A method for generating enterprise risk signals and related products

Abstract

The application discloses an enterprise risk signal generation method and related products. Among them, the method comprises: obtaining a preset risk rule and a plurality of preset threshold intervals corresponding to the preset risk rule; wherein each preset threshold interval corresponds to a signal strength; obtaining internal data and external data of a target enterprise, and processing the internal data and the external data respectively to obtain first feature data about the preset risk rule; for any first feature data, based on the preset threshold interval corresponding to the first feature data, a risk signal of the corresponding signal strength is generated. The technical scheme of the application realizes the dynamic alignment of the early warning sensitivity and the actual risk level of the enterprise in a data-driven and interval-adaptive manner, effectively reduces unnecessary control cost, significantly compresses the delay window of risk identification, and systematically solves the industry problem of coexistence of excessive early warning and early warning lag.

Description

Technical Field

[0001] This application relates to the field of data processing technology, specifically to a method for generating enterprise risk signals and related products. Background Technology

[0002] With the rapid development of information technology, data collection, storage, and processing have become increasingly easier. As data increases, generating appropriate risk signals becomes more and more important.

[0003] In related technologies, risk assessment thresholds are mostly fixed values, which leads to a mismatch between the risk assessment thresholds and the actual risk situation of enterprises. This results in either excessive warnings that increase management and control costs, or delayed warnings that miss the opportunity for prevention and control. Summary of the Invention

[0004] In view of this, this application provides a method for generating enterprise risk signals and related products, providing risk judgment thresholds that match risk judgment thresholds.

[0005] To solve the above problems, the technical solution provided in this application is as follows: In a first aspect of this application, a method for generating enterprise risk signals is provided, the method comprising: Obtain preset risk rules and multiple preset threshold intervals corresponding to the preset risk rules; wherein, each preset threshold interval corresponds to a signal strength. Acquire internal and external data of the target company, and process the internal and external data respectively to obtain the first feature data about the preset risk rules; For any first feature data, a risk signal with corresponding signal strength is generated based on the preset threshold range corresponding to the first feature data.

[0006] In one possible embodiment, the preset risk rules are constructed according to the following method: Acquire internal and external data from multiple enterprises, and process the internal and external data separately to obtain the second feature data; For each of the multiple enterprises, determine the enterprise's status label based on the enterprise's corresponding internal and external data; Based on the second feature data, status labels, and development indicators, the initial risk rule trigger generates early warning signal records, and sample data is obtained based on the signal records; Based on the first or second state indicated by the status label, and the customer groups with and without warning signals indicated by the sample data, the initial risk rules are filtered to obtain the preset risk rules.

[0007] In one possible embodiment, the preset threshold range and signal strength are set according to the following method: For any preset risk rule, the indicators corresponding to the preset risk rule are divided into multiple preset threshold intervals by quantiles; Calculate the risk signal hit rate within each preset threshold interval, and determine the signal strength corresponding to each preset threshold interval based on the risk signal hit rate.

[0008] In one possible embodiment, the first state characterizes poorly, and the second state characterizes well. Based on the first or second state indicated by the state label, and the customer groups with and without warning signals indicated by the sample data, initial risk rules are filtered to obtain preset risk rules, including: Determine the percentage of the first-ranked company in poor condition among the customer group with warning signals, and the percentage of the second-ranked company in poor condition among the customer group without warning signals; If the ratio of the proportion of the first enterprise to the proportion of the second enterprise is greater than or equal to a set threshold, a preset risk rule is determined.

[0009] In one possible embodiment, the second feature data are standardized.

[0010] In one possible embodiment, internal data includes basic credit information data, financial data, collateral data, corporate relationship data, internal rating data, financing data, and early warning data; external data includes business registration data, judicial data, penalty data, public opinion data, credit reporting data, external rating data, intellectual property data, industry data, environmental protection data, invoice verification data, tax rating data, local government data, urban investment data, bond data, real estate enterprise data, bidding data, and recruitment data.

[0011] In one possible embodiment, the method further includes: Risk signals are pushed to target companies so that they can respond accordingly.

[0012] In a second aspect of this application, an enterprise risk signal generation device is provided, the device comprising: a first acquisition module, a second acquisition module, and a generation module; The first acquisition module is configured to acquire a preset risk rule and multiple preset threshold intervals corresponding to the preset risk rule; wherein, each preset threshold interval corresponds to a signal strength. The second acquisition module is configured to acquire the target company's internal and external data, and process the internal and external data respectively to obtain the first feature data about the preset risk rules; The generation module is configured to generate a risk signal with corresponding signal strength for any first feature data, based on a preset threshold range corresponding to the first feature data.

[0013] In one possible embodiment, the apparatus further includes: a construction module; The construction module is configured to acquire internal and external data from multiple enterprises, process the internal and external data respectively to obtain second feature data; for each enterprise, determine the enterprise's status label based on the corresponding internal and external data; based on the second feature data, status label, and development indicators, process the initial risk rules to trigger the generation of early warning signal records, and obtain sample data based on the signal records; based on the first or second status indicated by the status label, and the customer groups with and without early warning signals indicated by the sample data, filter the initial risk rules to obtain preset risk rules.

[0014] In one possible embodiment, the device further includes: a setting module; The setting module is configured to set preset threshold intervals and signal strengths in the following manner: for any preset risk rule, the indicator corresponding to the preset risk rule is divided into multiple preset threshold intervals by quantiles; the risk signal hit rate in each preset threshold interval is calculated, and the signal strength corresponding to each preset threshold interval is determined based on the risk signal hit rate.

[0015] In one possible embodiment, the construction module is configured to, when the first state characterization is poor and the second state characterization is excellent, determine the proportion of the first enterprise with poor characterization in the customer group with warning signals and the proportion of the second enterprise with poor characterization in the customer group without warning signals; and determine a preset risk rule when the ratio of the proportion of the first enterprise to the proportion of the second enterprise is greater than or equal to a set threshold.

[0016] In one possible embodiment, the second feature data are standardized.

[0017] In one possible embodiment, internal data includes at least one of the following: basic credit information data, financial data, collateral data, corporate relationship data, internal rating data, financing data, and early warning data; external data includes at least one of the following: business registration data, judicial data, penalty data, public opinion data, credit data, external rating data, intellectual property data, industry data, environmental protection data, invoice verification data, tax rating data, local government data, urban investment data, bond data, real estate enterprise data, bidding data, and recruitment data.

[0018] In one possible embodiment, the device further includes: a push module; The push module is configured to push risk signals to the target company so that the target company can respond based on the risk signals.

[0019] In a third aspect, this application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the enterprise risk signal generation method as described in any embodiment of the first aspect.

[0020] In a fourth aspect, this application provides a computer-readable storage medium storing instructions that, when executed on a terminal device, cause the terminal device to perform the enterprise risk signal generation method as described in any embodiment of the first aspect.

[0021] This application's technical solution fundamentally breaks through the rigid judgment mode of the traditional fixed threshold "one-size-fits-all" approach by introducing a multi-level threshold range and signal strength grading mechanism, thereby accurately resolving the contradiction between risk warnings and actual conditions. Specifically, firstly, multiple preset threshold ranges are set for each preset risk rule, and each range is associated with a specific signal strength (such as weak, medium, and strong). This is equivalent to constructing a continuous and hierarchical risk measurement scale, rather than a single "trigger / non-trigger" dichotomy point. Subsequently, the internal data (such as financial and operational indicators) and external data (such as industry sentiment and market fluctuations) of the target company are integrated and processed to generate first feature data that reflects the company's current true risk profile. Finally, by mapping this dynamic feature data to the corresponding threshold range, the system outputs a matching signal strength—releasing a weak signal when the feature value falls into a lower range to avoid invalid interference, and issuing a strong signal when it falls into a higher range to ensure timely response. The core of this process lies in the diversity of threshold ranges and the gradient of signal strength. This allows risk assessment to be smoothly adjusted according to changes in the company's real-time data, preventing the underreporting of progressive risks due to excessively high fixed values ​​and avoiding frequent alarms for normal fluctuations due to excessively low fixed values. Therefore, this application's technical solution, through a data-driven and range-adaptive approach, achieves dynamic alignment between early warning sensitivity and the company's actual risk level. While effectively reducing unnecessary control costs, it significantly shortens the risk identification delay window, thus systematically solving the industry problem of both excessive and delayed early warnings. Attached Figure Description

[0022] Figure 1 A flowchart for constructing risk rules is provided as an embodiment of this application; Figure 2 A flowchart illustrating a method for generating enterprise risk signals provided in this application embodiment; Figure 3 This is a schematic diagram of an enterprise risk signal generation device provided in an embodiment of this application. Detailed Implementation

[0023] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0024] The terms "first" and "second," etc., used in the specification and claims of this application are used to distinguish different objects, not to describe a specific order of objects. For example, "first operation instruction" and "second operation instruction," etc., are used to distinguish different operation instructions, not to describe a specific order of operation instructions.

[0025] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0026] In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more, for example, multiple processing units means two or more processing units, multiple elements means two or more elements, etc.

[0027] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the embodiments of this application will be further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0028] For ease of understanding, the embodiments of this application will be introduced in two parts: first, offline construction of risk rules, that is, the preset risk rules used in the embodiments of this application; second, generation of risk signals.

[0029] See Figure 1 The figure is a flowchart of a risk rule construction method provided in an embodiment of this application.

[0030] like Figure 1 As shown, the method for constructing risk rules includes the following steps: S1100: Acquire internal and external data from multiple enterprises, and process the internal and external data respectively to obtain the second feature data.

[0031] For example, the internal data in this application embodiment includes: basic credit information data, financial data, collateral data, corporate relationship data, internal rating data, financing data, and early warning data.

[0032] For example, the external data in this application embodiment includes: business registration data, judicial data, penalty data, public opinion data, credit data, external rating data, intellectual property data, industry data, environmental protection data, invoice verification data, tax rating data, local government data, urban investment data, bond data, real estate enterprise data, bidding data, and recruitment data.

[0033] Based on the acquired internal and external data, this application embodiment can integrate the internal and external data of multiple enterprises, clean, integrate, analyze the data, and construct a unified calculation standard for feature data.

[0034] First, it is necessary to standardize the governance of data sources and indicators. This means that the system should establish a global indicator dictionary to clarify the unique business meaning, calculation formula (such as "current assets / current liabilities"), and data source table for each risk characteristic (such as "enterprise short-term solvency"). At the same time, the raw data should be cleaned by ETL tools to eliminate the differences in format and unit between different business systems. For example, the "1000 yuan" in system A and the "1000.00 yuan" in system B should be unified into a numerical standard format.

[0035] Secondly, a unified computation and processing engine is constructed, requiring the system to have a powerful built-in feature engineering and indicator calculation module capable of automatically processing raw data into standardized secondary feature data based on the logic embedded in the indicator dictionary. For example, for the feature "debt-to-asset ratio," regardless of whether the data comes from the total liabilities of the financial system or the total assets of another system, the engine will strictly execute the calculation according to the formula "total liabilities / total assets." For scenarios with high real-time requirements, streaming computing frameworks (such as Flink) will be introduced for window calculation and state aggregation. In addition, by recording data lineage, the computation path of any feature value is ensured to be traceable.

[0036] For example, suppose the system needs to construct a risk feature called "corporate cash flow pressure." Without a unified standard, the sales department might calculate it as "(accounts receivable + inventory) / current assets," while the finance department might calculate it as "(accounts payable - cash and cash equivalents) / total assets," leading to drastically different stress scores for the same company. This solution explicitly defines this feature in the indicator dictionary as "(accounts receivable turnover days + inventory turnover days) / 2," and forces all calculation tasks to extract standardized fields such as "accounts receivable" and "inventory" from a unified data mart. When internal and external data are updated, the calculation engine automatically pulls data from this unified source, executes the predetermined formula, and generates a standardized feature value that is consistent and comparable across different systems, completely eliminating deviations caused by different calculation logics.

[0037] S1200: For each of multiple enterprises, determine the enterprise's status label based on the enterprise's corresponding internal and external data.

[0038] In the risk identification and assessment process, clearly defining "good" and "bad" and calculating enterprise labels accordingly is the cornerstone step of the entire modeling and decision-making system.

[0039] For example, after completing the collection and standardization of multi-source data, the system first needs to establish the criteria for judging good and bad samples based on the collected internal historical data (such as repayment records and aging distribution).

[0040] Among them, "bad" usually refers to enterprises that have been overdue for more than 5 days in principal or more than 10 days in interest, or have extended their loans, or have borrowed new or old loans, or have been classified into the last three categories of the five-level classification, or have an account status of write-off; "good" refers to enterprises that are not "bad".

[0041] In one possible implementation, to balance objectivity and adaptability, a dual calibration approach of "expert rules + data-driven" can be adopted: first, a basic threshold is set based on business experience, and then cross-validation is performed using quantiles of historical data distribution (e.g., marking the bottom 10% of companies in terms of overall financial score as "bad") to dynamically optimize the judgment boundary. When calculating company labels, the system substitutes each company's key characteristic values ​​(such as debt service ratio and negative public opinion rate) into the above definition, and outputs a binary result through logical judgment or a scoring card model. If the overall characteristic score is lower than the preset "good" lower limit, a "bad" label is assigned; otherwise, a "good" label is assigned.

[0042] For example, for a group of small and medium-sized manufacturing enterprises, if an enterprise's principal is overdue for more than 5 days, interest is overdue for more than 10 days, or it has extended its loan, borrowed new or old loans, or is classified into the last three categories of the five-level risk assessment, or its account status is write-off, the system will automatically classify it as a "bad" sample. After the label calculation is completed, backtesting is required using an independent validation set to ensure that the distinction between "good" and "bad" is significant and the false positive rate is controllable. The definition benchmark is also updated regularly according to changes in the macro environment so that the labels always reflect the latest real risk status of the enterprises, providing reliable monitoring signals for subsequent rule training, threshold optimization, and early warning strategies. This process not only transforms qualitative risk perception into quantitative and operable indicators, but also ensures the objectivity, traceability, and continuous adaptability of the labels.

[0043] S1300: Based on the second feature data, status label and development indicators, process the initial risk rule trigger to generate early warning signal records, and obtain sample data based on the early warning signal records.

[0044] The warning signals in this application embodiment include valid warning signals and invalid warning signals.

[0045] A warning signal is considered "invalid" if it meets any one of the following three criteria: (1) Taking the time of occurrence of a rule as the observation point, only the enterprise status label within 12 months (inclusive) before and after that time point is considered as a "valid" label. For a certain time point of a rule signal, if the status label of an enterprise entity is outside 12 months before and after that time point, it is considered as a signal that needs to be excluded, that is, an "invalid" warning signal - the enterprise to which the signal belongs is not within the monitoring scope when the signal is generated.

[0046] (2) If the time point of occurrence of a certain rule signal is taken as the observation point, and the corresponding enterprise status has already shown a "bad" label 12 months (inclusive) before this time point, then the signal is considered to have no warning effect and the signal is regarded as an "invalid" warning signal.

[0047] (3) Signals with empty values: Since the early warning indicators are generated by creating a wide table based on the data information dimension, a company will have all indicator columns for that data dimension, but not every indicator can be generated. Therefore, there may be situations where the values ​​are empty. In actual use, such situations should not be regarded as generating signals, so they are considered "invalid" and removed during the analysis process.

[0048] In this application embodiment, a "valid" warning signal is defined as the signal remaining after excluding "invalid" warning signals from all signals.

[0049] Under the premise that the warning signal is "valid," taking the occurrence time of a certain rule signal as the observation point, if the enterprise's status was "good" for the 12 months (inclusive) before that time point, and then changed to "bad" for the 12 months (inclusive) after that time point, then the "warning is valid," also known as a "hit signal." In other words, if the enterprise was not in a "bad" state before the warning signal occurred, and then changed from "good" to "bad" after the signal was generated, this signal is counted as a "hit." In the embodiments of this application, "valid" warning signals and "hit" warning signals can be selected as sample data for analysis during the analysis process.

[0050] S1400: Based on the first or second state indicated by the status label, the sample data indicates the customer groups with warning signals and the customer groups without warning signals, and the initial risk rules are filtered to obtain the preset risk rules.

[0051] For example, if the amount announced by a state-owned enterprise implementing a risk rule is greater than 0, it is considered a warning signal, and the enterprise's "good" or "bad" status is determined. The rule's ability to differentiate between enterprise and customer risks is analyzed using four quadrants based on two dimensions (enterprise "good" / "bad" status and the presence or absence of a risk signal). This is shown in Table 1 below. Table 1 Enterprise Status Early Warning Signals

[0052] The risk rule divides enterprise customers into two groups within the industry customer base: customers with warning signals and customers without warning signals. The percentage of "bad" customers in the customer group with warning signals and the percentage of "bad" customers in the customer group without warning signals are calculated. The amplification of the risk rule is then calculated as (percentage of "bad" customers in the customer group with warning signals / percentage of "bad" customers in the customer group without warning signals). A threshold of 1.5 is set. When the ratio is greater than or equal to 1.5, the risk rule is considered significant; when the ratio is less than 1.5, it is considered insignificant, and samples with significant risk rules are retained. It should be noted that the threshold of 1.5 in this embodiment is merely exemplary, and no specific value is limited in this embodiment.

[0053] If the risk rule is significant for the entire industry customer group and also significant for the sub-industry customer groups, the risk rule will apply to the entire industry customer group; if the risk rule is significant only for the sub-industry customer groups, the risk rule will not apply to the entire industry customer group, but only to the specific industry customer group.

[0054] The preset risk rules described in this application embodiment are samples with significant risk rules in the above examples.

[0055] In addition, the risk rule threshold is determined in the following ways in the embodiments of this application.

[0056] (1) Calculate the risk signal hit rate of the risk rule in different threshold ranges.

[0057] After sorting the risk rule indicators by the number of enterprise customers, they are divided into quantiles. Currently, they are divided into 13 parts: P1, P5, P10, P20, P30, P40, P50, P60, P70, P80, P90, P95, and P99. The hit rate of risk signals within the upper and lower thresholds of each quantile is calculated: this determines whether a non-"bad" enterprise becomes "bad" within one year after the risk signal is generated. If it becomes "bad," it is considered a "hit"; otherwise, it is considered a "miss."

[0058] (2) Set risk rule thresholds and risk signal strength through data analysis.

[0059] By analyzing the risk signal hit rate of different thresholds of risk rules one by one, when the risk rule threshold is monotonically increasing with respect to enterprise risk, the risk signal hit rate at the quantile value of the cut-off point is viewed one by one from the minimum threshold range. When the risk rule threshold is monotonically decreasing with respect to enterprise risk, the risk signal hit rate at the quantile value of the cut-off point is viewed from the maximum threshold range.

[0060] For example, risk signal strength is set according to the risk signal hit rate. Currently, the comparison between setting risk rule strength and risk signal hit rate is as follows: None (no push notifications): [0,5%); Weak: [5%,20%); Middle: [20%, 50%); Strong: [50%, 80%); Gang: [80%, 100%].

[0061] Taking "Amount of Execution Announcement for Non-State-Owned Enterprises" as an example, the execution announcement amount field is divided by quantiles, and the percentage of hit signals above (>=) and below (<) quantiles is statistically analyzed. As shown in the table below, the larger the execution announcement amount, the higher the percentage of enterprises that ultimately "go bad," so the percentage of hit signals above (>=) quantiles is used. Based on the relationship between the hit rate and intensity, the P60 quantile is selected as the threshold cutoff point, with a quantile value of 600,000. At this point, two rules are generated for the execution announcement amount of non-state-owned enterprises: 1. Industry-wide applicable rule, threshold (0, 600,000), intensity: medium; 2. Industry-wide applicable rule, threshold [600,000, inf), intensity: strong. Additionally, if the analysis indicator "quotient of positive examples with / without signal" is greater than a certain threshold (default 1.5) in both the entire industry and sub-industry dimensions, then the rule is considered to be "significantly effective" in terms of discrimination across the entire industry, meaning it can be used across the entire industry without further industry segmentation; otherwise, it is necessary to conduct analysis at the sub-industry level.

[0062] For example, this application provides a statistical table of legal signal dimensions, as shown in Table 2 below: Table 2 Statistics on Legal Signal Dimensions

[0063] It should be understood that the above-mentioned S1100-S1400 can be completed offline on the server or online on the server, and no specific limitation is made in this application embodiment.

[0064] The following section will introduce the generation of enterprise risk signals, and the corresponding flowchart is shown below. Figure 2 As shown: S2100: Obtain preset risk rules and multiple preset threshold intervals corresponding to the preset risk rules; wherein, each preset threshold interval corresponds to a signal strength.

[0065] It should be understood that the degree of preset risk rules and the setting of preset threshold ranges have been specifically described in the foregoing embodiments, and will not be repeated here.

[0066] S2200: Obtain the target company's internal and external data, and process the internal and external data respectively to obtain the first feature data about the preset risk rules.

[0067] S2300: For any first feature data, generate a risk signal with corresponding signal strength based on the preset threshold range corresponding to the first feature data.

[0068] For example, each target company's characteristic data is checked to see if it triggers a preset risk rule. If it does, a risk signal and its intensity are generated and stored in the server database; if it does not, there is no risk signal.

[0069] In addition, in this embodiment of the application, risk signals can also be pushed to the target enterprise so that the target enterprise can respond based on the risk signals.

[0070] Specifically, risk signals are pushed to the target company's account manager via WeChat Work messages. This allows the account manager to check the target company's situation based on the risk signals, provide feedback on the accuracy and validity of the information, and store the content in a server database. It should be understood that using WeChat Work as the push method in the above example is merely exemplary. In addition, other methods can be used for push notifications in this embodiment, such as SMS and email.

[0071] This application's technical solution fundamentally breaks through the rigid judgment mode of the traditional fixed threshold "one-size-fits-all" approach by introducing a multi-level threshold range and signal strength grading mechanism, thereby accurately resolving the contradiction between risk warnings and actual conditions. Specifically, firstly, multiple preset threshold ranges are set for each preset risk rule, and each range is associated with a specific signal strength (such as weak, medium, and strong). This is equivalent to constructing a continuous and hierarchical risk measurement scale, rather than a single "trigger / non-trigger" dichotomy point. Subsequently, the internal data (such as financial and operational indicators) and external data (such as industry sentiment and market fluctuations) of the target company are integrated and processed to generate first feature data that reflects the company's current true risk profile. Finally, by mapping this dynamic feature data to the corresponding threshold range, the system outputs a matching signal strength—releasing a weak signal when the feature value falls into a lower range to avoid invalid interference, and issuing a strong signal when it falls into a higher range to ensure timely response. The core of this process lies in the diversity of threshold ranges and the gradient of signal strength. This allows risk assessment to be smoothly adjusted according to changes in the company's real-time data, preventing the underreporting of progressive risks due to excessively high fixed values ​​and avoiding frequent alarms for normal fluctuations due to excessively low fixed values. Therefore, this application's technical solution, through a data-driven and range-adaptive approach, achieves dynamic alignment between early warning sensitivity and the company's actual risk level. While effectively reducing unnecessary control costs, it significantly shortens the risk identification delay window, thus systematically solving the industry problem of both excessive and delayed early warnings.

[0072] Based on the foregoing embodiments, this application also provides an enterprise risk signal generation device, the schematic diagram of which is shown below. Figure 3 .

[0073] like Figure 3 As shown, the device includes: a first acquisition module 1000, a second acquisition module 2000, and a generation module 3000; The first acquisition module 1000 is configured to acquire a preset risk rule and multiple preset threshold intervals corresponding to the preset risk rule; wherein, each preset threshold interval corresponds to a signal strength. The second acquisition module 2000 is configured to acquire the target company's internal and external data, and process the internal and external data respectively to obtain the first feature data about the preset risk rules; The generation module 3000 is configured to generate a risk signal with corresponding signal strength for any first feature data based on a preset threshold range corresponding to the first feature data.

[0074] This application's technical solution fundamentally breaks through the rigid judgment mode of the traditional fixed threshold "one-size-fits-all" approach by introducing a multi-level threshold range and signal strength grading mechanism, thereby accurately resolving the contradiction between risk warnings and actual conditions. Specifically, firstly, multiple preset threshold ranges are set for each preset risk rule, and each range is associated with a specific signal strength (such as weak, medium, and strong). This is equivalent to constructing a continuous and hierarchical risk measurement scale, rather than a single "trigger / non-trigger" dichotomy point. Subsequently, the internal data (such as financial and operational indicators) and external data (such as industry sentiment and market fluctuations) of the target company are integrated and processed to generate first feature data that reflects the company's current true risk profile. Finally, by mapping this dynamic feature data to the corresponding threshold range, the system outputs a matching signal strength—releasing a weak signal when the feature value falls into a lower range to avoid invalid interference, and issuing a strong signal when it falls into a higher range to ensure timely response. The core of this process lies in the diversity of threshold ranges and the gradient of signal strength. This allows risk assessment to be smoothly adjusted according to changes in the company's real-time data, preventing the underreporting of progressive risks due to excessively high fixed values ​​and avoiding frequent alarms for normal fluctuations due to excessively low fixed values. Therefore, this application's technical solution, through a data-driven and range-adaptive approach, achieves dynamic alignment between early warning sensitivity and the company's actual risk level. While effectively reducing unnecessary control costs, it significantly shortens the risk identification delay window, thus systematically solving the industry problem of both excessive and delayed early warnings.

[0075] In one possible embodiment, the apparatus further includes: a construction module; The construction module is configured to acquire internal and external data from multiple enterprises, process the internal and external data respectively to obtain second feature data; for each enterprise, determine the enterprise's status label based on the corresponding internal and external data; based on the second feature data, status label, and development indicators, process the initial risk rules to trigger the generation of early warning signal records, and obtain sample data based on the signal records; based on the first or second status indicated by the status label, and the customer groups with and without early warning signals indicated by the sample data, filter the initial risk rules to obtain preset risk rules.

[0076] In one possible embodiment, the device further includes: a setting module; The setting module is configured to set preset threshold intervals and signal strengths in the following manner: for any preset risk rule, the indicator corresponding to the preset risk rule is divided into multiple preset threshold intervals by quantiles; the risk signal hit rate in each preset threshold interval is calculated, and the signal strength corresponding to each preset threshold interval is determined based on the risk signal hit rate.

[0077] In one possible embodiment, the construction module is configured to, when the first state characterization is poor and the second state characterization is excellent, determine the proportion of the first enterprise with poor characterization in the customer group with warning signals and the proportion of the second enterprise with poor characterization in the customer group without warning signals; and determine a preset risk rule when the ratio of the proportion of the first enterprise to the proportion of the second enterprise is greater than or equal to a set threshold.

[0078] In one possible embodiment, the second feature data are standardized.

[0079] In one possible embodiment, internal data includes basic credit information data, financial data, collateral data, corporate relationship data, internal rating data, financing data, and early warning data; external data includes business registration data, judicial data, penalty data, public opinion data, credit reporting data, external rating data, intellectual property data, industry data, environmental protection data, invoice verification data, tax rating data, local government data, urban investment data, bond data, real estate enterprise data, bidding data, and recruitment data.

[0080] In one possible embodiment, the device further includes: a push module; The push module is configured to push risk signals to the target company so that the target company can respond based on the risk signals.

[0081] In addition, this application also provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the enterprise risk signal generation method as described in any of the foregoing embodiments.

[0082] In addition, embodiments of this application also provide a computer-readable storage medium storing instructions that, when executed on a terminal device, cause the terminal device to perform the enterprise risk signal generation method as described in any of the foregoing embodiments.

[0083] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0084] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for generating enterprise risk signals, characterized in that, The methods include: Obtain preset risk rules and multiple preset threshold intervals corresponding to the preset risk rules; wherein, each preset threshold interval corresponds to a signal strength; Acquire internal and external data of the target company, and process the internal and external data respectively to obtain first feature data about the preset risk rule; For any of the first feature data, a risk signal corresponding to the signal strength is generated based on the preset threshold range corresponding to the first feature data.

2. The method according to claim 1, characterized in that, The preset risk rules are constructed according to the following method: Acquire internal and external data from multiple enterprises, and process the internal and external data respectively to obtain second feature data; For each of the multiple enterprises, a status label for the enterprise is determined based on the internal data and external data corresponding to that enterprise; Based on the second feature data, the status label, and the development indicators, the initial risk rule is processed to trigger the generation of early warning signal records, and sample data is obtained based on the early warning signal records; Based on the first or second state indicated by the status label, and the customer groups with and without warning signals indicated by the sample data, the preset risk rules are obtained by filtering the initial risk rules.

3. The method according to claim 2, characterized in that, The preset threshold range and the signal strength are set according to the following method: For any of the preset risk rules, the indicators corresponding to the preset risk rules are divided into multiple preset threshold intervals by quantiles; Calculate the risk signal hit rate within each preset threshold interval, and determine the signal strength corresponding to each preset threshold interval based on the risk signal hit rate.

4. The method according to claim 2, characterized in that, If the first state characterization is poor and the second state characterization is excellent, the process of filtering the initial risk rules to obtain the preset risk rules based on the first or second state indicated by the state label, and the customer groups with and without warning signals indicated by the sample data, includes: Determine the percentage of the first enterprise with poor condition in the customer group with warning signals, and the percentage of the second enterprise with poor condition in the customer group without warning signals; If the ratio of the proportion of the first enterprise to the proportion of the second enterprise is greater than or equal to a set threshold, the preset risk rule is determined.

5. The method according to claim 2, characterized in that, The second feature data should be standardized.

6. The method according to any one of claims 1-5, characterized in that, The internal data includes at least one of the following: basic credit information data, financial data, collateral data, corporate relationship data, internal rating data, financing data, and early warning data; the external data includes at least one of the following: business registration data, judicial data, penalty data, public opinion data, credit data, external rating data, intellectual property data, industry data, environmental protection data, invoice verification data, tax rating data, local government data, urban investment data, bond data, real estate enterprise data, bidding data, and recruitment data.

7. The method according to claim 1, characterized in that, The method further includes: The risk signal is pushed to the target company so that the target company can respond based on the risk signal.

8. A business risk signal generation device, characterized in that, The device includes: a first acquisition module, a second acquisition module, and a generation module; The first acquisition module is configured to acquire a preset risk rule and a plurality of preset threshold intervals corresponding to the preset risk rule; wherein each preset threshold interval corresponds to a signal strength. The second acquisition module is configured to acquire internal and external data of the target enterprise, and process the internal and external data respectively to obtain first feature data about the preset risk rule; The generation module is configured to generate a risk signal corresponding to the signal strength for any of the first feature data, based on the preset threshold range corresponding to the first feature data.

9. A computer device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the enterprise risk signal generation method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed on a terminal device, cause the terminal device to perform the enterprise risk signal generation method as described in any one of claims 1-7.

Images