A training method, device, medium and equipment for enterprise risk early warning model

Abstract

This specification discloses a training method, device, medium and equipment for an enterprise risk warning model, including: when training the enterprise risk warning model, the enterprise data of the enterprise determined based on the basic data of the enterprise and the supervisory data for supervising the enterprise is used as a training sample, and then the labeling of the training sample is determined based on the list of untrustworthy enterprises. Afterwards, the basic data in the training sample is first input into the first prediction layer of the enterprise risk warning model to be trained to obtain an intermediate prediction result. The intermediate prediction result and the supervisory data are then input into the second prediction layer to obtain the risk prediction result. The enterprise risk warning model to be trained is trained with the minimum difference between the risk prediction result and the labeling of the training sample as the training goal. When the trained enterprise risk warning model issues a risk warning to the enterprise, the risk prediction result obtained is more accurate, so that supervisors can accurately intervene in the enterprise early and reduce enterprise risks.

Description

Technical Field

[0001] This specification relates to the field of computer technology, and in particular to a training method, device, medium, and equipment for an enterprise risk early warning model. Background Art

[0002] With the continuous advancement of technology, the application of machine learning models is becoming increasingly widespread. In the public credit sector, enterprise risk warning models can be used to provide risk warnings to enterprises. Based on the predictions of these models, regulators can make precise and early interventions to mitigate risks and ensure the sustainable and healthy development of enterprises. For example, using an enterprise risk warning model to provide a risk warning for a specific enterprise, the enterprise's probability of default can be determined. If the enterprise has repeatedly failed to pay taxes on time, its probability of default may be high. Regulators can alert the enterprise to this risk of default, enabling it to take early action to mitigate risks and maintain sustainable and healthy development.

[0003] However, with the continuous application of enterprise risk warning models and the early intervention of regulators in enterprises, the prediction results of enterprise risk warning models have become increasingly inaccurate.

[0004] Therefore, how to accurately provide risk warnings to enterprises is an urgent problem to be solved. Summary of the Invention

[0005] This specification provides a training method, device, medium and equipment for an enterprise risk early warning model to partially solve the above-mentioned problems existing in the prior art.

[0006] This manual adopts the following technical solutions:

[0007] This specification provides a method for training an enterprise risk early warning model, wherein the enterprise risk early warning model includes a first prediction layer and a second prediction layer; the method includes:

[0008] Determining enterprise data of the enterprise as a training sample based on basic data of the enterprise and regulatory data for supervising the enterprise;

[0009] Determining the labeling of the training samples according to the list of dishonest enterprises;

[0010] Inputting the basic data in the training sample into the first prediction layer of the enterprise risk early warning model to be trained to obtain an intermediate prediction result;

[0011] Inputting the intermediate prediction result and the supervision data into the second prediction layer of the enterprise risk early warning model to be trained to obtain a risk prediction result;

[0012] The enterprise risk warning model to be trained is trained with the minimum difference between the risk prediction result and the labeling of the training sample as the training goal, wherein the trained enterprise risk warning model is used to respond to risk control requests and determine the risk prediction result of the enterprise to be warned based on the enterprise data of the enterprise to be warned.

[0013] Optionally, the method further includes:

[0014] The first prediction layer of the enterprise risk early warning model to be trained is obtained through pre-training, wherein:

[0015] Determining enterprise data of enterprises within a first historical period, using basic data in the enterprise data as training samples, and determining labels for the training samples based on a list of untrustworthy enterprises within the first historical period;

[0016] The training samples are input into the first prediction layer of the enterprise risk warning model to be trained to obtain prediction results, and the first prediction layer of the enterprise risk warning model to be trained is trained with the minimum difference between the prediction results and the labels of the training samples as the training goal.

[0017] Optionally, the enterprise risk warning model to be trained is trained with the minimum difference between the risk prediction result and the annotation of the training sample as the training goal, specifically including:

[0018] The second prediction layer of the enterprise risk early warning model to be trained is trained with the minimum difference between the risk prediction result and the annotation of the training sample as the training goal.

[0019] Optionally, determining the labeling of the training samples based on the list of dishonest enterprises specifically includes:

[0020] For each training sample, determine whether the enterprise corresponding to the training sample is on the list of dishonest enterprises. If so, set the label of the training sample to have dishonest risk; otherwise, set the label of the training sample to have no dishonest risk.

[0021] Optionally, inputting the intermediate prediction result and the supervision data into the second prediction layer of the enterprise risk early warning model to be trained specifically includes:

[0022] Encoding the supervision data in the training sample to obtain a supervision feature vector of the training sample;

[0023] The intermediate prediction result and the supervision feature vector are input into the second prediction layer of the enterprise risk early warning model to be trained.

[0024] Optionally, the supervision data includes at least the number of supervisions, the form of supervision and the time of occurrence;

[0025] Encoding the supervision data in the training sample to obtain the supervision feature vector of the training sample specifically includes:

[0026] According to a preset coding rule, the number of supervisions, the form of supervisions, and the occurrence time in the training samples are respectively encoded to obtain feature vectors, and the obtained feature vectors are spliced;

[0027] The concatenated vectors are normalized to obtain the supervisory feature vector of the training sample.

[0028] Optionally, the original model includes the first prediction layer;

[0029] The method further comprises:

[0030] Determine enterprise data of enterprises within the second historical period as test samples, and determine labels for the test samples based on the list of untrustworthy enterprises within the second historical period;

[0031] Inputting the test sample into the trained enterprise risk early warning model and the original model respectively, obtaining a first test result output by the original model and a second test result output by the trained enterprise risk early warning model;

[0032] According to the first test result, the second test result, and the annotation of the test sample, determining, from the original model and the enterprise risk early warning model, a model whose output test result has the smallest difference with the annotation of the test sample;

[0033] The model with the smallest difference is used to provide risk warning to the enterprise.

[0034] Optionally, the supervision data includes at least a supervision behavior type;

[0035] The method further comprises:

[0036] Determining supervision data in the training sample, and determining each type of supervision behavior based on the supervision data;

[0037] For each type of regulatory behavior included in the regulatory data, determining an intermediate prediction result and a risk prediction result of a training sample corresponding to the type of regulatory behavior;

[0038] Determining the impact of the regulatory behavior type on the enterprise corresponding to the training sample based on the intermediate prediction result and the risk prediction result;

[0039] According to the determined influence of each regulatory behavior type, the regulatory behavior type with the highest influence is used as the main regulatory behavior type of the enterprise corresponding to the training sample.

[0040] Optionally, in response to the risk control request, determining the risk prediction result of the enterprise to be subject to risk warning based on the enterprise data of the enterprise to be subject to risk warning specifically includes:

[0041] Responding to risk control requests, determining the enterprise data of enterprises to be subject to risk warnings;

[0042] Inputting the basic data in the enterprise data into the first prediction layer of the trained enterprise risk warning model to obtain the intermediate prediction results of the enterprise to be warned of risk;

[0043] Inputting the supervision data in the enterprise data and the intermediate prediction results into the second prediction layer of the enterprise risk warning model to obtain the risk prediction results of the enterprise to be warned;

[0044] Based on the risk prediction results, risk warnings are issued to the enterprises to be issued risk warnings.

[0045] This specification provides a training device for an enterprise risk early warning model, wherein the enterprise risk early warning model includes a first prediction layer and a second prediction layer; the device includes:

[0046] A first determination module is configured to determine enterprise data of the enterprise as a training sample based on basic data of the enterprise and supervisory data for supervising the enterprise;

[0047] A second determination module is used to determine the labeling of the training sample based on the list of untrustworthy enterprises;

[0048] A first result module is used to input the basic data in the training sample into the first prediction layer of the enterprise risk early warning model to be trained to obtain an intermediate prediction result;

[0049] A second result module is used to input the intermediate prediction result and the supervision data into the second prediction layer of the enterprise risk early warning model to be trained to obtain a risk prediction result;

[0050] The first training module is used to train the enterprise risk warning model to be trained with the minimum difference between the risk prediction result and the label of the training sample as the training goal, wherein the trained enterprise risk warning model is used to respond to risk control requests and determine the risk prediction result of the enterprise to be warned based on the enterprise data of the enterprise to be warned.

[0051] Optionally, the device further comprises:

[0052] The second training module is used to determine the enterprise data of enterprises within the first historical period, use the basic data in the enterprise data as training samples, and determine the labels of the training samples based on the list of untrustworthy enterprises within the first historical period; input the training samples into the first prediction layer of the enterprise risk warning model to be trained to obtain the prediction results, and train the first prediction layer of the enterprise risk warning model to be trained with the minimum difference between the prediction results and the labels of the training samples as the training goal.

[0053] Optionally, the first training module is specifically configured to train the second prediction layer of the enterprise risk warning model to be trained, taking minimizing the difference between the risk prediction result and the annotation of the training sample as a training goal.

[0054] Optionally, the second determination module is specifically used to determine, for each training sample, whether the enterprise corresponding to the training sample is on the list of untrustworthy enterprises. If so, the label of the training sample is set to have a risk of untrustworthiness; otherwise, the label of the training sample is set to have no risk of untrustworthiness.

[0055] Optionally, the second result module is specifically used to encode the supervision data in the training sample to obtain the supervision feature vector of the training sample; and input the intermediate prediction result and the supervision feature vector into the second prediction layer of the enterprise risk warning model to be trained.

[0056] Optionally, the supervision data includes at least the number of supervisions, the form of supervision and the time of occurrence;

[0057] The second result module is specifically used to encode the number of supervisions, supervision forms and occurrence time in the training sample according to preset coding rules to obtain each feature vector, and splice the obtained feature vectors; normalize the spliced ​​vectors to obtain the supervision feature vector of the training sample.

[0058] Optionally, the original model includes the first prediction layer;

[0059] The device further comprises:

[0060] The testing module is used to determine the enterprise data of enterprises within the second historical period as test samples, and determine the labels of the test samples based on the list of untrustworthy enterprises within the second historical period; input the test samples into the trained enterprise risk warning model and the original model respectively, and obtain the first test result output by the original model and the second test result output by the trained enterprise risk warning model; based on the first test result, the second test result and the labeling of the test samples, determine the model with the smallest difference between the output test result and the labeling of the test sample from the original model and the enterprise risk warning model; and use the model with the smallest difference to issue a risk warning to the enterprise.

[0061] Optionally, the supervision data includes at least a supervision behavior type;

[0062] The device further comprises:

[0063] An analysis module is used to determine the regulatory data in the training sample and determine each regulatory behavior type based on the regulatory data; for each regulatory behavior type included in the regulatory data, determine the intermediate prediction result and risk prediction result of the training sample corresponding to the regulatory behavior type; based on the intermediate prediction result and the risk prediction result, determine the impact of the regulatory behavior type on the enterprise corresponding to the training sample; based on the determined impact of each regulatory behavior type, use the regulatory behavior type with the highest impact as the main regulatory behavior type for the enterprise corresponding to the training sample.

[0064] Optionally, the device further comprises:

[0065] An application module is used to respond to risk control requests and determine the enterprise data of enterprises to be subject to risk warning; input the basic data in the enterprise data into the first prediction layer of the trained enterprise risk warning model to obtain the intermediate prediction results of the enterprises to be subject to risk warning; input the supervision data in the enterprise data and the intermediate prediction results into the second prediction layer of the enterprise risk warning model to obtain the risk prediction results of the enterprises to be subject to risk warning; and issue a risk warning to the enterprises to be subject to risk warning based on the risk prediction results.

[0066] This specification provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the training method of the above-mentioned enterprise risk early warning model.

[0067] This specification provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the training method for the enterprise risk warning model is implemented.

[0068] At least one of the above technical solutions adopted in this specification can achieve the following beneficial effects:

[0069] In the training method of the enterprise risk warning model provided in this specification, the enterprise data of the enterprise can be first determined as a training sample based on the basic data of the enterprise and the regulatory data for supervising the enterprise. Then, the annotation of the training sample is determined based on the list of untrustworthy enterprises. Afterwards, the basic data in the training sample is input into the first prediction layer of the enterprise risk warning model to be trained to obtain an intermediate prediction result, and the intermediate prediction result and the regulatory data are input into the second prediction layer of the enterprise risk warning model to be trained to obtain a risk prediction result. Then, the enterprise risk warning model to be trained is trained with the minimum difference between the risk prediction result and the annotation of the training sample as the training goal, wherein the trained enterprise risk warning model is used to respond to risk control requests and determine the risk prediction result of the enterprise to be risk warned based on the enterprise data of the enterprise to be risk warned.

[0070] It can be seen from the above method that when training the enterprise risk warning model, the present application uses the enterprise data of the enterprise determined based on the basic data of the enterprise and the supervisory data for supervising the enterprise as a training sample, and then determines the labeling of the training sample based on the list of untrustworthy enterprises. Afterwards, the basic data in the training sample is first input into the first prediction layer of the enterprise risk warning model to be trained to obtain an intermediate prediction result. The intermediate prediction result and the supervisory data are then input into the second prediction layer of the enterprise risk warning model to be trained to obtain a risk prediction result. Then, the enterprise risk warning model to be trained is trained with the minimum difference between the risk prediction result and the labeling of the training sample as the training goal. Since supervisory data is added as a training sample when training the enterprise risk warning model, the risk prediction results obtained when the trained enterprise risk warning model issues a risk warning to the enterprise are more accurate, so that supervisors can intervene in the enterprise more accurately and earlier based on the risk prediction results, so that the enterprise can reduce risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] The drawings described herein are used to provide a further understanding of this specification and constitute a part of this specification. The exemplary embodiments and descriptions of this specification are used to explain this specification and do not constitute an improper limitation of this specification. In the drawings:

[0072] Figure 1 A flowchart of a training method for an enterprise risk early warning model provided in this specification;

[0073] Figure 2 A schematic diagram of the structure and training process of an enterprise risk early warning model provided in this manual;

[0074] Figure 3This is a schematic diagram of the structure of a training device for an enterprise risk early warning model provided in this manual;

[0075] Figure 4 This manual provides a corresponding Figure 1 Schematic diagram of the structure of the electronic equipment. DETAILED DESCRIPTION

[0076] To make the objectives, technical solutions, and advantages of this specification more clear, the following will clearly and completely describe the technical solutions of this specification in conjunction with the specific embodiments of this specification and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of this specification.

[0077] The technical solutions provided by the embodiments of this specification are described in detail below with reference to the accompanying drawings.

[0078] Figure 1 This is a flow chart of a method for training an enterprise risk early warning model provided in this specification. The enterprise risk early warning model includes a first prediction layer and a second prediction layer, and specifically includes the following steps:

[0079] S100: Determine the enterprise data of the enterprise as a training sample based on the basic data of the enterprise and the supervisory data for supervising the enterprise.

[0080] When using the enterprise risk warning model to provide risk warning to enterprises, regulators can intervene in the enterprise accurately and early based on the prediction results of the enterprise risk warning model to reduce the enterprise's risks. However, due to the early intervention of regulators in the enterprise, the enterprise regulates its own behavior to reduce its risks. As a result, when using the enterprise risk warning model to provide risk warning to enterprises, the enterprise regulates its own behavior and changes certain characteristics used to determine whether the enterprise has risks, which will affect the prediction results of the enterprise risk warning model and may make the prediction results of the enterprise risk warning model increasingly inaccurate.

[0081] Based on this, the device used to train the enterprise risk warning model determines the enterprise's data as training samples based on the enterprise's basic data and the supervisory data used to regulate the enterprise. The device used to train the enterprise risk warning model can be a server or an electronic device such as a desktop computer or laptop. For ease of description, the following description of the enterprise risk warning model training method provided in this specification uses the server as the execution entity.

[0082] Specifically, the server obtains the basic data of the enterprise and the supervisory data used to supervise the enterprise, identifies the obtained basic data and supervisory data as the enterprise data of the enterprise, and uses them as training samples for the enterprise risk warning model. The enterprise risk warning model can be a model that includes a first prediction layer and a second prediction layer. The basic data of the enterprise can include basic enterprise information, enterprise operating information, enterprise relationships, enterprise judicial arbitration information, and enterprise social evaluation. Supervisory data refers to data used to supervise the enterprise, and can include information such as the number of supervisions on the enterprise, the form of supervision, the person in charge of supervision, the connection method, and the time when the supervision occurred. The form of supervision refers to the type of supervisory behavior, which can include reminders, warnings, interviews, inspections, commitments, and reviews.

[0083] S102: Determine the labeling of the training samples according to the list of untrustworthy enterprises.

[0084] The server can determine the labels for training samples based on the list of untrustworthy enterprises. Enterprises on the list of untrustworthy enterprises are already untrustworthy. Specifically, the server can determine the list of untrustworthy enterprises it has collected and, for each training sample, determine whether the enterprise corresponding to the training sample is on the list. If so, the label for the training sample is set to have a risk of untrustworthiness; otherwise, the label for the training sample is set to have no risk of untrustworthiness. When determining whether the enterprise corresponding to the training sample is on the list of untrustworthy enterprises, the name of the enterprise corresponding to the training sample can be determined, and then a determination can be made whether the name of the enterprise corresponding to the training sample is on the list of untrustworthy enterprises.

[0085] For example, assuming that the enterprises corresponding to each training sample are Party A, Party B, and Party C respectively, for the training samples corresponding to Party A, determine whether Party A is on the list of untrustworthy enterprises. If so, the label of the training sample corresponding to Party A is set to have a default risk. Otherwise, the label of the training sample corresponding to Party A is set to have no default risk.

[0086] S104: Inputting the basic data in the training sample into the first prediction layer of the enterprise risk early warning model to be trained to obtain an intermediate prediction result.

[0087] S106: Input the intermediate prediction result and the supervision data into the second prediction layer of the enterprise risk early warning model to be trained to obtain a risk prediction result.

[0088] S108: The enterprise risk warning model to be trained is trained with the minimum difference between the risk prediction result and the labeling of the training sample as the training goal, wherein the trained enterprise risk warning model is used to respond to risk control requests and determine the risk prediction result of the enterprise to be warned based on the enterprise data of the enterprise to be warned.

[0089] First, the server can input the basic data from the training samples into the first prediction layer of the enterprise risk early warning model to be trained, obtaining an intermediate prediction result output by the first prediction layer of the enterprise risk early warning model to be trained. The server can then input the intermediate prediction result and the supervisory data from the training samples into the second prediction layer of the enterprise risk early warning model to be trained, obtaining a risk prediction result output by the second prediction layer of the enterprise risk early warning model to be trained.

[0090] The server can then train the enterprise risk warning model, with the goal of minimizing the difference between the risk prediction results and the annotations of the training samples. The trained enterprise risk warning model can then be used to respond to risk control requests and determine the risk prediction results for the enterprise to be warned based on the enterprise data of the enterprise to be warned.

[0091] Specifically, the server can input the risk prediction results and the annotations of the training samples into a preset loss function, calculate the loss according to the loss function, determine the gradient that minimizes the loss, and train the model to be trained according to the determined gradient.

[0092] As can be seen from the above method, when training an enterprise risk warning model, the server can identify the enterprise's basic data and supervisory data as the enterprise's enterprise data and use them as training samples. The training samples are then labeled based on the list of dishonest enterprises. Subsequently, the basic data from the training samples is first input into the first prediction layer of the enterprise risk warning model to be trained to obtain an intermediate prediction result. The intermediate prediction result and the supervisory data are then input into the second prediction layer to obtain a risk prediction result. The enterprise risk warning model is then trained with the goal of minimizing the difference between the risk prediction result and the annotations of the training samples. Because supervisory behavior can affect the accuracy of the enterprise risk warning model's prediction results, the impact of supervisory behavior on prediction results is considered when training the enterprise risk warning model. Both the supervisory data and the basic data of the enterprise are used as training samples to train the enterprise risk warning model. This results in more accurate risk predictions when the trained enterprise risk warning model issues risk warnings to the enterprise. This allows supervisors to more accurately intervene in the enterprise earlier based on the risk prediction results to mitigate risk.

[0093] When inputting the intermediate prediction results and supervisory data into the second prediction layer of the enterprise risk early warning model to be trained in step S106, the server may first encode the supervisory data in the training sample to obtain the supervisory feature vector of the training sample. Thereafter, the intermediate prediction results and supervisory feature vector are input into the second prediction layer of the enterprise risk early warning model to be trained.

[0094] When encoding the supervision data in the training sample to obtain the supervision feature vector for the training sample, the supervision data includes at least the number of supervisions, the type of supervision, and the time of occurrence. The server can encode the number of supervisions, the type of supervision, and the time of occurrence in the training sample according to a preset encoding rule to obtain separate feature vectors, and then concatenate the obtained feature vectors. The concatenated vectors are then normalized to obtain the supervision feature vector for the training sample. The preset encoding rule can be set based on the specific information in the supervision data.

[0095] Specifically, first, the server can encode the number of supervisions, supervision forms, and occurrence times in the training samples according to the preset encoding rules to obtain each feature vector. Among them, when encoding the number of supervisions and supervision forms in the supervision data, the number of supervisions corresponding to each supervision form (i.e., supervision behavior type) in the supervision data can be determined first, and then encoding can be performed based on the determined information. For example, Enterprise 1 was supervised once in the form of "warning", so the feature vector obtained after encoding is a vector of (0, 1, 0, 0, 0), and for example, Enterprise 2 was supervised 3 times in the form of "warning" and 1 time in the form of "inspection", so the feature vector obtained after encoding is a vector of (0, 3, 0, 1, 0, 0).

[0096] The above method can also be used to encode other similar data contained in regulatory data, such as the method of communication, existing issues, whether the communication was on schedule, whether feedback was provided, whether commitments were made, whether commitments were fulfilled, and the position of the contact person. Communication methods can include phone calls, emails, meetings, unannounced visits, and factory visits. Existing issues can also include various types of issues, and the contact person's position can include various types, such as clerk, department manager, and general manager. The encoding process for this type of regulatory data is similar to the encoding of the number and type of supervision in the regulatory data described above. For example, if the communication method for enterprise 1 is phone calls, the resulting vector after encoding is (1, 0, 0, 0, 0), corresponding to (phone calls, emails, meetings, unannounced visits, factory visits), where 1 indicates that the communication method was used and 0 indicates that the communication method was not used. Binary information such as whether the communication was on schedule, whether feedback was provided, whether commitments were made, and whether commitments were fulfilled can be represented by 0 and 1, with 1 representing yes and 0 representing no. For example, if enterprise 1 communicated on schedule, this information would be encoded as 1.

[0097] When encoding time-related information in regulatory data, the time-related information can be encoded based on the difference between the time-related information and the predicted time. Time-related information can include information such as the occurrence date and commitment period, and the predicted time is the time when the risk warning is issued to the enterprise. When training an enterprise risk warning model, the predicted time can be the time determined by the annotation of the training sample. For example, when training an enterprise risk warning model, enterprise data between January 1st and March 1st is used as training samples. The annotation of the training sample can be the time when the list of untrustworthy enterprises is collected between January 1st and March 1st, and the predicted time can be the last time point when the training sample annotation can be determined, that is, March 1st. When using the enterprise risk warning model to issue a risk warning to the enterprise, the predicted time is the time when the risk warning is issued to the enterprise. For example, if the enterprise risk warning model is planned to issue a risk warning to the enterprise on March 1st, the predicted time is March 1st.

[0098] Continuing with the previous example, when encoding the occurrence date in the regulatory data of Enterprise 1, the occurrence date can be encoded based on the difference between the occurrence date and the predicted time. Assuming that the date of Enterprise 1’s most recent regulatory action is 2022-07-15, and the model predicts the time is 2022-09-01, the calculated difference is (47).

[0099] Afterwards, the server can splice the obtained feature vectors in a preset order, perform dimensionality reduction and normalization on the spliced ​​vectors, and obtain the supervisory feature vector of the training sample. Among them, the preset order can be an order arranged according to a certain set rule, or an order of arbitrary random arrangement, which is not specifically limited in this specification. Dimensionality reduction can be performed on the spliced ​​vectors according to the annotation of the training sample, or by any current means, which is not specifically limited in this specification. Continuing with the above example, the (0,1,0,0,0,0) feature vector corresponding to enterprise 1 and the feature vector (47) of the occurrence date obtained above are spliced ​​to obtain a spliced ​​vector of (0,1,0,0,0,0,47). According to the annotation of the training sample corresponding to enterprise 1, the spliced ​​vector is subjected to dimensionality reduction and normalization, and (0.12,0.38,0.69) is obtained as the supervisory feature vector of the training sample corresponding to enterprise 1.

[0100] When the intermediate prediction results and the regulatory feature vector are input into the second prediction layer of the enterprise risk warning model to be trained, the server can concatenate the intermediate prediction results and the regulatory feature vector, and then input the concatenated vector into the enterprise risk warning model to be trained to obtain the second prediction layer. Among them, the intermediate prediction result can be the probability of enterprise default. Continuing with the above example, assuming that the basic data corresponding to enterprise 1 is input into the first prediction layer of the enterprise risk warning model to be trained, the default probability obtained is 0.86. (0.86) can be concatenated with the regulatory feature vector (0.12, 0.38, 0.69) obtained in the above example to obtain (0.86, 0.12, 0.38, 0.69). The obtained (0.86, 0.12, 0.38, 0.69) is input into the enterprise risk warning model to be trained to obtain the second prediction layer.

[0101] Because regulators haven't yet overseen enterprises extensively in the early stages, resulting in insufficient regulatory data or the lack of predictive features when enterprises regulate their own behavior, a model can be pre-trained using the enterprise's basic data to predict whether the enterprise is risky. This is referred to as the first prediction layer of the enterprise risk warning model to be trained in this specification. Therefore, in this specification, the server can determine the enterprise data for the first historical period, use the basic data within the enterprise data as training samples, and determine the labels for the training samples based on the list of untrustworthy enterprises in the first historical period. The training samples are then input into the first prediction layer of the enterprise risk warning model to be trained, obtaining prediction results. The first prediction layer of the enterprise risk warning model to be trained is trained with the goal of minimizing the difference between the prediction results and the labels of the training samples. The first prediction layer of the enterprise risk warning model to be trained can be pre-trained. The first historical period can be a preset time period, such as three months, or a period containing a preset amount of data, meaning a period in which a preset amount of data can be determined, such as a period containing 10,000 pieces of data. Other rules can also be used, and this specification does not provide specific limitations.

[0102] Based on this, since the first prediction layer of the enterprise risk warning model to be trained has been pre-trained, in the above step S108, the training goal is to minimize the difference between the risk prediction results and the labels of the training samples. When training the enterprise risk warning model to be trained, the server can use the minimum difference between the risk prediction results and the labels of the training samples as the training goal and only train the second prediction layer of the enterprise risk warning model to be trained.

[0103] Before applying a trained enterprise risk warning model, it is usually necessary to test it. Furthermore, since the trained enterprise risk warning model is trained using regulatory data on the enterprise, its predictive performance may be inferior to that of a model trained without such data. Therefore, the server needs to test which of the two models has better predictive performance and use the model with the better predictive performance for subsequent enterprise risk warnings.

[0104] Therefore, the server can determine the enterprise data of enterprises within the second historical period as test samples and, based on the list of untrustworthy enterprises within the second historical period, determine the labels for the test samples. The test samples are then input into the trained enterprise risk warning model and the original model, respectively, to obtain a first test result output by the original model and a second test result output by the trained enterprise risk warning model. Then, based on the first test result, the second test result, and the labels of the test samples, the server determines the model from the original model and the enterprise risk warning model that minimizes the difference between the test results it outputs and the labels of the test samples. The model with the smallest difference is then used to issue a risk warning for the enterprise. The original model includes a first prediction layer. The second historical period is similar to the first historical period described above and can be a preset time period, a time period containing a preset amount of data, or any other set of rules. This specification does not impose specific limitations. However, the second historical period is not the same time period or time period as the first historical period.

[0105] Since the prediction effect of the original model may be better than the prediction effect of the enterprise risk warning model (i.e., the model trained with regulatory data as training samples) in a certain period, but the prediction effect of the original model may not be better than the prediction effect of the enterprise risk warning model (i.e., the model trained with regulatory data as training samples) in another period, at the same time, the enterprise data of the enterprise in each period is also different, and the enterprise data is constantly updated and changed, so in order to ensure the accuracy of the prediction results of the model for risk warning of the enterprise, to be able to more accurately provide risk warning to the enterprise and provide the correct supervision basis for the regulators, the server needs to periodically train the enterprise risk warning model, that is, to periodically execute the above steps S100 to S108, and then determine which model has better training effect between the model trained with regulatory data as training samples and the model trained without regulatory data as training samples (i.e., the original model), and determine the model with better prediction effect to provide risk warning to the enterprise.

[0106] In this specification, since the intermediate prediction results are the prediction results obtained without adding regulatory data, and the risk prediction results are the prediction results obtained by adding regulatory data, it is possible to determine which type of regulatory behavior or which types of regulatory behavior have a more significant impact on the enterprise based on the intermediate prediction results and risk prediction results of the enterprise risk warning model, that is, to determine which type of regulatory behavior or which types of regulatory behavior have a better regulatory effect on the enterprise, so that supervisors can use this type of regulatory behavior to supervise the enterprise in the subsequent process and improve supervision efficiency. At the same time, it provides a reference for supervisors to specify supervision strategies and gives supervisors certain guidance.

[0107] Therefore, the server can determine the regulatory data in the training sample and, based on the regulatory data, identify each regulatory action type. Next, for each regulatory action type included in the regulatory data, the server determines the intermediate prediction results and risk prediction results for the training sample corresponding to that regulatory action type. Based on these intermediate prediction results and risk prediction results, the server determines the impact of that regulatory action type on the enterprise corresponding to the training sample. Then, based on the determined impact of each regulatory action type, the regulatory action type with the highest impact is selected as the primary regulatory action type for the enterprise corresponding to the training sample.

[0108] When determining the impact of a regulatory action type on the enterprise corresponding to the training sample based on the intermediate prediction results and the risk prediction results, the difference between the intermediate prediction results and the risk prediction results can be used as the impact of the regulatory action type on the enterprise corresponding to the training sample. Since a regulatory action type may correspond to multiple training samples, resulting in multiple differences, the average of these differences is used as the impact of the regulatory action type on the enterprise corresponding to the training sample.

[0109] In this specification, to more accurately target each type of enterprise with the most effective regulatory actions, the server can also bin the training samples corresponding to the enterprises according to a certain rule, with the training samples in each bin corresponding to the relevant type of enterprise. Then, for each training sample in each bin, that is, for each type of enterprise, the server determines which type of regulatory action, or which types of regulatory actions, are most effective for that type of enterprise.

[0110] Specifically, the server can also classify the training samples according to preset rules to obtain various types of training samples. For each type of training sample, the server determines the regulatory data within that type of training sample and, based on the regulatory data, determines the various types of regulatory behavior. Subsequently, for each type of regulatory behavior included in the regulatory data, the server determines the intermediate prediction results and risk prediction results for the training samples corresponding to that type of regulatory behavior within that type of training sample. Based on the intermediate prediction results and risk prediction results, the server determines the impact of that type of regulatory behavior on the enterprise corresponding to that type of training sample. Subsequently, based on the determined impact of each type of regulatory behavior, the type of regulatory behavior with the highest impact is selected as the primary regulatory behavior type for the enterprise corresponding to that type of training sample.

[0111] In step S100, the server may determine the enterprise data of the enterprise as a training sample based on the basic data of the enterprise and the supervisory data for supervising the enterprise within a historical period. The historical period may be a preset time period, such as three months, or a period containing a preset amount of data, or may be set according to any other rule, which is not specifically limited in this specification.

[0112] After determining the enterprise data in step S100, the server can use the determined enterprise data as a sample. Then, based on the list of untrustworthy enterprises, the server determines the labels corresponding to the sample. Based on the labels corresponding to the sample, the server then performs stratified random sampling. The sampled samples are then divided into training samples and test samples according to a preset ratio, for example, a ratio of 8:2.

[0113] In this manual, in order to incorporate the supervisory behavior of regulators into the enterprise risk warning model, it is necessary to pre-build supervisory behavior norms and recording mechanisms, that is, to sort out and standardize the supervisory methods used by regulators on risky enterprises, and finally build enterprise public credit supervision record collection norms to guide regulators to do a good job in collecting supervision records and improve the work efficiency of regulators. At the same time, it is also convenient for obtaining supervision feature vectors based on supervision data when training the enterprise risk warning model in the future.

[0114] The standards for collecting enterprise public credit supervision records include information such as the collection method, content, operation examples (i.e., filling in, collaborating, viewing, and modifying), storage method, and storage address. Record content includes information such as the form of supervision, supervisor, subject, connection method, timeframe, content, and feedback.

[0115] The training process and model structure of the enterprise risk early warning model in this manual are as follows: Figure 2 As shown, Figure 2This is a schematic diagram illustrating the structure and training process of an enterprise risk early warning model provided in this specification. The server first inputs basic data into the first prediction layer of the enterprise risk early warning model to be trained, obtaining an intermediate prediction result output by the first prediction layer. Subsequently, the intermediate prediction result output by the first prediction layer is used, along with supervisory data, to input the intermediate prediction result into the second prediction layer of the enterprise risk early warning model to obtain a risk prediction result.

[0116] The above are one or more implementation methods of this specification. Based on the same idea, this specification also provides a corresponding enterprise risk warning model training device, such as Figure 3 shown.

[0117] Figure 3 This is a schematic diagram of a training device for an enterprise risk early warning model provided in this specification, wherein the enterprise risk early warning model includes a first prediction layer and a second prediction layer; the device includes:

[0118] A first determination module 200 is configured to determine enterprise data of an enterprise as a training sample based on the basic data of the enterprise and the supervisory data for supervising the enterprise;

[0119] The second determining module 202 is used to determine the label of the training sample according to the list of untrustworthy enterprises;

[0120] The first result module 204 is used to input the basic data in the training sample into the first prediction layer of the enterprise risk early warning model to be trained to obtain an intermediate prediction result;

[0121] A second result module 206 is configured to input the intermediate prediction result and the supervision data into the second prediction layer of the enterprise risk early warning model to be trained to obtain a risk prediction result;

[0122] The first training module 208 is used to train the enterprise risk warning model to be trained with the minimum difference between the risk prediction result and the label of the training sample as the training goal, wherein the trained enterprise risk warning model is used to respond to risk control requests and determine the risk prediction result of the enterprise to be warned based on the enterprise data of the enterprise to be warned.

[0123] Optionally, the device further comprises:

[0124] The second training module 210 is used to determine the enterprise data of enterprises within the first historical period, use the basic data in the enterprise data as training samples, and determine the labels of the training samples based on the list of untrustworthy enterprises within the first historical period; input the training samples into the first prediction layer of the enterprise risk warning model to be trained to obtain the prediction results, and train the first prediction layer of the enterprise risk warning model to be trained with the minimum difference between the prediction results and the labels of the training samples as the training goal.

[0125] Optionally, the first training module 208 is specifically configured to train the second prediction layer of the enterprise risk warning model to be trained, taking minimizing the difference between the risk prediction result and the annotation of the training sample as a training goal.

[0126] Optionally, the second determination module 202 is specifically used to determine, for each training sample, whether the enterprise corresponding to the training sample is on the list of untrustworthy enterprises. If so, the label of the training sample is set to have a risk of untrustworthiness; otherwise, the label of the training sample is set to have no risk of untrustworthiness.

[0127] Optionally, the second result module 206 is specifically used to encode the supervision data in the training sample to obtain the supervision feature vector of the training sample; and input the intermediate prediction result and the supervision feature vector into the second prediction layer of the enterprise risk warning model to be trained.

[0128] Optionally, the supervision data includes at least the number of supervisions, the form of supervision and the time of occurrence;

[0129] The second result module 206 is specifically used to encode the number of supervisions, supervision forms and occurrence time in the training sample according to a preset coding rule to obtain each feature vector, and splice the obtained feature vectors; normalize the spliced ​​vectors to obtain the supervision feature vector of the training sample.

[0130] Optionally, the original model includes the first prediction layer;

[0131] The device further comprises:

[0132] The testing module 212 is used to determine the enterprise data of enterprises within the second historical period as test samples, and determine the labels of the test samples based on the list of untrustworthy enterprises within the second historical period; input the test samples into the trained enterprise risk warning model and the original model respectively, and obtain the first test result output by the original model and the second test result output by the trained enterprise risk warning model; based on the first test result, the second test result and the labels of the test samples, determine the model with the smallest difference between the output test result and the label of the test sample from the original model and the enterprise risk warning model; and use the model with the smallest difference to issue a risk warning to the enterprise.

[0133] Optionally, the supervision data includes at least a supervision behavior type;

[0134] The device further comprises:

[0135] The analysis module 216 is used to determine the regulatory data in the training sample and determine each regulatory behavior type based on the regulatory data; for each regulatory behavior type included in the regulatory data, determine the intermediate prediction result and risk prediction result of the training sample corresponding to the regulatory behavior type; based on the intermediate prediction result and the risk prediction result, determine the impact of the regulatory behavior type on the enterprise corresponding to the training sample; based on the determined impact of each regulatory behavior type, use the regulatory behavior type with the highest impact as the main regulatory behavior type for the enterprise corresponding to the training sample.

[0136] Optionally, the device further comprises:

[0137] Application module 214 is used to determine the enterprise data of the enterprise to be warned of risks in response to risk control requests; input the basic data in the enterprise data into the first prediction layer of the trained enterprise risk warning model to obtain the intermediate prediction result of the enterprise to be warned of risks; input the supervision data in the enterprise data and the intermediate prediction result into the second prediction layer of the enterprise risk warning model to obtain the risk prediction result of the enterprise to be warned of risks; and issue a risk warning to the enterprise to be warned of risks based on the risk prediction result.

[0138] This specification also provides a computer-readable storage medium, which stores a computer program that can be used to execute the above Figure 1 A training method for an enterprise risk early warning model is provided.

[0139] This manual also provides Figure 4 The one shown corresponds to Figure 1 Schematic diagram of the electronic equipment. Figure 4As shown, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1 The training method of the enterprise risk early warning model.

[0140] Of course, in addition to software implementation, this specification does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc. That is to say, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0141] In the 1990s, technological improvements could be clearly distinguished as either hardware improvements (for example, improvements to circuit structures like diodes, transistors, and switches) or software improvements (improvements to process flows). However, with the advancement of technology, many process flow improvements today can now be considered direct improvements to hardware circuit structures. Designers almost always create the corresponding hardware circuit structure by programming the improved process flow into the hardware circuit. Therefore, it cannot be said that a process flow improvement cannot be implemented using hardware modules. For example, a programmable logic device (PLD), such as a field programmable gate array (FPGA), is an integrated circuit whose logical function is determined by user programming. Designers can "integrate" a digital system on a PLD through their own programming, without having to hire a chip manufacturer to design and manufacture a dedicated integrated circuit chip. Moreover, nowadays, instead of manually fabricating integrated circuit chips, this programming is mostly done using "logic compiler" software. This is similar to the software compiler used when developing programs. Before compilation, the original code must also be written in a specific programming language, called a hardware description language (HDL). There is not just one HDL, but many, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art will also understand that by simply programming the method flow in one of these hardware description languages ​​and then programming it into an integrated circuit, a hardware circuit that implements the logic method flow can be easily obtained.

[0142] The controller can be implemented in any suitable manner. For example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also know that in addition to implementing the controller in a purely computer-readable program code format, the controller can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, such a controller can be considered a hardware component, and the devices included therein for implementing various functions can also be considered as structures within the hardware component. Or even, the devices for implementing various functions can be considered as both software modules that implement the method and structures within the hardware component.

[0143] The systems, devices, modules, or units described in the above embodiments may be implemented by computer chips or entities, or by products having certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0144] For the convenience of description, the above devices are described as being divided into various units according to their functions. Of course, when implementing this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0145] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0146] This specification is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0147] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0148] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0149] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0150] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0151] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules 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 technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0152] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0153] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Thus, this specification may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0154] This specification may be described in the general context of computer-executable instructions, such as program modules, executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform specific tasks or implement specific abstract data types. This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media, including storage devices.

[0155] The various embodiments in this specification are described in a progressive manner. Similar parts between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the system embodiments are generally similar to the method embodiments, so the description is relatively simple. For relevant parts, refer to the description of the method embodiments.

[0156] The foregoing is merely an example of the present invention and is not intended to limit the present invention. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be included within the scope of the claims of the present invention.

Claims

1. A training method for an enterprise risk early warning model, characterized in that: The enterprise risk early warning model includes a first prediction layer and a second prediction layer; the method includes: Determining enterprise data of the enterprise as a training sample based on basic data of the enterprise and regulatory data for supervising the enterprise; Determining the labeling of the training samples according to the list of dishonest enterprises; Inputting the basic data in the training sample into the first prediction layer of the enterprise risk early warning model to be trained to obtain an intermediate prediction result; Inputting the intermediate prediction result and the regulatory data into the second prediction layer of the enterprise risk early warning model to be trained to obtain a risk prediction result; wherein, according to the regulatory rules corresponding to different types of regulatory data, each type of regulatory data is encoded to obtain a regulatory feature vector, and the regulatory feature vector is spliced ​​with the intermediate prediction result and then input into the second prediction layer; the different types of regulatory data include at least: the number of regulatory times, the form of regulatory action, and the time of occurrence; The enterprise risk warning model to be trained is trained with the minimum difference between the risk prediction result and the labeling of the training sample as the training goal, wherein the trained enterprise risk warning model is used to respond to risk control requests and determine the risk prediction result of the enterprise to be warned based on the enterprise data of the enterprise to be warned.

2. The method according to claim 1, wherein The method further comprises: The first prediction layer of the enterprise risk early warning model to be trained is obtained through pre-training, wherein: Determining enterprise data of enterprises within a first historical period, using basic data in the enterprise data as training samples, and determining labels for the training samples based on a list of untrustworthy enterprises within the first historical period; The training samples are input into the first prediction layer of the enterprise risk warning model to be trained to obtain prediction results, and the first prediction layer of the enterprise risk warning model to be trained is trained with the minimum difference between the prediction results and the labels of the training samples as the training goal.

3. The method according to claim 2, wherein The enterprise risk warning model to be trained is trained with the minimum difference between the risk prediction result and the annotation of the training sample as the training goal, specifically including: The second prediction layer of the enterprise risk early warning model to be trained is trained with the minimum difference between the risk prediction result and the annotation of the training sample as the training goal.

4. The method according to claim 1, wherein According to the list of dishonest enterprises, the labeling of the training samples is determined, specifically including: For each training sample, determine whether the enterprise corresponding to the training sample is on the list of dishonest enterprises. If so, set the label of the training sample to have dishonest risk; otherwise, set the label of the training sample to have no dishonest risk.

5. The method according to claim 1, wherein The original model includes the first prediction layer; The method further comprises: Determine enterprise data of enterprises within the second historical period as test samples, and determine labels for the test samples based on the list of untrustworthy enterprises within the second historical period; Inputting the test sample into the trained enterprise risk early warning model and the original model respectively, obtaining a first test result output by the original model and a second test result output by the trained enterprise risk early warning model; According to the first test result, the second test result, and the annotation of the test sample, determining, from the original model and the enterprise risk early warning model, a model whose output test result has the smallest difference with the annotation of the test sample; The model with the smallest difference is used to provide risk warning to the enterprise.

6. The method according to claim 1, wherein The regulatory data at least includes the type of regulatory behavior; The method further comprises: Determining supervision data in the training sample, and determining each type of supervision behavior based on the supervision data; For each type of regulatory behavior included in the regulatory data, determining an intermediate prediction result and a risk prediction result of a training sample corresponding to the type of regulatory behavior; Determining the impact of the regulatory behavior type on the enterprise corresponding to the training sample based on the intermediate prediction result and the risk prediction result; According to the determined influence of each regulatory behavior type, the regulatory behavior type with the highest influence is used as the main regulatory behavior type of the enterprise corresponding to the training sample.

7. The method according to claim 1, wherein In response to the risk control request, the risk prediction result of the enterprise to be warned is determined based on the enterprise data of the enterprise to be warned, specifically including: Responding to risk control requests, determining the enterprise data of enterprises to be subject to risk warnings; Inputting the basic data in the enterprise data into the first prediction layer of the trained enterprise risk warning model to obtain the intermediate prediction results of the enterprise to be warned of risk; Inputting the supervision data in the enterprise data and the intermediate prediction results into the second prediction layer of the enterprise risk warning model to obtain the risk prediction results of the enterprise to be warned; Based on the risk prediction results, risk warnings are issued to the enterprises to be issued risk warnings.

8. A training device for an enterprise risk warning model, characterized in that: The enterprise risk early warning model includes a first prediction layer and a second prediction layer; the device includes: A first determination module is configured to determine enterprise data of the enterprise as a training sample based on basic data of the enterprise and supervisory data for supervising the enterprise; A second determination module is used to determine the labeling of the training sample based on the list of untrustworthy enterprises; A first result module is configured to input the basic data in the training sample into the first prediction layer of the enterprise risk early warning model to be trained to obtain an intermediate prediction result; wherein, according to the supervision rules corresponding to different types of supervision data, each type of supervision data is encoded to obtain a supervision feature vector, and the supervision feature vector is concatenated with the intermediate prediction result and then input into the second prediction layer; the different types of supervision data include at least: the number of supervisions, the form of supervision, and the time of occurrence; A second result module is used to input the intermediate prediction result and the supervision data into the second prediction layer of the enterprise risk early warning model to be trained to obtain a risk prediction result; The first training module is used to train the enterprise risk warning model to be trained with the minimum difference between the risk prediction result and the label of the training sample as the training goal, wherein the trained enterprise risk warning model is used to respond to risk control requests and determine the risk prediction result of the enterprise to be warned based on the enterprise data of the enterprise to be warned.

9. The device according to claim 8, wherein The device further comprises: The second training module is used to determine the enterprise data of enterprises within the first historical period, use the basic data in the enterprise data as training samples, and determine the labels of the training samples based on the list of untrustworthy enterprises within the first historical period; input the training samples into the first prediction layer of the enterprise risk warning model to be trained to obtain the prediction results, and train the first prediction layer of the enterprise risk warning model to be trained with the minimum difference between the prediction results and the labels of the training samples as the training goal.

10. The device according to claim 9, characterized in that The first training module is specifically used to train the second prediction layer of the enterprise risk warning model to be trained, with the difference between the risk prediction result and the label of the training sample being minimized as a training goal.

11. The device according to claim 8, wherein The second determination module is specifically used to determine, for each training sample, whether the enterprise corresponding to the training sample is on the list of untrustworthy enterprises. If so, the label of the training sample is set to have untrustworthy risk; otherwise, the label of the training sample is set to have no untrustworthy risk.

12. The device according to claim 8, wherein The original model includes the first prediction layer; The device further comprises: The testing module is used to determine the enterprise data of enterprises within the second historical period as test samples, and determine the labels of the test samples based on the list of untrustworthy enterprises within the second historical period; input the test samples into the trained enterprise risk warning model and the original model respectively, and obtain the first test result output by the original model and the second test result output by the trained enterprise risk warning model; based on the first test result, the second test result and the labeling of the test samples, determine the model with the smallest difference between the output test result and the labeling of the test sample from the original model and the enterprise risk warning model; and use the model with the smallest difference to issue a risk warning to the enterprise.

13. The device according to claim 8, wherein The regulatory data at least includes the type of regulatory behavior; The device further comprises: An analysis module is used to determine the regulatory data in the training sample and determine each regulatory behavior type based on the regulatory data; for each regulatory behavior type included in the regulatory data, determine the intermediate prediction result and risk prediction result of the training sample corresponding to the regulatory behavior type; based on the intermediate prediction result and the risk prediction result, determine the impact of the regulatory behavior type on the enterprise corresponding to the training sample; based on the determined impact of each regulatory behavior type, use the regulatory behavior type with the highest impact as the main regulatory behavior type for the enterprise corresponding to the training sample.

14. The device according to claim 8, wherein The device further comprises: An application module is used to respond to risk control requests and determine the enterprise data of enterprises to be subject to risk warning; input the basic data in the enterprise data into the first prediction layer of the trained enterprise risk warning model to obtain the intermediate prediction results of the enterprises to be subject to risk warning; input the supervision data in the enterprise data and the intermediate prediction results into the second prediction layer of the enterprise risk warning model to obtain the risk prediction results of the enterprises to be subject to risk warning; and issue a risk warning to the enterprises to be subject to risk warning based on the risk prediction results.

15. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

16. An electronic 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 program, the method according to any one of claims 1 to 7 is implemented.

Images