CN-122022466-A - Enterprise operation risk assessment system based on multi-source heterogeneous data fusion

CN122022466ACN 122022466 ACN122022466 ACN 122022466ACN-122022466-A

Abstract

The invention relates to the technical field of enterprise risk management and big data analysis, in particular to an enterprise management risk assessment system based on multi-source heterogeneous data fusion, which comprises a data acquisition module, a data fusion characteristic engineering module, a risk assessment module and an early warning visualization module; according to the invention, the multi-source heterogeneous data related to the enterprise are obtained, the multi-source heterogeneous data are subjected to feature coding to generate the numerical feature vector, the graph structure feature vector and the semantic risk feature vector, and then the comprehensive modeling is performed through cross-mode feature depth fusion and the attention mechanism to output the management risk probability value, so that the risk level of enterprise management is determined, and the enterprise risk state is intuitively presented in a color identification mode, so that the comprehensive, accurate and visual display of enterprise management risk assessment is realized.

Inventors

LIU YANG
SONG ZIYU
GUO CHUNYANG

Assignees

全联征信有限公司

Dates

Publication Date: 20260512
Application Date: 20260128

Claims (10)

1. The utility model provides an enterprise management risk assessment system based on heterogeneous data of multisource fuses, includes data acquisition module for gather the relevant heterogeneous data of multisource of enterprise, multisource heterogeneous data includes structured data, semi-structured data and unstructured data, its characterized in that still includes: the data fusion feature engineering module is used for carrying out feature coding analysis processing on the multi-source heterogeneous data so as to determine a numerical feature vector, a graph structure feature vector and a semantic risk feature vector; The risk assessment module is used for comprehensively assessing the enterprise management risk based on the numerical feature vector, the graph structure feature vector and the semantic risk feature vector so as to determine a management risk probability value; and the early warning visualization module is used for carrying out grading judgment on the enterprise management risk based on the management risk probability value so as to determine the management risk level and carrying out early warning display in a visualization mode.
2. The system for evaluating risk of enterprise business based on multi-source heterogeneous data fusion according to claim 1, wherein the numerical feature vector is determined by performing a fusion of a basic numerical feature and a derivative numerical feature.
3. The enterprise business risk assessment system based on multi-source heterogeneous data fusion of claim 2, wherein the process of solving and analyzing the basic numerical features and the derivative numerical features is as follows: after field analysis and time alignment processing are carried out on the structured data, deletion and abnormality detection are carried out on various financial indexes in the structured data, and when the deletion data of various financial indexes are detected, the deletion data are corrected by adopting a mean value filling mode based on a historical sliding window; when abnormal fluctuation of various financial indexes is detected, identifying and judging abnormal data through a preset standard deviation threshold value, and cutting off the data identified as abnormal; after the deletion correction and the exception processing are completed, normalization processing is carried out on various processed financial indexes to obtain basic numerical characteristics; Constructing derivative numerical features on the basis of the basic numerical features, wherein the derivative numerical features at least comprise one or more of repayment capability features, profitability features and cash flow stability features; wherein the repayment capability feature is a flow rate, a snap-action rate and an asset liability rate calculated based on the asset liability table data within a preset time window; The profit capability features are gross profit rate, net profit rate and asset profit rate calculated based on profit table data in a preset time window; The cash flow stability characteristics are calculated based on cash flow meter data to obtain an operational cash flow duty ratio and a cash flow fluctuation rate within a preset time window.
4. The enterprise business risk assessment system based on multi-source heterogeneous data fusion according to claim 1, wherein the process of solving and analyzing the feature vector of the graph structure is as follows: establishing an enterprise associated knowledge graph based on semi-structured data, specifically taking an enterprise main body, an advanced manager and a stakeholder as nodes in the graph, and taking investment relations, job-holding relations, control relations and other business association relations as edges in the graph to form a heterogeneous graph structure containing multiple types of nodes and multiple types of relations; meanwhile, in the knowledge graph construction process, a dynamic weighting mechanism is introduced, and weight modeling is carried out on importance of different relation types and different nodes; Performing feature representation learning on the heterogeneous graph structure by adopting a graph neural network, and particularly, distributing different attention weights to different neighbor nodes of the same node through a node level attention mechanism; Meanwhile, through a semantic level attention mechanism, sub-graph representations corresponding to different relation types are subjected to weighted fusion; thus, a higher-order graph structure is obtained and a corresponding graph structure feature vector is output.
5. The enterprise business risk assessment system based on multi-source heterogeneous data fusion of claim 1, wherein the solving and analyzing process of the semantic risk feature vector is as follows: After text denoising, clause, word segmentation and irrelevant symbol cleaning processing are completed on unstructured data, semantic coding is carried out on the unstructured data by adopting a pre-trained Chinese BERT semantic coding model, and a high-dimensional semantic vector is obtained; On the basis of a high-dimensional semantic vector, a risk tendency classifier is introduced to judge the risk tendency of the text information, so that a positive risk tendency, a neutral risk tendency and a negative risk tendency are obtained, and corresponding probability values are respectively output through a Softmax function; The risk tendency classifier takes a high-dimensional semantic vector as an input characteristic, and performs risk tendency discrimination through a classification layer; and taking probabilities corresponding to the positive risk tendency, the neutral risk tendency and the negative risk tendency as semantic risk characteristics to quantitatively represent, and obtaining a semantic risk characteristic vector.
6. The enterprise business risk assessment system based on multi-source heterogeneous data fusion according to claim 1, wherein the solving and analyzing process of the business risk probability value is as follows: The method comprises the steps of calling a numerical feature vector, a graph structure feature vector and a semantic risk feature vector of an enterprise, and projecting the numerical feature vector, the graph structure feature vector and the semantic risk feature vector to a unified feature space in a linear mapping mode to obtain corresponding modal features; After the unified feature mapping is completed, combining the three types of modal features into a cross-modal feature sequence according to a preset sequence; in the depth fusion layer, a cross-modal feature sequence is subjected to fusion modeling by adopting a cross-modal attention mechanism, and corresponding attention weights are distributed to different modal features to obtain a depth fusion feature direction; And then, the depth fusion feature vector is input into a risk prediction network to perform risk probability calculation, so as to obtain an operation risk probability value corresponding to the enterprise.
7. The enterprise business risk assessment system based on multi-source heterogeneous data fusion according to claim 1, wherein the process of determining and analyzing the business risk level is as follows: And matching the management risk probability value with the management risk state judging table to determine corresponding management risk levels, wherein the management risk levels comprise a low risk level, a medium risk level, a higher risk level and a high risk level.
8. The enterprise business risk assessment system based on multi-source heterogeneous data fusion according to claim 7, wherein for the determined business risk level, a color identification mode corresponding to each business risk level one by one is adopted for visual presentation, and the corresponding color is used as the risk identification color of the business risk level.
9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the system of any one of claims 1 to 8 when executing the computer program.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the system of any one of claims 1 to 8.

Description

Enterprise operation risk assessment system based on multi-source heterogeneous data fusion Technical Field The invention relates to the technical field of enterprise risk management and big data analysis, in particular to an enterprise management risk assessment system based on multi-source heterogeneous data fusion. Background The enterprise management risk assessment is an important link for decision making of financial institutions, supervision departments and enterprises, has important significance for guaranteeing healthy development of the enterprises and reducing potential loss, and the traditional risk assessment method is mainly based on a statistical model of a single data set and judges the risk condition of the enterprises, however, the prior art still has significant defects in practical application and mainly comprises the following aspects: firstly, the data source is single and the evaluation is lagged, the existing method mainly depends on structured data such as financial reports and the like issued by enterprises at regular intervals, has obvious lagging, and is difficult to reflect the influence of market change or emergency in time, so that the instant early warning capability of risk evaluation in a fast-evolving environment is limited; Secondly, the heterogeneous data fusion capability is insufficient, the traditional model is difficult to effectively integrate unstructured data (such as news reports, social media public opinion), semi-structured data (such as lawsuit information, equity structure and associated transactions) and structured financial data, so that the risk assessment result is single in dimension, and potential non-financial risks of enterprises such as compliance risks and reputation risks are easily ignored; And meanwhile, under the market environment of emerging industries or complex and changeable, the generalization capability of the model is limited, the evaluation accuracy is easy to be obviously reduced, and the requirements of the risk management of diversified enterprises are difficult to be met. In order to solve the above-mentioned defect, a technical scheme is provided. Disclosure of Invention Aiming at the defects existing in the prior art, the invention provides an enterprise management risk assessment system based on multi-source heterogeneous data fusion, which can effectively solve the problems of data lag, insufficient heterogeneous data fusion capability and limited model robustness and accuracy existing in the prior art. In order to achieve the above object, the present invention can be achieved by the following technical scheme: The invention provides an enterprise management risk assessment system based on multi-source heterogeneous data fusion, which comprises the following steps: the data acquisition module is used for acquiring multi-source heterogeneous data related to enterprises, wherein the multi-source heterogeneous data comprises structured data, semi-structured data and unstructured data; the data fusion feature engineering module is used for carrying out feature coding analysis processing on the multi-source heterogeneous data so as to determine a numerical feature vector, a graph structure feature vector and a semantic risk feature vector; The risk assessment module is used for comprehensively assessing the enterprise management risk based on the numerical feature vector, the graph structure feature vector and the semantic risk feature vector so as to determine a management risk probability value; and the early warning visualization module is used for carrying out grading judgment on the enterprise management risk based on the management risk probability value so as to determine the management risk level and carrying out early warning display in a visualization mode. Further, the numerical feature vector is determined by splicing and fusing the basic numerical feature and the derivative numerical feature. Further, the process of solving and analyzing the basic numerical characteristics and the derivative numerical characteristics is as follows: after field analysis and time alignment processing are carried out on the structured data, deletion and abnormality detection are carried out on various financial indexes in the structured data, and when the deletion data of various financial indexes are detected, the deletion data are corrected by adopting a mean value filling mode based on a historical sliding window; when abnormal fluctuation of various financial indexes is detected, identifying and judging abnormal data through a preset standard deviation threshold value, and cutting off the data identified as abnormal; after the deletion correction and the exception processing are completed, normalization processing is carried out on various processed financial indexes to obtain basic numerical characteristics; Constructing derivative numerical features on the basis of the basic numerical features, wherein the derivative numerical featur