CN-122022468-A - Enterprise risk intelligent early warning method and system based on dynamic data fusion and graph reasoning

CN122022468ACN 122022468 ACN122022468 ACN 122022468ACN-122022468-A

Abstract

The embodiment of the invention discloses an enterprise risk intelligent early warning method and system based on dynamic data fusion and graph reasoning. The method comprises the steps of carrying out dynamic trusted fusion on enterprise multidimensional data acquired from a plurality of heterogeneous data sources to generate an enterprise unified fact image, extracting multidimensional risk indexes based on the unified fact image, constructing an enterprise dynamic risk knowledge graph taking enterprise entities, risk indexes, relatives and risk events as nodes, carrying out reasoning and simulation on the graph by utilizing a risk reasoning model based on a graph neural network, identifying risk nodes, evaluating a conducting path and outputting a comprehensive risk score and a tracing path, triggering early warning according to the scores by combining an adaptive threshold value, and generating an interpretable report containing risk causes, conducting visualization and relief strategies based on attribution analysis. The implementation mode can realize accurate, deep and interpretable intelligent early warning of enterprise risks.

Inventors

CHEN FEI
CHENG JIANJUN
SHEN DAWEI
HAN YANAN
GU TAO

Assignees

北京阿尔法风控科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260128

Claims (10)

1. An enterprise risk early warning method based on dynamic data fusion and graph reasoning is characterized by comprising the following steps: Performing dynamic trusted fusion on the enterprise multidimensional data to generate an enterprise unified fact image, wherein the dynamic trusted fusion comprises performing weighted fusion on heterogeneous data based on real-time estimated data source dynamic confidence weights; based on the enterprise unified fact drawing, constructing an enterprise dynamic risk knowledge graph, wherein the graph comprises a network structure taking an enterprise entity as a root node and taking a risk index, an associated party entity and a risk event as associated nodes; Carrying out reasoning analysis on the enterprise dynamic risk knowledge graph by using a risk reasoning model based on a graph neural network so as to generate a comprehensive risk score and a risk tracing path; And carrying out early warning judgment according to the comprehensive risk score and the dynamically generated risk threshold value, and generating an interpretable risk report based on the risk tracing path.
2. The method of claim 1, wherein the dynamically trusted fusing of the enterprise multidimensional data comprises: collecting data from at least two data sources in a government database, an enterprise information system, a public opinion platform and a supply chain platform; Calculating the data source dynamic confidence weight of each data source aiming at the current fusion task based on at least one dimension of timeliness, historical supply stability, cross-source verification consistency rate and self logic conflict rate of the data sources; And carrying out weighted fusion on the same semantic data from different data sources according to the dynamic confidence weights of the data sources.
3. The method of claim 1, wherein the constructing the enterprise dynamic risk knowledge graph comprises: extracting at least two of a financial risk index, an operation risk index, a market risk index, a compliance risk index and a public opinion risk index from the enterprise unified fact image; identifying an associated party of the enterprise, wherein the associated party at least comprises a stakeholder, a provider, a client and a guarantee party; Extracting historical risk events and real-time risk signals of the enterprise; and establishing association edges between the enterprise entity and the risk index, between the association party entity and between the enterprise entity and the risk event by taking the enterprise entity as a root node, wherein the relationship types of the association edges at least comprise stock right control, supply dependence, guarantee association and event involvement.
4. The method of claim 3, wherein the performing the inference analysis using a graph neural network-based risk inference model further comprises: And simulating the conducting process of the risk along the associated side based on the enterprise dynamic risk knowledge graph so as to quantitatively evaluate the impact range and strength of the risk on the associated enterprise.
5. A method according to claim 3, characterized in that the method further comprises: carrying out multi-scale time sequence prediction on the key risk indexes to obtain a prediction result; and supplementing the prediction result as a future risk characteristic node to the enterprise dynamic risk knowledge graph.
6. The method of claim 1, wherein the dynamic risk threshold is generated by a machine learning model whose input features include industry characteristics, macro economic cycle phases, and enterprise historical risk profiles to which the enterprise belongs.
7. The method of claim 1, wherein the generating an interpretable risk report based on the risk tracing path comprises: positioning a risk factor with the highest contribution to the comprehensive risk score and an associated path by using a model interpretability technology; and generating a template based on a predefined natural language, and integrating the risk factors, the association paths and the corresponding suggestions into a structured text and a chart.
8. An enterprise risk intelligent early warning system based on dynamic data fusion and graph reasoning is characterized by comprising: the trusted data fusion center is configured to dynamically and trusted fuse the enterprise multidimensional data to generate an enterprise unified fact figure; The intelligent atlas analysis engine is configured to construct an enterprise dynamic risk knowledge atlas based on the enterprise unified fact image, and perform reasoning analysis on the atlas by utilizing a risk reasoning model based on a graph neural network so as to generate a comprehensive risk score and a risk tracing path; And the interpretable early warning decision center is configured to perform early warning judgment according to the comprehensive risk score and the dynamic risk threshold value and generate an interpretable risk report based on the risk tracing path.
9. The system of claim 8, wherein the system further comprises: And the model evolution feedback module is configured to collect early warning feedback and actual risk evolution data and is used for performing incremental learning and optimization on the risk reasoning model and the dynamic risk threshold model.
10. An electronic device, comprising: one or more processors; a storage device having one or more programs stored thereon; the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method of any of claims 1-7.

Description

Enterprise risk intelligent early warning method and system based on dynamic data fusion and graph reasoning Technical Field The invention relates to the technical field of Internet information processing, enterprise risk management and information, in particular to a method and a system for enterprise risk intelligent early warning based on dynamic data fusion and graph reasoning, which are used for cross application of big data analysis, knowledge graph and artificial intelligent technology in enterprise risk early warning. Background With the acceleration of the enterprise digital process, the internal and external data generated in the operation process of the enterprise show explosive growth, and the data types cover multiple dimensions such as finance, operation, market, public opinion and the like. The traditional enterprise risk early warning method mainly relies on periodic analysis of static financial reports or a rule model based on a single data source, and has obvious limitations. First, the data sources are single and the updating is lagged, so that the real-time and dynamic risk condition of the enterprise is difficult to reflect. Secondly, the existing analysis method focuses on isolated index judgment, lacks systematic consideration on conduction and amplification of risks in enterprise associated networks (such as supply chains and guarantee chains), and cannot early warn the linkage risk of domino. Moreover, most risk early warning systems based on complex machine learning models (such as deep learning) are like 'black boxes', the early warning results of the risk early warning systems lack of interpretability, and business staff are difficult to understand the risk sources, so that decision making is difficult and response is slow. In the prior art, there are some proposals that attempt to improve. For example, some schemes do risk analysis by introducing more data sources, but do not address the problem of trusted fusion when the multi-source data are of varying quality and conflict with each other. Other schemes attempt to express enterprise associations using graph technology, but at the static relationship presentation level, lack the ability to deep reasoning and dynamically simulate risk conduction. Still other schemes focus on improving model prediction accuracy, but neglect the interpretability of the early warning results, making high risk early warning difficult to be signaled by the management layer and converted into effective countermeasures. Therefore, there is a need in the art for an intelligent enterprise risk early warning solution that can integrate multi-source heterogeneous data, deep mining risk association and conduction rules, and can clearly explain the early warning source, so as to improve the accuracy, prospectivity and operability of risk management. Disclosure of Invention The disclosure is in part intended to introduce concepts in a simplified form that are further described below in the detailed description. The disclosure is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used to limit the scope of the claimed subject matter. Some embodiments of the disclosure provide an enterprise risk intelligent early warning method and system based on dynamic data fusion and interpretable graph reasoning, so as to solve the problems of low reliability of data fusion, surfacing of risk correlation analysis, unexplained early warning results and the like in the background art. In a first aspect, some embodiments of the present disclosure provide an enterprise risk early warning method based on dynamic data fusion and graph reasoning, the method including: s1, dynamic trusted fusion is carried out on the enterprise multidimensional data to generate an enterprise unified fact portrait. The dynamic trusted fusion comprises the step of carrying out weighted fusion on heterogeneous data based on the dynamic confidence weight of the data source estimated in real time. S2, constructing an enterprise dynamic risk knowledge graph based on the enterprise unified fact image. The map comprises a network structure taking a business entity as a root node and a risk index, an associated party entity and a risk event as associated nodes. And S3, carrying out reasoning analysis on the enterprise dynamic risk knowledge graph by using a risk reasoning model based on the graph neural network so as to generate a comprehensive risk score and a risk tracing path. And S4, early warning judgment is carried out according to the comprehensive risk score and the dynamically generated risk threshold value, and an interpretable risk report is generated based on the risk tracing path. In a second aspect, some embodiments of the present disclosure provide an enterprise risk intelligent early warning system based on dynamic data fusion and graph reasoning, the system comprising: A trusted data fusion center configured to perform the st