信洁,吴志成,龙向阳..混合计算架构与知识库增强的资源受限财务报表分析方法[J].网络安全与数据治理,2026,45(8):37-44,8.
