Enterprise intelligent financial risk prediction method fused with big data analysis

Abstract

The invention relates to the technical field of enterprise risk management, in particular to an enterprise intelligent financial risk prediction method fused with big data analysis, comprising: collecting structured financial data and unstructured text data of a target enterprise, the text data comprising a supply chain contract, a supervision punishment announcement and an industry policy document; generating a supply chain default risk signal and a policy sensitivity risk signal through entity relationship extraction; inputting the structured data and the risk signal into a time sequence diagram neural network, and outputting a dynamic risk probability value and a risk conduction node sequence; constructing an industrial chain risk conduction map according to the result, setting a dynamic risk probability value double threshold to trigger a grading early warning signal, and further pushing a key intervention path and risk control suggestion; the method has the multi-modal risk modeling capability and a map-level response mechanism, and can be widely applied to enterprise credit assessment, supply chain financial risk control and macroscopic supervision scenes.

Description

Technical Field

[0001] This invention relates to the field of enterprise risk management technology, and in particular to an intelligent financial risk prediction method for enterprises that integrates big data analysis. Background Technology

[0002] As the business environment becomes increasingly complex and supply chain collaboration deepens, financial risk is no longer solely a result of internal financial indicator fluctuations. It is increasingly influenced by unstructured factors such as external supply chain relationships, policy guidance, and regulatory compliance. Supported by big data technology, a growing body of research is attempting to integrate structured financial data with unstructured text data for comprehensive assessment of corporate risk. Simultaneously, novel deep learning methods such as graph neural networks offer new possibilities for corporate relationship modeling and risk propagation path mining, laying the theoretical and technological foundation for dynamic risk prediction and intelligent risk control decision-making.

[0003] However, existing technologies still have significant shortcomings in the following aspects: First, their ability to integrate multi-source heterogeneous data is limited. Most methods rely solely on structured financial data and cannot effectively mine potential risk signals in unstructured texts such as supply chain contracts, regulatory penalty announcements, and industry policy documents. Second, risk prediction models mostly adopt static scoring or rule-driven approaches, lacking modeling of risk transmission mechanisms in the industrial chain and failing to reflect the dynamic evolution of risks. Third, there is a lack of mechanisms to structure prediction results into visualized knowledge graphs and actionable early warning paths, making it difficult to support real-time intervention and decision-making by financial institutions, regulatory authorities, and enterprises. Summary of the Invention

[0004] Based on the above objectives, this invention provides an intelligent financial risk prediction method for enterprises that integrates big data analysis. It provides an intelligent financial risk prediction method that integrates multimodal data, has dynamic transmission modeling capabilities, and can generate a highly operational risk knowledge graph.

[0005] A method for predicting enterprise financial risks by integrating big data analytics includes the following steps:

[0006] S1, Data Acquisition: Collect structured financial data and unstructured text data of the target enterprise. The unstructured text data includes supply chain contracts, regulatory penalty announcements, and industry policy documents.

[0007] S2, Risk Signal Extraction: Extract entity relationships from the unstructured text data in S1 to generate supply chain default risk signals and policy sensitivity risk signals;

[0008] S3, Multimodal Risk Dynamic Fusion Prediction: The structured financial data of S1 and the supply chain default risk signal and policy sensitivity risk signal of S2 are input into the time series graph neural network, and the dynamic risk probability value and risk transmission node sequence are output.

[0009] S4, Risk Transmission Knowledge Graph Construction: Using the dynamic risk probability value as a node attribute and the risk transmission node sequence as an edge connection relationship, construct the industrial chain risk transmission graph, and trigger graded early warning signals based on the dynamic risk probability value threshold.

[0010] Optionally, S1 includes:

[0011] S11, Multi-source Enterprise Data Acquisition: This involves collecting structured financial data and unstructured text data from target enterprises via API interfaces, including:

[0012] The structured financial data includes a balance sheet, a cash flow statement, and an income statement;

[0013] The unstructured text data includes supply chain contracts, regulatory penalty announcements, and industry policy documents;

[0014] S12, Cross-modal data cleaning and alignment: Missing value filling and outlier correction are performed on the structured financial data collected in S11, while the encoding format is standardized and noise is removed from the unstructured text data to generate cleaned structured financial data and cleaned unstructured text data.

[0015] S13, Data Pool Construction: The cleaned structured financial data and cleaned unstructured text data generated in S12 are associated and stored according to the enterprise entity ID and timestamp to form a multi-source heterogeneous enterprise data pool with spatiotemporal tags.

[0016] Optionally, S2 includes:

[0017] S21, Entity Relationship Extraction from Multi-Type Text: Entity relations are extracted from the unstructured text data in the multi-source heterogeneous enterprise data pool constructed in S1, where:

[0018] Extract payment terms entities and breach of contract liability entities from supply chain contracts to generate a set of contract payment terms entities and a set of contract breach of contract liability entities;

[0019] Extract the entities of penalty type and penalty intensity from the regulatory penalty announcements to generate a set of entities of regulatory penalty type and a set of entities of regulatory penalty intensity.

[0020] Extract policy keyword entities and regulation direction entities from industry policy documents to generate a policy keyword entity set and a policy regulation direction entity set;

[0021] S22, Risk Signal Quantization Calculation: Based on the output of S21, risk signal quantization calculation is performed, including:

[0022] The correlation between the entity set of contract payment terms and the entity set of contract breach liability is calculated to generate a supply chain default risk signal.

[0023] Semantic conflict analysis is performed on the entity set of regulatory penalty types, the entity set of regulatory penalty intensity, and the entity set of policy keywords to generate policy sensitivity risk signals.

[0024] S23, Dynamic Weighted Signal Fusion: The supply chain default risk signal and policy sensitivity risk signal generated by S22 are weighted according to timeliness, and the enhanced supply chain default risk signal and policy sensitivity risk signal are output.

[0025] Optionally, S3 includes:

[0026] S31, Heterogeneous Data Vectorization Mapping: The structured financial data of S1 is converted into financial feature vectors, and the supply chain default risk signal and policy sensitivity risk signal of S2 are mapped into default signal vectors and policy signal vectors respectively, generating a multimodal risk vector set;

[0027] S32, Dynamic Propagation of Temporal Graph Neural Network: Construct an initial graph structure with enterprises as nodes and supply chain relationships as edges, input multimodal risk vector groups into the temporal graph neural network, dynamically update node states and edge weights through a time sliding window, and output node risk state matrix and edge propagation strength sequence;

[0028] S33, Risk Quantification and Transmission Path Generation: Based on the node risk state matrix of S32, calculate the dynamic risk probability value of each enterprise node, and extract the cross-enterprise risk propagation path according to the edge transmission strength sequence, and generate the risk transmission node sequence in descending order of transmission probability.

[0029] Optionally, S4 includes:

[0030] S41, Node Attribute and Edge Relationship Mapping: Maps the dynamic risk probability value output by S3 to the node attributes of enterprise nodes, and at the same time transforms the risk transmission node sequence into edge connection relationships, generating a set of attributed nodes and edge relationship topology.

[0031] S42, Construction of the Supply Chain Risk Transmission Graph: Based on the attributed node set and edge relationship topology structure of S41, a force-directed layout algorithm is used to generate the supply chain risk transmission graph, where:

[0032] Node size is positively correlated with dynamic risk probability value;

[0033] Edge thickness is positively correlated with the intensity of risk transmission;

[0034] S43, Graded Early Warning and Intervention Path Generation: In the constructed industrial chain risk transmission map, if the dynamic risk probability value of any node exceeds the preset threshold, a risk early warning signal of the corresponding level is triggered, and a key intervention path from the high-risk source to the target enterprise is generated based on the risk transmission node sequence.

[0035] Optionally, in S43, a dual-threshold triggering mechanism for dynamic risk probability values ​​is set. If the dynamic risk probability value of an enterprise node satisfies: 0.6 ≤ dynamic risk probability value < 0.8, a yellow warning signal is triggered. If the dynamic risk probability value of an enterprise node satisfies: dynamic risk probability value ≥ 0.8, a red warning signal is triggered.

[0036] Optionally, when a red warning signal is triggered in S43, a red warning response strategy is executed, including:

[0037] The top three enterprise nodes with the highest dynamic risk probability values ​​are identified as high-risk sources from key intervention paths;

[0038] The system will push a recommendation to freeze credit lines for the high-risk sources.

[0039] Optionally, after receiving a credit limit freeze suggestion, the supply chain finance system executes a step-by-step risk control strategy based on the correlation of the frozen nodes and the historical default probability, and generates a risk response record table for auditing and tracing.

[0040] Optionally, the time-series graph neural network adopts a time-sliding window mechanism based on gated recurrent units to achieve dynamic updates of enterprise node states and improve the response capability to sudden risk events.

[0041] The beneficial effects of this invention are:

[0042] This invention, through the construction of a multi-source heterogeneous enterprise data pool with spatiotemporal tags, for the first time deeply integrates enterprise financial indicators with unstructured text data from supply chain contracts, regulatory penalty announcements, and industry policy documents. Furthermore, by employing entity relationship extraction and semantic quantification mechanisms, it extracts and merges supply chain default risk signals and policy sensitivity risk signals, significantly expanding the coverage of external risk factors by traditional financial models. This solves the problem of blind spots in risk identification caused by relying solely on financial statements in existing technologies, and improves the semantic richness and upstream / downstream interpretability of risk identification.

[0043] This invention introduces a time-series graph neural network to vectorize financial features and textual signals to construct a multimodal risk vector set, and then models the dynamic propagation of risk status on the enterprise supply chain network. This mechanism updates the enterprise node status through a time-sliding window, outputting the enterprise's dynamic risk probability value and risk transmission node sequence. This effectively captures the evolution trend and transmission path of risk in the supply chain, improving the response speed to sudden risks and the foresight and sensitivity of predictions, thus overcoming the limitations of static risk scoring models in terms of timeliness and linkage modeling.

[0044] This invention maps prediction results to enterprise node attributes and risk transmission relationships, employs a force-directed layout algorithm to generate a visually expressive supply chain risk transmission map, and distinguishes between red and yellow warning levels through a dual-threshold triggering mechanism with dynamic risk probability values. Simultaneously, it generates key intervention paths based on the backtracking of risk transmission node sequences, providing credit freeze suggestions and risk response records to the supply chain finance system. This solution not only achieves visualized insight into risk status but also transforms risk prediction results into actionable management actions, enhancing the overall system's risk governance capabilities and automation level. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention;

[0047] Figure 2 This is a schematic diagram of the S3 process in an embodiment of the present invention. Detailed Implementation

[0048] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should also be noted that, to make the embodiments more comprehensive, the following embodiments are the best and preferred embodiments, and those skilled in the art can use other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0049] It should be noted that the use of terms such as "an embodiment," "an embodiment," "an exemplary embodiment," and "some embodiments" in the specification indicates that the described embodiment may include a specific feature, structure, or characteristic, but not every embodiment necessarily includes that specific feature, structure, or characteristic. Furthermore, when a specific feature, structure, or characteristic is described in connection with an embodiment, implementing such a feature, structure, or characteristic in conjunction with other embodiments (whether explicitly described or not) should be within the knowledge of those skilled in the art.

[0050] Generally, terms can be understood at least partly from their use in context. For example, depending at least partly on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in a singular sense, or a combination of features, structures, or characteristics in a plural sense. Additionally, the term "based on" can be understood not necessarily to convey an exclusive set of factors, but rather, alternatively, depending at least partly on the context, to allow for the presence of other factors that are not necessarily explicitly described.

[0051] like Figures 1-2 As shown, the enterprise intelligent financial risk prediction method integrating big data analysis includes the following steps:

[0052] S1, Data Acquisition: Collect structured financial data and unstructured text data of the target company. Unstructured text data includes supply chain contracts, regulatory penalty announcements, and industry policy documents.

[0053] S2, Risk Signal Extraction: Extract entity relationships from the unstructured text data in S1 to generate supply chain default risk signals and policy sensitivity risk signals;

[0054] S3, Multimodal Risk Dynamic Fusion Prediction: The structured financial data of S1 and the supply chain default risk signal and policy sensitivity risk signal of S2 are input into the time series graph neural network, and the dynamic risk probability value and risk transmission node sequence are output.

[0055] S4, Risk Transmission Knowledge Graph Construction: Using dynamic risk probability values ​​as node attributes and risk transmission node sequences as edge connections, a risk transmission graph of the industrial chain is constructed, and graded early warning signals are triggered based on dynamic risk probability value thresholds.

[0056] S11, Multi-source Enterprise Data Acquisition: First, collect structured financial data and unstructured text data of the target enterprise through API interface.

[0057] Structured financial data originates from corporate annual reports, audit reports, and publicly available financial statement interfaces, and mainly includes:

[0058] Balance sheet (including indicators such as accounts receivable, inventory, and current liabilities);

[0059] Cash flow statement (e.g., net cash flow from operating activities, cash flow from investing activities);

[0060] Income statement (e.g., operating revenue, operating costs, net profit).

[0061] Unstructured text data comes from the following three sources:

[0062] Supply chain contracts (obtained through channels such as enterprise information disclosure websites and bidding platforms);

[0063] Regulatory penalty announcements (announcements from authoritative platforms such as the China Securities Regulatory Commission and the State Financial Regulatory Commission);

[0064] Industry policy documents (such as policy guidance texts issued by the Ministry of Industry and Information Technology, the National Development and Reform Commission, and the Ministry of Finance).

[0065] The data acquisition is designed with an asynchronous distributed architecture, supporting timed acquisition and event-triggered update mechanisms.

[0066] Example: Taking a certain company as an example, the system retrieves its publicly available profit statement for 2024, with the "Net Profit" field showing ¥12,300,000. At the same time, it retrieves an environmental regulatory document related to the company from the National Development and Reform Commission website, with record number ZC2024-HJ-0012.

[0067] S12, Cross-modal data cleaning and alignment: After acquisition, the data exhibits significant differences in structure, format, and accuracy, necessitating cleaning and alignment. This process handles both structured and unstructured data separately.

[0068] Structured data processing method: Missing value fields are filled using a regression model based on industry mean and time window.

[0069]

[0070] in, The field value after being filled. The same field value from the previous time point. This is the average value of this field across its respective industry.

[0071] Outlier detection is achieved using the Z-score method:

[0072] If |Z i If |>3, it is marked as an anomaly and corrected or removed.

[0073] Unstructured text data processing methods: Convert multi-source text to UTF-8 encoding; remove HTML tags, special characters, and template formatting noise; use regular expressions to filter invalid segments (such as footnotes, endnotes, and duplicate document numbers); supplement contract data with punctuation and paragraph structure identification for subsequent relationship extraction.

[0074] Example: The original regulatory announcement stating "...the company was ordered to rectify its violations of the 'Information Disclosure Regulations'..." is standardized to: "Reason for Regulatory Penalty, Type of Violation, Rectification"

[0075] S13, Data Pool Construction: The cleaned structured financial data generated in S12 is merged with unstructured text data. A traceable, multi-source, heterogeneous enterprise data pool is constructed using enterprise entity IDs (such as unified social credit codes) and data occurrence timestamps as a composite index. Each data item records the following fields:

[0076] Company ID;

[0077] Data type (structured or unstructured);

[0078] Data subtypes (balance sheet, supply chain contracts, etc.);

[0079] Timestamp (e.g., 2024-12-31 or document publication date);

[0080] Source file ID (original source index);

[0081] Content fields (or sets of relational entities).

[0082] S2 includes:

[0083] S21, Multi-type text entity relation extraction: Using named entity recognition (NER) and relation extraction models, key semantic units and their relationships are extracted from unstructured text data to generate multiple entity sets, which serve as the semantic basis for subsequent risk signal calculations, including:

[0084] (1) Supply chain contract processing: Use contract intent recognition models (such as RoBERTa-BiLSTM-CRF) to identify payment terms and liability statements for breach of contract;

[0085] Extract payment terms, payment methods, late payment fee rates, etc., to form a set of contract payment terms entities;

[0086] Extract relevant information regarding delayed delivery, non-performance, and liquidated damages to form a set of contractual liability entities.

[0087] Example:

[0088] The original contract text states: "If Party B fails to make payment within 30 days, Party A has the right to terminate the contract and claim compensation."

[0089] →Extracted Entity: Payment Period = 30 days, Liability for Breach of Contract = Contract Termination + Claim;

[0090] (2) Handling of regulatory penalty announcements:

[0091] Based on keyword triggering rules and deep learning models (such as BERT + rule fusion), extract the core description of the penalty event;

[0092] Identify entities such as "order to rectify", "fine", and "revocation of license" to form a set of regulatory penalty types;

[0093] Extract fields such as "fine amount", "regulatory period", and "scope of business affected" to form an entity set of regulatory penalty intensity;

[0094] Example:

[0095] The original announcement stated: "The company was fined 5 million yuan for making false disclosures."

[0096] →Extracted Entity: Penalty Type = False Disclosure, Penalty Amount = 5 million yuan;

[0097] (3) Processing of industry policy documents:

[0098] Perform part-of-speech tagging and dependency parsing on the main text, and extract policy-oriented terms using a policy dictionary;

[0099] Extract keywords such as "accelerate elimination", "key support", and "strict restriction" to generate a set of policy keyword entities;

[0100] Analyze the industry actions (support / restriction) behind the policies to form a set of entities that reflect the policy control direction;

[0101] Example:

[0102] The document states: "Accelerate the elimination of high-energy-consuming enterprises."

[0103] →Extracting entities: Policy keywords = high-energy-consuming enterprises, regulation direction = elimination (restriction)

[0104] All entity relationships are uniformly formatted as 5-tuples:

[0105] Tuple = <source text segment, entity type, entity value, context weight, timestamp>;

[0106] S22, Risk Signal Quantification Calculation: After entity extraction, each entity set is transformed into quantifiable risk indicators. The core objective is to compress semantic information into two key risk signals—supply chain default risk signals and policy sensitivity risk signals.

[0107] (1) Calculation of supply chain default risk signals:

[0108] Construct a relationship graph between the entity set of contract payment terms and the entity set of contract breach liability;

[0109] Using the weighted matching score function R 违约 Quantification:

[0110] in, and w represents the entity of the i-th pair of contract terms and breach of contract clauses. i The weight of this pair of clauses is δ, which is the entity matching function (1 for a match, 0 for no match);

[0111] If a company is identified as having ≥3 high-risk default combinations in the past 6 months, its supply chain default risk signal is determined to be strong.

[0112] (2) Calculation of policy sensitivity risk signals:

[0113] Calculate the following two types of semantic conflict indices in sequence:

[0114] Policy Entity Conflict Index C p : Reflects the degree of policy direction conflict between entities with regulatory penalty types and entities with policy keywords;

[0115] Penalty Intensity Modification Item S p The scalar normalized value is based on the severity of the entity's violation of regulatory penalties; the comprehensive calculation formula is as follows:

[0116] R 政策 =γ·C p +(1-γ)·S p ;

[0117] Where γ∈[0,1] is the empirical weight parameter (e.g., 0.6), C p S is obtained through nested semantic comparison calculations (e.g., the policy keyword "phase-out" and the penalty type "illegal emissions" constitute a strong conflict). p The penalties will be processed uniformly based on the amount of the penalty and the processing period.

[0118] S23, Dynamically Weighted Signal Fusion: Since policy and contractual events have varying impacts on enterprises at different times, timeliness adjustments are necessary to enhance the dynamic expressiveness of the signals. Let the signal time decay weighting function be:

[0119]

[0120] Among them, T now T represents the current time. eventWhere λ is the signal occurrence time, and λ is the attenuation coefficient (e.g., 0.05 / month). The final fused output risk signal is:

[0121]

[0122] Example:

[0123] Company A signed a contract with a downstream client in January 2024 with lenient payment terms and vague liability for breach of contract.

[0124] In March of the same year, it was fined 800,000 yuan by the Environmental Protection Bureau. In May of the same year, industry policy proposed to "restrict companies that fail to meet emission standards from participating in bidding."

[0125] →After the above quantification and integration, Company A's overall risk signal increased significantly in June 2024.

[0126] S3 includes:

[0127] S31, Heterogeneous Data Vectorization Mapping: The structured financial data from S1 and the supply chain default risk signals and policy sensitivity risk signals from S2 are uniformly encoded into numerical vectors to form a multimodal risk vector set.

[0128] (1) Transforming structured financial data into financial feature vectors:

[0129] Core financial indicators (such as current ratio, gross profit margin, net operating cash flow, and inventory turnover) are extracted from the balance sheet, income statement, and cash flow statement to form a financial vector F. To enhance the dynamic expression over time, time window difference vectors are used to represent continuous quarterly data.

[0130] ΔF t =F t -F t-1 ;

[0131] (2) Vectorization of supply chain default risk signals:

[0132] The risk scoring model maps supply chain default risk signals into a default signal vector V. 违约 Each dimension corresponds to a different contract risk factor (such as the frequency of payment defaults, the risk intensity of contract terms, etc.);

[0133] Example vector structure is as follows:

[0134] V 违约 = [0.8, 0.6, 0.2];

[0135] (3) Vectorization of policy sensitivity risk signals:

[0136] Policy keywords and penalty events are used to construct a policy risk space. A semantic distance and penalty weight vector V are generated by calculating the semantic distance and penalty weight using a word embedding model (such as BERT or Word2Vec). 政策 ;

[0137] Example: If "emission restrictions + environmental penalties" exist, then the higher value of the relevant dimension will be used;

[0138] (4) Construction of multimodal risk vector sets:

[0139] Finally, the above three types of vectors are concatenated to form a unified enterprise input feature:

[0140] X t =concat(ΔF) t V 违约 V 政策 );

[0141] S32, Dynamic Propagation of Temporal Graph Neural Network: Construct a risk transmission graph between enterprises and propagate node states based on a temporal graph neural network to capture the dynamic evolution of risks in supply chain relationships;

[0142] Initial graph structure construction:

[0143] In the graph G = (V, E), node V represents a business entity, and edge E represents a supply chain relationship (such as contract transactions, invoice associations, etc.).

[0144] Each edge e ij ∈E has static weights Defined based on transaction frequency, contract amount, etc.;

[0145] (2) Temporal propagation mechanism: Using a temporal graph neural network, the node state is updated within a sliding time window [t-τ,t]. The node state update formula is as follows:

[0146]

[0147] in, Let W be the set of neighboring nodes of node i, and let W,b be the parameters of the graph neural network. It is a dynamic edge weight.

[0148] (3) Output results: The output contains two core tensors: the node risk state matrix and the edge propagation strength sequence.

[0149] Node risk state matrix:

[0150] Edge conduction intensity sequence: for each edge e ij Output its time series weights

[0151] S33, Risk Quantification and Transmission Path Generation: Based on the output of S32, perform quantitative calculations to obtain the dynamic risk probability value of each enterprise node and identify the risk transmission node sequence across nodes.

[0152] (1) Calculation of dynamic risk probability value: The Sigmoid mapping function is used to map the final hidden state of a node to a probability value, expressed as:

[0153]

[0154] Among them, w p ,b p These are the parameters for risk prediction mapping.

[0155] (2) Risk propagation path extraction: The propagation probability is calculated based on the edge propagation strength sequence, and is expressed as:

[0156]

[0157] Starting from each high-risk node, perform a breadth-first search (BFS) based on the propagation probability. Arrange them from high to low to extract the risk transmission chain.

[0158] (3) Generate a risk transmission node sequence: form an ordered sequence:

[0159] Risk transmission node sequence = {v r 1,v r2 ,…,v rk};

[0160] The nodes are sorted by weighted average of the number of times they act as relay nodes in the transmission path and the transmission strength.

[0161] Example explanation:

[0162] Suppose that company A has a high supply chain default risk signal and a negative cash flow trend, company B has a procurement contract with company A, and company C has a raw material supply relationship with company B. After propagation through a time-series graph neural network, the node state of company C is continuously enhanced, and its dynamic risk probability value increases from 0.25 to 0.74, and it is included in the final risk transmission node sequence.

[0163] S4 includes:

[0164] S41, Node Attribute and Edge Relationship Mapping: Transforms the prediction results output from S3 into the graph structure data required for graph construction, including:

[0165] (1) Node attribute mapping:

[0166] The dynamic risk probability value of each enterprise node is used as the node attribute P. i (t), and assign the value to the corresponding node in the graph node set V;

[0167] The node attribute structure is illustrated as follows: v i ={Enterprise ID} i ,P i (t) Industry tags, timestamps;

[0168] (2) Generation of edge connection relationships:

[0169] The risk transmission node sequence in S3 is parsed into a set of directed edges E, where each edge e ij This indicates that the risk propagates from node i to node j;

[0170] At the same time, edge attribute weights are introduced to represent the propagation strength. The edge conduction intensity sequence originating from S3.

[0171] (3) Generation of attributed graph structures:

[0172] The final graph triplet is formed in the following format: G = (V, E, A), where A = {node attributes + edge attributes};

[0173] S42, Construction of Supply Chain Risk Transmission Map: Based on the graph structure constructed in S41, a visual layout algorithm is used to generate a supply chain risk transmission map to reflect the risk impact links between enterprises and their strength relationships, including:

[0174] (1) Graph layout algorithm:

[0175] A force-directed layout algorithm is used to calculate the position of each node in the graph, ensuring that high-risk nodes are focused and the transmission path is intuitive.

[0176] The graph layout algorithm simulates the elastic repulsion between nodes and the attraction of edges, and iteratively converges to ensure a reasonable distribution of nodes.

[0177] (2) Visual mapping rules for graphs:

[0178] The size of a node is proportional to its corresponding dynamic risk probability value: NodeSize(v i )=α·P i (t) ;

[0179] Where α is the visualization scaling factor;

[0180] The thickness of the edge and its corresponding risk transmission strength Proportional:

[0181] Where β is the edge weight display scaling factor;

[0182] All atlas information is updated in a rolling time window, supporting dynamic graph evolution display.

[0183] Example:

[0184] If the risk transmission strength from company A to company B is 0.85, and from company B to company C it is 0.62, and the dynamic risk probability values ​​of A, B, and C are 0.91, 0.76, and 0.65 respectively, then node A in the graph is the largest, and the edge from A to B is the thickest, indicating that the main source of risk is company A.

[0185] S43, Generation of Hierarchical Early Warning and Intervention Paths: Based on the node attributes and structure in the graph, intelligent early warning and intervention path calculation is performed, including:

[0186] (1) Dynamic risk probability value dual threshold triggering mechanism:

[0187] Two risk level thresholds are set: when 0.6 ≤ P i (t) When P < 0.8, it is marked as a yellow warning signal. i (t) A red alert signal is triggered when the value is ≥0.8.

[0188] It can be formalized as:

[0189]

[0190] (2) Key intervention path generation: In the graph, starting from the red warning node, trace back the upstream propagation path along the risk transmission node sequence, extract the dynamic risk probability value of the nodes in the path, and select the top three enterprise nodes with the highest risk values, marking them as the high-risk source node set S. 高风 ;

[0191] Example: If the path is and Therefore, D, B, and C are selected as high-risk sources.

[0192] (3) Red Alert Response Strategy Execution: Send the following instructions to the supply chain finance system:

[0193] Indicate the ID of the high-risk source node;

[0194] Includes its dynamic risk probability value and critical path ID sequence;

[0195] It is recommended to freeze their credit limit or adjust their risk rating.

[0196] (4) Risk intervention record generation: An intervention record table including the following fields is generated synchronously for the audit and compliance system to trace:

[0197] Risk node ID, level, warning trigger time, intervention suggestion type (freeze / prudent lending), and corresponding path.

[0198] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0199] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting enterprise intelligent financial risks by integrating big data analytics, characterized in that: Includes the following steps: S1, Data Acquisition: Collect structured financial data and unstructured text data of the target enterprise. The unstructured text data includes supply chain contracts, regulatory penalty announcements, and industry policy documents. S2, Risk Signal Extraction: Extract entity relationships from the unstructured text data in S1 to generate supply chain default risk signals and policy sensitivity risk signals; S3, Multimodal Risk Dynamic Fusion Prediction: The structured financial data of S1 and the supply chain default risk signal and policy sensitivity risk signal of S2 are input into the time series graph neural network, and the dynamic risk probability value and risk transmission node sequence are output. S4, Risk Transmission Knowledge Graph Construction: Using the dynamic risk probability value as a node attribute and the risk transmission node sequence as an edge connection relationship, construct the industrial chain risk transmission graph, and trigger graded early warning signals based on the dynamic risk probability value threshold.

2. The enterprise intelligent financial risk prediction method integrating big data analysis according to claim 1, characterized in that, S1 includes: S11, Multi-source Enterprise Data Acquisition: This involves collecting structured financial data and unstructured text data from target enterprises via API interfaces, including: The structured financial data includes a balance sheet, a cash flow statement, and an income statement; The unstructured text data includes supply chain contracts, regulatory penalty announcements, and industry policy documents; S12, Cross-modal data cleaning and alignment: Missing value filling and outlier correction are performed on the structured financial data collected in S11, while the encoding format is standardized and noise is removed from the unstructured text data to generate cleaned structured financial data and cleaned unstructured text data. S13, Data Pool Construction: The cleaned structured financial data and cleaned unstructured text data generated in S12 are associated and stored according to the enterprise entity ID and timestamp to form a multi-source heterogeneous enterprise data pool with spatiotemporal tags.

3. The enterprise intelligent financial risk prediction method integrating big data analysis according to claim 2, characterized in that, S2 includes: S21, Entity Relationship Extraction from Multi-Type Text: Entity relations are extracted from the unstructured text data in the multi-source heterogeneous enterprise data pool constructed in S1, where: Extract payment terms entities and breach of contract liability entities from supply chain contracts to generate a set of contract payment terms entities and a set of contract breach of contract liability entities; Extract the entities of penalty type and penalty intensity from the regulatory penalty announcements to generate a set of entities of regulatory penalty type and a set of entities of regulatory penalty intensity. Extract policy keyword entities and regulation direction entities from industry policy documents to generate a policy keyword entity set and a policy regulation direction entity set; S22, Risk Signal Quantization Calculation: Based on the output of S21, risk signal quantization calculation is performed, including: The correlation between the entity set of contract payment terms and the entity set of contract breach liability is calculated to generate a supply chain default risk signal. Semantic conflict analysis is performed on the entity set of regulatory penalty types, the entity set of regulatory penalty intensity, and the entity set of policy keywords to generate policy sensitivity risk signals. S23, Dynamic Weighted Signal Fusion: The supply chain default risk signal and policy sensitivity risk signal generated by S22 are weighted according to timeliness, and the enhanced supply chain default risk signal and policy sensitivity risk signal are output.

4. The enterprise intelligent financial risk prediction method integrating big data analysis according to claim 3, characterized in that, S3 includes: S31, Heterogeneous Data Vectorization Mapping: The structured financial data of S1 is converted into financial feature vectors, and the supply chain default risk signal and policy sensitivity risk signal of S2 are mapped into default signal vectors and policy signal vectors respectively, generating a multimodal risk vector set; S32, Dynamic Propagation of Temporal Graph Neural Network: Construct an initial graph structure with enterprises as nodes and supply chain relationships as edges, input multimodal risk vector groups into the temporal graph neural network, dynamically update node states and edge weights through a time sliding window, and output node risk state matrix and edge propagation strength sequence; S33, Risk Quantification and Transmission Path Generation: Based on the node risk state matrix of S32, calculate the dynamic risk probability value of each enterprise node, and extract the cross-enterprise risk propagation path according to the edge transmission strength sequence, and generate the risk transmission node sequence in descending order of transmission probability.

5. The enterprise intelligent financial risk prediction method integrating big data analysis according to claim 4, characterized in that, S4 includes: S41, Node Attribute and Edge Relationship Mapping: Maps the dynamic risk probability value output by S3 to the node attributes of enterprise nodes, and at the same time transforms the risk transmission node sequence into edge connection relationships, generating a set of attributed nodes and edge relationship topology. S42, Construction of the Supply Chain Risk Transmission Graph: Based on the attributed node set and edge relationship topology structure of S41, a force-directed layout algorithm is used to generate the supply chain risk transmission graph, where: Node size is positively correlated with dynamic risk probability value; Edge thickness is positively correlated with the intensity of risk transmission; S43, Graded Early Warning and Intervention Path Generation: In the constructed industrial chain risk transmission map, if the dynamic risk probability value of any node exceeds the preset threshold, a risk early warning signal of the corresponding level is triggered, and a key intervention path from the high-risk source to the target enterprise is generated based on the risk transmission node sequence.

6. The enterprise intelligent financial risk prediction method integrating big data analysis according to claim 5, characterized in that, S43 sets a dual threshold triggering mechanism for dynamic risk probability values. If the dynamic risk probability value of an enterprise node meets the following conditions: 0.6 ≤ dynamic risk probability value < 0.8, a yellow warning signal is triggered. If the dynamic risk probability value of an enterprise node meets the following conditions: dynamic risk probability value ≥ 0.8, a red warning signal is triggered.

7. The enterprise intelligent financial risk prediction method integrating big data analysis according to claim 6, characterized in that, When a red warning signal is triggered in S43, a red warning response strategy is executed, including: The top three enterprise nodes with the highest dynamic risk probability values ​​are identified as high-risk sources from key intervention paths; The system will push a recommendation to freeze credit lines for the high-risk sources.

8. The enterprise intelligent financial risk prediction method integrating big data analysis according to claim 7, characterized in that, After receiving a credit limit freeze suggestion, the supply chain finance system executes a step-by-step risk control strategy based on the correlation of the frozen nodes and the historical default probability, and generates a risk response record table for auditing and tracing.

9. The enterprise intelligent financial risk prediction method integrating big data analysis according to claim 8, characterized in that, The time-series graph neural network adopts a time-sliding window mechanism based on gated recurrent units to achieve dynamic updates of enterprise node states and improve the response capability to sudden risk events.

Images