网络舆情引导下企业财务危机深度动态预警研究

Under the situation of network public opinion big data, this project intends to study on dynamic early warning of financial crisis of the listed companies with direction of the network public opinion big data, exploiting the sentiment analysis. This project will deal with two scientific issues, based on multimodal deep learning and multiple kernel learning theory and methods. The first scientific issue is multimodal feature fusion of non-financial indicators under the guidance of network public opinion and financial indicators. The second issue is deeply dynamic warning of financial crisis by considering the updating of both the company samples and the financial crisis features. This project will study on four main research contents. The first is design of financial crisis early warning index system while network public opinion is introduced. This content will study on the design and extraction of the warning index system with the aid of the network public opinion and generate multimodal crisis features. The second is deep fusion of the multimodal crisis features. This content will develop effective fusion method by introducing multimodal deep learning theory. The fusion method is expected to well integrate the crisis features from the network public opinion and the traditional company crisis features. The third is establishment of dynamic early warning model by exploiting incremental multiple kernel learning theory. This content will introduce the incremental multiple kernel learning and propose good mechanism of dynamic choosing company samples. it is expected to well deal with the concept drift and updating the crisis features. The last content is empirical research, analysis and evaluation. This project is expected to construct deep and dynamic early warning model of financial crisis of the listed companies while the network public opinion is well explored. This project will have important theoretical value and practical significance for the normal operation of social and economic security of listed companies.

本项目在网络大数据背景下，通过网络舆评情感分析信息挖掘手段，拟开展网络舆情引导下企业财务危机动态预警研究。针对网络舆情语义特征构成的非财务指标同传统财务指标多模融合、网络舆情引导下财务危机预测动态性两大科学问题，立足深度学习、多核学习理论，主要研究：(1)网络舆情变量引导下危机预警指标体系设计，拟研究网络舆情引导下预警指标体系设计和提取，形成网络舆情引导下财务危机多模特征；(2)网络舆情引导下多模财务危机特征融合，拟基于多模深度学习，研究多模财务危机深度特征融合，有效整合网络舆情类危机特征和财务指标类危机特征；3)网络舆情引导下多核动态预警模型，拟设计有效地动态样本筛选机制，引入增量多核学习，解决动态预测中存在的概念漂移和危机特征更新问题；(4)实证研究与评价分析。本项目预期构建网络舆情自适应融入的企业财务危机动态预警模型，对于保障上市企业正常运行和社会经济稳定有重要的理论价值和实际意义。

本项目在网络大数据背景下，通过网络舆评情感分析信息挖掘手段，拟开展网络舆情引导下企业财务危机动态预警研究。针对网络舆情语义特征构成的非财务指标同传统财务指标多模融合、网络舆情引导下财务危机预测动态性两大科学问题，立足深度学习、多核学习理论，主要研究：1)网络舆情变量引导下危机预警指标体系设计，拟研究网络舆情引导下预警指标体系设计和提取，形成网络舆情引导下财务危机多模特征；2)网络舆情引导下多模财务危机特征融合，拟基于多模深度学习，研究多模财务危机深度特征融合，有效整合网络舆情类危机特征和财务指标类危机特征；3)网络舆情引导下多核动态预警模型，拟设计有效地动态样本筛选机制，引入增量多核学习，解决动态预测中存在的概念漂移和危机特征更新问题；4)实证研究与评价分析。项目研究取得的主要进展包括：1)融合舆情信息的财务危机预警指标设计；2)基于多尺度多核学习的财务危机静态预警模型及方法；3)基于稀疏多核学习的财务危机预测中指标影响力分析；4)基于增量型多核学习的财务危机动态预警模型及方法。整体上，设计了融合舆情信息的企业财务危机预警指标体系，构建了多尺度多核学习网络的静态预测模型及方法，提出了稀疏性约束下的企业财务指标危机影响力分析方法，构建了增量型动态预警模型，进一步分析了舆情数据用于财务危机预警仍存在的问题。在基金项目资助下，课题组在财务困境预测方法、动态预警方面取得了一定研究成果。目前已发表论文4篇，其中1篇发表在《运筹与管理》上，3篇发表在国际SCI期刊上，完成动态预测论文1篇(待投稿)。申请国家发明专利1项(审理中)，完成企业财务困境预测方法专著1部(校稿中)。