A multi-source data fusion enterprise credit comprehensive prediction method, device and medium
By standardizing and classifying enterprise-related data and integrating static and dynamic characteristics, the problem of risk transmission among related parties in enterprise credit forecasting is solved, enabling dynamic monitoring and early warning of enterprise credit risk.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Filing Date
- 2026-03-12
- Publication Date
- 2026-07-14
Smart Images
Figure CN122390850A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of enterprise credit prediction technology, and in particular to a method, device and medium for comprehensive enterprise credit prediction by multi-source data fusion. Background Technology
[0002] Corporate credit forecasting is a technology that quantitatively assesses a company's future willingness and ability to fulfill its obligations by analyzing various types of information. It is widely used in core risk control areas such as bank lending, bond investment, supply chain finance, and commercial transactions. Existing mainstream technologies are primarily built upon companies' regularly disclosed structured financial reports (such as quarterly and annual statements) and historical credit records. Depending on the data dimensions, model algorithms, and analytical logic used, existing solutions exhibit significant differences in key performance aspects such as forecast timeliness, relevance, and interpretability.
[0003] In existing technologies, the collection of multi-source data typically involves static data, which, due to limitations caused by time lags, struggles to capture risk events occurring at a specific assessment point. For the assessment targets of corporate credit forecasting, existing forecasting models usually treat the target as an independent entity, analyzing only its own credit data, failing to anticipate chain risks arising from the transmission of risks from related parties. Therefore, there is an urgent need for a corporate credit forecasting method that can integrate multi-source real-time data and perform dynamic iteration. Summary of the Invention
[0004] This application provides a method, device, and medium for comprehensive enterprise credit prediction based on multi-source data fusion, which solves the technical problem in the prior art that enterprise credit prediction cannot predict the transmission of risks to related parties.
[0005] In a first aspect, embodiments of this application provide a method for comprehensive corporate credit prediction through multi-source data fusion. The method includes: collecting relevant corporate data from a target company and standardizing the data to obtain a standardized event stream for the target company; wherein the relevant corporate data includes: judicial litigation data, business operation data, public opinion and news data, financial market data, and supply chain data; classifying the standardized event stream by risk identification to obtain the risk intensity corresponding to each type of event, and based on the risk intensity, performing risk correlation analysis on the standardized event stream to determine the dynamic risk characteristics of the target company; calling static fundamental features from the target company's corporate database and fusing the static fundamental features and dynamic risk features to obtain risk fusion features; calculating the corporate default probability corresponding to the risk fusion features and allocating the feature contribution of the risk fusion features to obtain risk attribution data for the target company; and extracting the contribution value from the risk attribution data using natural language to determine the corporate credit prediction text for the target company.
[0006] In one implementation of this application, enterprise-related data is standardized to obtain a standardized event flow for the target enterprise. Specifically, this includes: extracting time data from enterprise-related data and converting the time data into timestamps; extracting case data from judicial litigation data and converting the case data and corresponding timestamps into a coded form to obtain a first standardized event flow; extracting changes to the enterprise entity from business registration data and converting the changes and corresponding timestamps into a coded form to obtain a second standardized event flow; wherein the changes include: changes in legal representative, changes in key personnel, and changes in registered capital; extracting positive and negative events from public opinion and news data to determine the event categories occurring to the enterprise entity, and converting the event categories, the enterprise entity, and the corresponding timestamps into a coded form to obtain a third standardized event flow; extracting abnormal data from financial market data and supply chain data and converting the abnormal data and corresponding timestamps into a coded form to obtain a fourth standardized event flow; and integrating the first, second, third, and fourth standardized event flows into a single standardized event flow.
[0007] In one implementation of this application, risk identification and classification are performed on the standardized event flow to obtain the risk intensity corresponding to each type of event. Specifically, this includes: splitting the standardized event flow, determining the standardized event sub-flows, and dividing the time windows of the standardized time sub-flows to obtain a list of risk events within the corresponding time windows; locating the source node of each risk event in the risk event list in a preset association graph in the target enterprise, and querying the association path of the target enterprise based on the source node; and performing risk transmission simulation on the association path to calculate the residual risk intensity corresponding to each risk event transmitted to the target enterprise, thereby determining the risk intensity.
[0008] In one implementation of this application, risk correlation analysis is performed on standardized event flows based on risk intensity to determine the dynamic risk characteristics of the target enterprise. Specifically, this includes: superimposing all risk intensities transmitted to the target enterprise to obtain the risk transmission intensity of each risk event correlation network; and integrating the features of risk transmission intensities within the same time window to determine the dynamic risk characteristics of the target enterprise.
[0009] In one implementation of this application, static fundamental features and dynamic risk features are fused and concatenated to obtain risk fusion features. Specifically, this includes: aligning the static fundamental features and dynamic risk features with data to obtain standardized static fundamental features and dynamic risk features; and concatenating the standardized static fundamental features and dynamic risk feature vectors to obtain risk fusion features.
[0010] In one implementation of this application, calculating the corporate default probability corresponding to the risk fusion feature specifically includes: determining the root node corresponding to the risk fusion feature, and based on the root node, determining the leaf node flow direction corresponding to the risk fusion feature by judging the tree node flow direction; calculating the corresponding leaf node scores of all leaf nodes according to the leaf node flow direction; and mapping the leaf node branches to the probability interval through the Sigmoid function to obtain the corporate default probability corresponding to the risk fusion feature.
[0011] In one implementation of this application, risk fusion features are assigned feature contributions to obtain risk attribution data for the target enterprise. Specifically, this includes: calculating the average probability value corresponding to the enterprise default probability to determine the benchmark value for feature contribution assignment; comparing the enterprise default probability corresponding to the risk fusion feature with the benchmark value to obtain the probability difference; and assigning a corresponding feature contribution value to each risk fusion feature based on the probability difference to obtain risk attribution data for the target enterprise.
[0012] In one implementation of this application, the corporate credit prediction text of the target enterprise is determined by extracting contribution values from the risk attribution data using natural language. Specifically, this includes: selecting several feature contribution values with the highest absolute values in the risk attribution data, and converting these feature contribution values into corresponding natural language through natural language extraction to determine the corporate credit prediction text of the target enterprise.
[0013] Secondly, this application also provides a multi-source data fusion enterprise credit comprehensive prediction device, characterized in that the device includes: a data acquisition module, used to collect enterprise-related data from the target enterprise and standardize the enterprise-related data to obtain a standardized event stream of the target enterprise; wherein, the enterprise-related data includes: judicial litigation data, business operation data, public opinion news data, financial market data, and supply chain data; a dynamic risk feature generation module, used to identify and classify the standardized event stream for risk, to obtain the risk intensity corresponding to each type of event, and based on the risk intensity, to perform risk correlation analysis on the standardized event stream to determine the dynamic risk features of the target enterprise; a fusion prediction module, used to call the static fundamental features in the enterprise database of the target enterprise, and to fuse and splice the static fundamental features and dynamic risk features to obtain risk fusion features; the fusion prediction module is also used to calculate the enterprise default probability corresponding to the risk fusion features, and to allocate the feature contribution of the risk fusion features to obtain the risk attribution data of the target enterprise; and an enterprise credit prediction text generation module, used to extract the contribution value from the risk attribution data through natural language to determine the enterprise credit prediction text of the target enterprise.
[0014] Thirdly, embodiments of this application also provide a non-volatile computer storage medium for enterprise credit comprehensive prediction based on multi-source data fusion, storing computer-executable instructions, characterized in that the computer-executable instructions, when executed, can realize an enterprise credit comprehensive prediction method based on multi-source data fusion.
[0015] This application provides a method, device, and medium for comprehensive corporate credit prediction based on multi-source data fusion. By performing risk intensity analysis and correlation graph transmission path analysis on the collected corporate-related data, it enables the judgment of the risk transmission of the target enterprise. Furthermore, by performing default probability analysis based on a combination of dynamic and static data, it enables corporate credit prediction that includes risk transmission information, thus solving the technical problem in the prior art that corporate credit prediction cannot predict the risk transmission of related parties. Attached Figure Description
[0016] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart of a multi-source data fusion method for comprehensive enterprise credit prediction is provided in this application embodiment; Figure 2 This is a schematic diagram of a multi-source data fusion enterprise credit comprehensive prediction device provided in an embodiment of this application. Detailed Implementation
[0017] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0018] This application provides a method, device, and medium for comprehensive corporate credit prediction based on multi-source data fusion. By performing risk intensity analysis and correlation graph transmission path analysis on the collected corporate-related data, it enables the judgment of the risk transmission of the target enterprise. Furthermore, by performing default probability analysis based on a combination of dynamic and static data, it enables corporate credit prediction that includes risk transmission information, thus solving the technical problem in the prior art that corporate credit prediction cannot predict the risk transmission of related parties.
[0019] The technical solutions proposed in the embodiments of this application will be described in detail below with reference to the accompanying drawings.
[0020] Figure 1This document provides a flowchart of a multi-source data fusion method for comprehensive enterprise credit prediction, as illustrated in an embodiment of this application. Figure 1 As shown in the figure, the enterprise credit comprehensive prediction method provided by the embodiment of this application through multi-source data fusion specifically includes the following steps: Step 101: Collect relevant enterprise data from the target enterprise and standardize the relevant enterprise data to obtain the standardized event flow of the target enterprise.
[0021] The enterprise-related data includes: judicial litigation data, business operation data, public opinion and news data, financial market data, and supply chain data.
[0022] For example, this application collects enterprise-related data from the target enterprise and standardizes the enterprise-related data to obtain the standardized event flow of the target enterprise, transforming multi-source data into analyzable time-series data, thus solving the fusion difficulties caused by the heterogeneity of enterprise-related data and providing a data foundation for subsequent dynamic risk perception.
[0023] Specifically, the relevant enterprise data is standardized to obtain a standardized event flow for the target enterprise. This includes: extracting time data from the relevant enterprise data and converting it into timestamps; extracting case data from judicial litigation data and converting the case data and corresponding timestamps into a coded form to obtain a first standardized event flow; extracting changes to the enterprise entity from business registration data and converting the changes and corresponding timestamps into a coded form to obtain a second standardized event flow; these changes include changes in legal representative, key personnel, and registered capital; extracting positive and negative events from public opinion and news data to determine the event categories occurring to the enterprise entity, and converting the event categories, enterprise entity, and corresponding timestamps into a coded form to obtain a third standardized event flow; extracting abnormal data from financial market data and supply chain data and converting the abnormal data and corresponding timestamps into a coded form to obtain a fourth standardized event flow; and integrating the first, second, third, and fourth standardized event flows into a single standardized event flow.
[0024] In one embodiment, a data acquisition module is used to perform real-time credit monitoring on target company A.
[0025] First, the judicial data collector obtains court judgments and enforcement announcements related to Company A and its subsidiaries in which it holds more than 20% of the shares by calling the authorized API of a third-party judicial big data service provider. For the obtained JSON messages, the case cause field is extracted and matched with the built-in judicial case cause-risk event type mapping table.
[0026] For example, the sales contract dispute is converted into the event type code LAW01. At the same time, the system locates whether Company A or its subsidiaries are defendants or judgment debtors from the list of parties, extracts the amount involved from the document using regular expressions, and finally converts the date of the document into a timestamp.
[0027] Then, the business data collector polls the public interface of the enterprise credit information disclosure system to query the change records of Company A and its subsidiaries. When a change in the legal representative or key personnel field is detected, an event type code BIZ02 is automatically generated, and the name of the changed personnel is stored as an attribute field. If a reduction in registered capital of more than 20% is detected, an event type code BIZ05 is generated, and the change ratio is calculated.
[0028] Furthermore, the public opinion data collector subscribes to the news feed API of a professional public opinion service provider, receives news reports containing the names of Company A or its core suppliers and customers in real time, scores the news titles and summaries by calling the internal sentiment analysis microservice, outputs a sentiment polarity score between -1 and 1, and classifies the events into predefined categories such as production accidents and successful financing based on keyword hitting rules.
[0029] The obtained standardized event objects are uniformly pushed to the same topic in the message queue, sorted according to the timestamp field of the event, and out-of-order data is removed to obtain the standardized real-time event stream of Company A.
[0030] Step 102: Perform risk identification and classification on the standardized event flow to obtain the risk intensity corresponding to each type of event, and based on the risk intensity, perform risk correlation analysis on the standardized event flow to determine the dynamic risk characteristics of the target enterprise.
[0031] For example, this application identifies and classifies standardized event flows by risk to obtain the risk intensity corresponding to each type of event, and performs risk correlation analysis on the standardized event flows based on the risk intensity to determine the dynamic risk characteristics of the target enterprise, thereby realizing the perception of the risk evolution trend of the target enterprise and the analysis of the risk transmission path in the enterprise relationship network.
[0032] Specifically, standardized event flows are risk-identified and classified to obtain the risk intensity corresponding to each type of event. This includes: splitting the standardized event flow, determining standardized event sub-flows, and dividing the time windows of standardized time sub-flows to obtain a list of risk events within the corresponding time windows; locating the source node of each risk event in the risk event list in the pre-defined association graph of the target enterprise, and querying the association path of the target enterprise based on the source node; and performing risk transmission simulation on the association path to calculate the residual risk intensity corresponding to each risk event transmitted to the target enterprise, thereby determining the risk intensity.
[0033] Furthermore, based on risk intensity, risk correlation analysis is performed on standardized event flows to determine the dynamic risk characteristics of the target enterprise. Specifically, this includes: superimposing all risk intensities transmitted to the target enterprise to obtain the risk transmission intensity of each risk event correlation network; and integrating the features of risk transmission intensities within the same time window to determine the dynamic risk characteristics of the target enterprise.
[0034] In one embodiment, the dynamic risk feature generation module receives a standardized event stream from Company A and maintains a 90-day sliding time window for Company A. Whenever a new event arrives, the window automatically slides, and expired events are removed.
[0035] First, the event list in the window is sent to the risk pattern matcher, which pre-loads an expert library of risk control rules. The matcher performs a batch scan of the event list and outputs the set of pattern tags that Company A has currently triggered.
[0036] Then, through intensity analysis, the events to be executed are selected from the event list, and the number of times and total amount are statistically analyzed by week. The average weekly increase in the execution frequency and the average weekly increase in the execution amount of Company A are determined. At the same time, the total number of times in the window is compared with the historical distribution of companies in the same industry.
[0037] The event type is encoded into a character sequence, and the system scans for chain risk patterns defined in the rule base. It detects that equity pledge and change of legal representative (code BIZ02) occurred within 15 days of each other, and determines that the matching pattern is the actual controller risk chain. The confidence level is calculated based on the number of days between the two events and the proportion of pledged equity.
[0038] Finally, after receiving a public opinion event concerning a production safety accident involving Company B, a Tier 1 supplier of Company A, a query was initiated to the graph database to obtain the supply chain path from Company B to Company A. It was found that Company A relies on Company B for 60% of the procurement of a certain key component. Based on the preset transmission attenuation coefficient of 0.4, the residual intensity of the event transmitted to Company A was calculated.
[0039] The residual strength of all paths is superimposed to obtain the network transmission strength value of this event to Company A, and the dynamic risk feature vector is obtained by normalization.
[0040] Step 103: Call the static fundamental features in the target company's enterprise database, and merge and combine the static fundamental features and dynamic risk features to obtain risk fusion features.
[0041] For example, this application calls the static fundamental features in the target company's enterprise database and merges and splices the static fundamental features and dynamic risk features to obtain risk fusion features, thereby realizing the combination of dynamic and static features in enterprise credit prediction, which can meet the needs of long-term credit prediction and immediate risk prediction.
[0042] Specifically, static fundamental features and dynamic risk features are fused and concatenated to obtain risk fusion features, including: aligning the static fundamental features and dynamic risk features with data to obtain standardized static fundamental features and dynamic risk features; and concatenating the standardized static fundamental features and dynamic risk feature vectors to obtain risk fusion features.
[0043] In one embodiment, after Company A's dynamic risk feature vector is generated, the latest quarterly financial data of Company A is read, and six static indicators are calculated: debt-to-equity ratio of 65.2%, current ratio of 1.32, net profit margin of 5.8%, total asset turnover of 0.85, interest coverage ratio of 4.3, and zero overdue credit accounts in the past 24 months.
[0044] Then, Z-score standardization is performed on the static features. The standardized static feature vector and the dynamic risk feature vector are concatenated in sequence to form the risk fusion feature vector.
[0045] Finally, the risk fusion feature vector is encapsulated in JSON format.
[0046] Step 104: Calculate the enterprise default probability corresponding to the risk fusion feature, and allocate the feature contribution of the risk fusion feature to obtain the risk attribution data of the target enterprise.
[0047] For example, this application calculates the enterprise default probability corresponding to the risk fusion feature and assigns the feature contribution of the risk fusion feature to the risk feature to obtain the risk attribution data of the target enterprise, thereby realizing dynamic risk prediction of enterprise credit and improving the data foundation for visualization of enterprise credit violations.
[0048] Specifically, calculating the enterprise default probability corresponding to the risk fusion feature includes: determining the root node corresponding to the risk fusion feature, and based on the root node, determining the leaf node flow direction corresponding to the risk fusion feature by judging the flow direction of tree nodes; calculating the corresponding leaf node scores of all leaf nodes according to the leaf node flow direction; and mapping the leaf node branches to the probability interval through the Sigmoid function to obtain the enterprise default probability corresponding to the risk fusion feature.
[0049] Furthermore, the risk fusion features are assigned feature contributions to obtain risk attribution data for the target enterprise. Specifically, this includes: calculating the average probability value corresponding to the enterprise default probability to determine the benchmark value for feature contribution allocation; comparing the enterprise default probability corresponding to the risk fusion feature with the benchmark value to obtain the probability difference; and assigning a corresponding feature contribution value to each risk fusion feature based on the probability difference to obtain risk attribution data for the target enterprise.
[0050] In one embodiment, after receiving the risk fusion feature vector of Company A, the fusion prediction module sends it into the loaded LightGBM model.
[0051] The model consists of 50 decision trees, each with a depth limit of 6 layers. Starting from the first tree, the network transmission strength value (0.31) in the feature vector is compared with the splitting threshold of the root node (0.25), and the flow proceeds to the right child node. The execution frequency trend value (0.85) is compared with the node threshold (0.60), and the process continues to the right until a leaf node is reached, obtaining the tree's base score of -0.12. This process is repeated to traverse all 50 trees, and the scores of all leaf nodes are summed to obtain a total score of 2.37. The total score is then input into the Sigmoid function mapping layer to calculate the default probability.
[0052] Then, the global average predicted probability (i.e., the baseline value) of 0.15 is read from the training set. Based on the feature vector of Company A, the Shapley value of each feature is approximately calculated using the Monte Carlo sampling method.
[0053] Finally, the contribution value of the network transmission strength feature is +0.38, the contribution value of the execution frequency trend is +0.25, the contribution value of the supplier production accident sequence confidence is +0.12, and the contribution value of net profit margin is -0.05. The feature name, actual value of the feature, contribution value and contribution direction are integrated into the risk attribution data object.
[0054] Step 105: Extract the contribution value from the risk attribution data using natural language processing to determine the corporate credit prediction text for the target company.
[0055] Specifically, the corporate credit prediction text for the target company is determined by extracting contribution values from the risk attribution data using natural language. This includes: selecting the highest absolute value of several feature contribution values from the risk attribution data, and converting these feature contribution values into corresponding natural language through natural language extraction to determine the corporate credit prediction text for the target company.
[0056] In one embodiment, firstly, the feature contribution values are sorted in descending order of absolute value, and the top three features are selected, including: network transmission strength (contribution value +0.38), execution frequency trend (contribution value +0.25), and supplier production accident sequence confidence (contribution value +0.12). Network transmission strength is mapped to related-party risk transmission, execution frequency trend is mapped to a rapid increase in litigation risk, and supplier production accident sequence confidence is mapped to abnormal operations of core suppliers.
[0057] Then, the pre-defined attribution report template is invoked: "Based on assessment, Company A's probability of default in the next year is {probability}, significantly higher than the industry average. The primary risk is attributed to: {reason 1}; secondly, {reason 2}; and furthermore, {reason 3}."
[0058] Fill the corresponding placeholders with the mapped terms and attach the actual values of each feature to generate the final text: "Based on the assessment, Company A's probability of default in the next year is 91.4%, significantly higher than the industry average. The main risks are attributed to: related party risk transmission (transmission strength 0.31), with a sudden production accident at its key supplier causing a sharp increase in the risk of supply chain disruption; secondly, the risk of litigation is rising rapidly (the weekly increase in the frequency of enforcement in the past month is 85%), indicating that the company's liquidity pressure continues to worsen; in addition, the core supplier is operating abnormally (confidence level 0.82), with a single source of goods dependence as high as 60%."
[0059] The above are embodiments of the method proposed in this application. Based on the same inventive concept, embodiments of this application also provide a multi-source data fusion enterprise credit comprehensive prediction device, the structure of which is as follows: Figure 2 As shown.
[0060] Figure 2 This is a schematic diagram of a multi-source data fusion-based enterprise credit comprehensive prediction device provided in an embodiment of this application. Figure 2 As shown, the device includes: The data acquisition module 201 is used to collect enterprise-related data from the target enterprise and standardize the data to obtain a standardized event flow for the target enterprise. This enterprise-related data includes: judicial litigation data, business operation data, public opinion and news data, financial market data, and supply chain data. The dynamic risk feature generation module 202 is used to identify and classify the standardized event flow to obtain the risk intensity corresponding to each type of event. Based on the risk intensity, it performs risk correlation analysis on the standardized event flow to determine the dynamic risk characteristics of the target enterprise. The fusion prediction module 203 is used to call the static fundamental features from the target enterprise's database and fuse the static fundamental features and dynamic risk features to obtain risk fusion features. The fusion prediction module 203 is also used to calculate the enterprise default probability corresponding to the risk fusion features and allocate the feature contribution of the risk fusion features to obtain the risk attribution data of the target enterprise. The enterprise credit prediction text generation module 204 is used to extract the contribution value from the risk attribution data using natural language to determine the enterprise credit prediction text of the target enterprise.
[0061] Some embodiments of this application provide corresponding to Figure 1 A non-volatile computer storage medium for enterprise credit comprehensive prediction based on multi-source data fusion, storing computer-executable instructions, which, when executed, can realize an enterprise credit comprehensive prediction method based on multi-source data fusion.
[0062] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for IoT devices and media are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.
[0063] The systems, media, and methods provided in this application are one-to-one correspondences. Therefore, the systems and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the systems and media will not be repeated here.
[0064] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0065] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0066] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0067] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0068] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.
[0069] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.
[0070] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.
[0071] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.
[0072] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.
Claims
1. A comprehensive enterprise credit prediction method based on multi-source data fusion, characterized in that, The method includes: Collect enterprise-related data from the target enterprise and standardize the enterprise-related data to obtain the standardized event flow of the target enterprise; wherein, the enterprise-related data includes: judicial litigation data, business operation data, public opinion news data, financial market data, and supply chain data; The standardized event flow is risk-identified and classified to obtain the risk intensity corresponding to each type of event. Based on the risk intensity, risk correlation analysis is performed on the standardized event flow to determine the dynamic risk characteristics of the target enterprise. The static fundamental features from the target company's enterprise database are retrieved, and the static fundamental features and the dynamic risk features are fused and combined to obtain risk fusion features; Calculate the enterprise default probability corresponding to the risk fusion feature, and assign the feature contribution of the risk fusion feature to obtain the risk attribution data of the target enterprise; The contribution value of the risk attribution data is extracted using natural language to determine the corporate credit prediction text of the target enterprise.
2. The enterprise credit comprehensive prediction method based on multi-source data fusion according to claim 1, characterized in that, The relevant data of the enterprise is standardized to obtain a standardized event flow for the target enterprise, specifically including: Extract the relevant time data of the enterprise and convert the time data into a timestamp; Extract the case data from the judicial litigation data, and convert the case data and the corresponding timestamps into an encoded form to obtain the first standardized event stream; Extract the changes to the business entity from the business registration data, and convert the changes and their corresponding timestamps into a coded form to obtain a second standardized event flow; wherein, the changes include: changes to legal representative, changes to key personnel, and changes to registered capital; Positive and negative events are extracted from the public opinion news data to determine the event categories of the enterprise entity, and the event categories, the enterprise entity, and the corresponding timestamps are converted into coded forms to obtain a third standardized event stream; Extract abnormal data from the financial market data and the supply chain data, and convert the abnormal data and the corresponding timestamps into an encoded form to obtain the fourth standardized event stream; The first standardized event stream, the second standardized event stream, the third standardized event stream, and the fourth standardized event stream are integrated into the standardized event stream.
3. The enterprise credit comprehensive prediction method based on multi-source data fusion according to claim 1, characterized in that, The standardized event flow is subjected to risk identification and classification to obtain the risk intensity corresponding to each type of event, specifically including: The standardized event stream is split into sub-streams to determine the standardized event sub-streams, and the time windows of the standardized time sub-streams are divided to obtain a list of risk events within the corresponding time window. Locate the source node of each risk event in the risk event list within the preset association graph of the target enterprise, and query the association path of the target enterprise based on the source node; Risk transmission simulation is performed on the associated path to calculate the residual risk intensity corresponding to each risk event transmitted to the target enterprise, thereby determining the risk intensity.
4. The enterprise credit comprehensive prediction method based on multi-source data fusion according to claim 1, characterized in that, Based on the aforementioned risk intensity, risk correlation analysis is performed on the standardized event flow to determine the dynamic risk characteristics of the target enterprise, specifically including: The risk intensity of all risks transmitted to the target enterprise is superimposed to obtain the risk transmission intensity of the risk event association network; The risk transmission intensity within the same time window is feature-integrated to determine the dynamic risk characteristics of the target enterprise.
5. The enterprise credit comprehensive prediction method based on multi-source data fusion according to claim 1, characterized in that, The static fundamental features and the dynamic risk features are fused and combined to obtain risk fusion features, specifically including: Data alignment is performed on the static fundamental features and the dynamic risk features to obtain standardized static fundamental features and dynamic risk features; The standardized static fundamental features and dynamic risk feature vectors are concatenated to obtain the risk fusion feature.
6. The enterprise credit comprehensive prediction method based on multi-source data fusion according to claim 1, characterized in that, Calculating the corporate default probability corresponding to the risk fusion feature specifically includes: Determine the root node corresponding to the risk fusion feature, and based on the root node, determine the leaf node flow direction corresponding to the risk fusion feature by judging the tree node flow direction; Based on the flow direction of the leaf nodes, the corresponding leaf node scores of all leaf nodes are accumulated and calculated. By using the Sigmoid function, the leaf node branches are mapped to probability intervals to obtain the enterprise default probability corresponding to the risk fusion feature.
7. The enterprise credit comprehensive prediction method based on multi-source data fusion according to claim 6, characterized in that, The risk fusion features are assigned a feature contribution to obtain the risk attribution data of the target enterprise, specifically including: Calculate the average probability value corresponding to the enterprise's default probability to determine the benchmark value for feature contribution allocation; The enterprise default probability corresponding to the risk fusion feature is compared with the benchmark value to obtain the probability difference. Based on the probability difference, a corresponding feature contribution value is assigned to each risk fusion feature to obtain the risk attribution data of the target enterprise.
8. The enterprise credit comprehensive prediction method based on multi-source data fusion according to claim 1, characterized in that, The contribution value of the risk attribution data is extracted using natural language processing to determine the corporate credit prediction text for the target enterprise, specifically including: The risk attribution data is filtered to select the feature contribution values with the highest absolute values, and then the feature contribution values are converted into corresponding natural language through natural language extraction to determine the corporate credit prediction text of the target enterprise.
9. A multi-source data fusion enterprise credit comprehensive prediction device, characterized in that, The device includes: The data acquisition module is used to collect enterprise-related data from the target enterprise and to standardize the enterprise-related data to obtain the standardized event flow of the target enterprise; wherein, the enterprise-related data includes: judicial litigation data, business operation data, public opinion and news data, financial market data, and supply chain data; The dynamic risk feature generation module is used to identify and classify the standardized event flow to obtain the risk intensity corresponding to each type of event, and based on the risk intensity, to perform risk correlation analysis on the standardized event flow to determine the dynamic risk features of the target enterprise. The fusion prediction module is used to call the static fundamental features in the enterprise database of the target enterprise, and fuse and combine the static fundamental features and the dynamic risk features to obtain risk fusion features; The fusion prediction module is also used to calculate the enterprise default probability corresponding to the risk fusion feature, and to allocate the feature contribution of the risk fusion feature to obtain the risk attribution data of the target enterprise. The enterprise credit prediction text generation module is used to extract contribution values from the risk attribution data using natural language to determine the enterprise credit prediction text of the target enterprise.
10. A non-volatile computer storage medium for enterprise credit comprehensive prediction based on multi-source data fusion, storing computer-executable instructions, characterized in that, When the computer-executable instructions are executed, they can realize the enterprise credit comprehensive prediction method of multi-source data fusion as described in claims 1-8.