Abnormality level acquisition method and device based on error correction, medium and equipment

Abstract

The invention relates to the field of data processing, in particular to an error correction-based anomaly level acquisition method and device, a medium and equipment, and the method comprises the steps: determining a first prediction error of a reference set and a correlation level between the reference set according to a credit anomaly prediction error and a correlation level between a to-be-recognized subject, and when the reference set meets a second preset condition, adding the to-be-identified main body in the reference set into the temporary set, deleting the reference set, carrying out iterative combination on all the reference sets to obtain a target set, and combining a correlation level and an intermediate prediction error between the target sets to obtain the target set. And iteratively optimizing the first prediction error of the target set into the second prediction error, further determining a third prediction error corresponding to the to-be-identified subject, correcting the corresponding credit anomaly target probability according to the third prediction error, obtaining a credit anomaly correction probability with higher accuracy, and mapping to obtain a corresponding credit anomaly level. And the acquisition accuracy of the credit abnormality level is improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a method, device, medium and equipment for acquiring anomaly levels based on error correction. Background Art

[0002] The credit anomaly rating assessment of the subject can identify, qualitatively and quantitatively analyze the subject's potential abnormal factors, help organizations and individuals understand the various risks they may face, and better avoid and control abnormalities. It can also help organizations and individuals sort and prioritize credit anomalies and rationally allocate limited resources, which is of great significance in many fields such as finance and investment, corporate management, medical and health, information technology and network security. For example, in the field of corporate investment, asset management companies need to conduct credit anomaly assessments on financing entities when investing in them, establish an investable enterprise pool for asset management companies based on the assessment results of credit anomaly ratings, calculate the investable quota of financing companies, and determine credit risk pressure.

[0003] In the existing credit anomaly registration and assessment methods, the abnormal probability of the subject in each credit dimension is mainly measured, and the importance weights corresponding to each credit dimension are combined to obtain the overall credit anomaly level of the subject by integrating the abnormal probability of multiple credit dimensions. However, the above method ignores the impact of the relationship between subjects on the subject abnormality, which makes it difficult to reasonably assess the credit anomaly level of the subject. For example, in the field of corporate investment, the current credit anomaly level assessment method focuses on the financial indicators, industry status, market value and other risk factors of the financing subject, and does not consider the credit impact of related subjects such as the industrial chain supply chain network, investment and financing network, and guarantee circle where the financing subject is located on the financing subject, which makes it difficult to quantitatively calculate and evaluate the credit impact of related subjects on the financing subject, and thus it is difficult to meet the requirements of asset management companies for accurate risk assessment of financing subjects.

[0004] Therefore, when obtaining the credit anomaly rating of an entity, how to improve the accuracy of the credit anomaly rating becomes an urgent problem to be solved. Summary of the invention

[0005] In view of the above technical problems, the technical solution adopted by the present invention is a method for obtaining an abnormality level based on error correction, and the method for obtaining an abnormality level based on error correction comprises the following steps:

[0006] S100, obtaining N entities to be identified, the credit anomaly target probability and credit anomaly prediction error corresponding to each entity to be identified, and the correlation level between the entities to be identified, where N is an integer greater than 1.

[0007] S200, respectively assigning a corresponding reference set to each subject to be identified.

[0008] S300, determining the correlation level between the first prediction error corresponding to each reference set and the reference set according to the correlation level between the credit anomaly prediction error corresponding to each subject to be identified and the subject to be identified.

[0009] S400: randomly select a reference set from all reference sets as a temporary set.

[0010] S500: For any reference set, if the current reference set meets the second preset condition, add the subject to be identified in the current reference set to the temporary set, delete the current reference set, and use the temporary set as the reference set.

[0011] S600, returning to step S300, until the number of reference sets converges, Q reference sets are obtained, and the Q reference sets are determined as Q target sets.

[0012] S700, updating the first prediction error corresponding to each target set according to the first prediction error corresponding to each target set and the correlation level between the target sets, to obtain the second prediction error corresponding to each target set.

[0013] S800, for any subject to be identified, determining a third prediction error corresponding to the current subject to be identified according to a second prediction error corresponding to a target set to which the current subject to be identified belongs.

[0014] S900, obtaining the credit anomaly level corresponding to each subject to be identified according to the third prediction error and the credit anomaly target probability corresponding to each subject to be identified.

[0015] The present invention also provides an abnormality level acquisition device based on error correction, and the abnormality level acquisition device based on error correction includes:

[0016] The data acquisition module is used to obtain N subjects to be identified, the credit anomaly target probability and credit anomaly prediction error corresponding to each subject to be identified, and the correlation level between the subjects to be identified, where N is an integer greater than 1.

[0017] The reference set allocation module is used to allocate a corresponding reference set to each subject to be identified.

[0018] The correlation level acquisition module is used to determine the correlation level between the first prediction error corresponding to each reference set and the reference set according to the correlation level between the credit anomaly prediction error corresponding to each subject to be identified and the subject to be identified.

[0019] The temporary set selection module is used to randomly select a reference set from all reference sets as a temporary set.

[0020] The reference set merging module is used to add the subject to be identified in the current reference set to the temporary set for any reference set, if the current reference set meets the second preset condition, and delete the current reference set to use the temporary set as the reference set.

[0021] The target set acquisition module is used to return to execute the relevant level acquisition module until the number of reference sets converges to obtain Q reference sets, and determine the Q reference sets as Q target sets.

[0022] The prediction error updating module is used to update the first prediction error corresponding to each target set according to the first prediction error corresponding to each target set and the correlation level between the target sets to obtain the second prediction error corresponding to each target set.

[0023] The third prediction error acquisition module is used to determine, for any subject to be identified, a third prediction error corresponding to the current subject to be identified according to the second prediction error corresponding to the target set to which the current subject to be identified belongs.

[0024] The credit anomaly level acquisition module is used to acquire the credit anomaly level corresponding to each subject to be identified based on the third prediction error and credit anomaly target probability corresponding to each subject to be identified.

[0025] The present invention also provides a non-transitory computer-readable storage medium, in which at least one instruction or at least one program is stored, and the at least one instruction or at least one program is loaded and executed by a processor to implement the above-mentioned error correction-based abnormality level acquisition method.

[0026] The present invention also provides an electronic device, comprising a processor and the above-mentioned non-transitory computer-readable storage medium.

[0027] The present invention has at least the following beneficial effects: according to the correlation level between the credit anomaly prediction error corresponding to each subject to be identified and the subject to be identified, the correlation level between the first prediction error corresponding to each reference set and the reference set is determined to characterize the overall error situation of all subjects to be identified in each reference set, as well as the closeness of the correlation relationship between any two reference sets; according to the correlation level between the current reference set and the temporary set and the difference between the first prediction error, the merging result between the reference set and the temporary set is judged, and when the current reference set meets the second preset condition, the subject to be identified in the current reference set is added to the temporary set, and the current reference set is deleted, and then all reference sets are iteratively merged to obtain Q target sets. The set reduces the number of sets and improves the efficiency of error correction; further combines the correlation level and intermediate prediction error between target sets to analyze the error impact between target sets, and iteratively optimizes the first prediction error of all target sets to obtain the second prediction error corresponding to each target set, thereby improving the accuracy of the second prediction error; for any subject to be identified, the third prediction error corresponding to the current subject to be identified is determined according to the second prediction error corresponding to the target set to which the current subject to be identified belongs, and then the corresponding credit anomaly target probability is corrected according to the third prediction error to obtain a more accurate credit anomaly correction probability, which is further mapped to obtain the corresponding credit anomaly level, thereby improving the accuracy of obtaining the credit anomaly level. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0029] Figure 1 A flowchart of a method for obtaining an abnormality level based on error correction provided in Embodiment 1 of the present invention;

[0030] Figure 2 A flowchart of a method for obtaining an abnormal credit level provided in the second embodiment of the present invention;

[0031] Figure 3 A schematic diagram of the structure of an abnormality level acquisition device based on error correction provided in Embodiment 3 of the present invention. DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0033] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It is understood that, where appropriate, the above-mentioned terms used to distinguish similar objects can be interchanged so that the present invention can also implement other embodiments other than the above-mentioned illustrated embodiments or described embodiments. In addition, the terms "including" and "having" and any variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0034] Embodiment 1

[0035] The first embodiment provides a method for obtaining an abnormality level based on error correction. The method for obtaining an abnormality level based on error correction includes the following steps: Figure 1 As shown:

[0036] S100, obtaining N entities to be identified, the credit anomaly target probability and credit anomaly prediction error corresponding to each entity to be identified, and the correlation level between the entities to be identified, where N is an integer greater than 1.

[0037] Among them, the subject to be identified refers to the subject that needs to be analyzed and evaluated for credit anomaly level. By obtaining the credit anomaly level corresponding to the subject to be identified, it can help organizations and individuals understand the various risks they may face, so as to better avoid and control anomalies. It can also help organizations and individuals sort and prioritize credit anomalies and reasonably allocate limited resources. For example, in the field of enterprise management, the subject to be identified can refer to an enterprise that needs to conduct an operating risk assessment. In the field of medical care and health, the subject to be identified can refer to patients or hospitals that need to conduct risk assessments. In the field of information technology and network security, the subject to be identified can refer to equipment or information for security assessments.

[0038] This embodiment is explained using the financial and investment fields as an example. The subject to be identified may refer to an enterprise that needs financing. When investing in a financing subject, the asset management company needs to conduct a credit anomaly assessment on the financing subject and establish an investable enterprise pool for the asset management company based on the assessment results of the credit anomaly level, so as to calculate the investable amount of the financing enterprise, determine the credit risk pressure, and improve the financing reliability and security of the asset management company.

[0039] The target probability of credit anomaly may refer to the probability of a credit anomaly occurring in the subject to be identified.

[0040] The association levels include level 1, level 2, ..., and level M, where M refers to the highest level of the associated subject. Among them, the types of association relationships between the subjects to be identified include upstream and downstream relationships in the industrial chain, guarantee relationships, investment holding relationships, and indirect association relationships. For any subject to be identified, if the type of association relationship between the other subjects to be identified and the current subject to be identified is upstream and downstream relationships in the industrial chain, guarantee relationships, or investment holding relationships, the other subjects to be identified will be determined as the directly associated subjects corresponding to the current subject to be identified.

[0041] Furthermore, the directly associated subject corresponding to the current subject to be identified is determined as the first-level associated subject corresponding to the current subject to be identified, and the directly associated subject corresponding to the j-th level associated subject corresponding to the current subject to be identified is determined as the j+1-th level associated subject corresponding to the current subject to be identified, j=1, 2, ..., M-1, thereby obtaining the association level between the current subject to be identified and each corresponding associated subject, which is used to characterize the closeness of the association relationship between the corresponding subjects to be identified.

[0042] In a specific implementation, S100 further includes the following steps:

[0043] S110, for any subject to be identified, obtain historical scale data of the current subject to be identified, the historical scale data including first reference subject scale data corresponding to the current subject to be identified within a preset first reference time period, and second reference subject scale data corresponding to the preset second reference time period.

[0044] S120, inputting the historical scale data corresponding to the current subject to be identified into the trained abnormal probability identification model, and obtaining the historical abnormal probability corresponding to the current subject to be identified within a preset second reference time period.

[0045] S130, obtaining the credit anomaly prediction error corresponding to the current subject to be identified according to the actual anomaly value and the historical anomaly probability corresponding to the preset second reference time period.

[0046] The start and end time of the preset first reference time period and the start and end time of the preset second reference time period can be set by the implementer according to the actual situation. For example, the start time of the preset first reference time period can refer to a time point 3 years from the current time point, the end time of the preset first reference time period can refer to a time point 2 years from the current time point, the start time of the preset second reference time period can refer to a time point 2 years from the current time point, and the end time of the preset second reference time period can refer to a time point 1 year from the current time point.

[0047] Correspondingly, the first reference entity scale data can characterize the operating conditions of the subject to be identified within the preset first reference time period, and the second reference entity scale data can characterize the operating conditions of the subject to be identified within the preset second reference time period. Therefore, by extracting and analyzing the features of the historical scale data through the trained abnormal probability recognition model, the changes in the operating conditions of the subject to be identified within the preset first reference time period and the preset second reference time period can be analyzed, thereby mapping the initial credit abnormality probability corresponding to the subject to be identified within the preset second reference time period, which is used to characterize the probability of a credit abnormality occurring in the subject to be identified within the corresponding time period.

[0048] If the current subject to be identified has a credit anomaly within the preset second reference time period and the subsequent preset buffer time period, the actual anomaly value corresponding to the current subject to be identified within the preset second reference time period is obtained as 1. If the current subject to be identified has no credit anomaly within the preset second reference time period and the subsequent preset buffer time period, the actual anomaly value corresponding to the current subject to be identified within the preset second reference time period is obtained as 0. The specific value of the preset buffer time period can be set by the implementer according to the actual situation.

[0049] Calculate the difference between the actual anomaly value corresponding to the current subject to be identified and the historical anomaly probability, and determine the difference as the credit anomaly prediction error corresponding to the current subject to be identified. It is used to characterize the recognition accuracy of the anomaly probability recognition model for the current subject to be identified, and serves as the basis for correcting the credit anomaly target probability of the current subject to be identified.

[0050] S200, respectively assigning a corresponding reference set to each subject to be identified.

[0051] Among them, N subjects to be identified are correspondingly assigned to N reference sets as the basis for clustering and merging the subjects to be identified, thereby reducing the number of subjects and improving the efficiency of error correction. Based on the correlation between the clustered and merged subjects, the error data of the abnormal probability of each subject to be identified is measured, so as to correct the credit abnormality target probability of each subject to be identified according to the error data, thereby improving the accuracy of the credit abnormality level of each subject to be identified.

[0052] S300, determining the correlation level between the first prediction error corresponding to each reference set and the reference set according to the correlation level between the credit anomaly prediction error corresponding to each subject to be identified and the subject to be identified.

[0053] In a specific implementation, S300 further includes the following steps:

[0054] S310, for any reference set, according to the credit anomaly prediction errors corresponding to all the entities to be identified in the current reference set, obtain the first prediction error corresponding to the current reference set.

[0055] Among them, the average value of the credit anomaly prediction errors corresponding to all the subjects to be identified in the current reference set is determined as the first prediction error corresponding to the current reference set, which is used to characterize the overall error situation of all the subjects to be identified in the current reference set.

[0056] In a specific implementation, S300 further includes the following steps:

[0057] S320: For any two reference sets, all the subjects to be identified included in one of the two reference sets are used as first intermediate subjects, and all the subjects to be identified included in the other reference set are used as second intermediate subjects.

[0058] S330: Determine the minimum value of the association level between the first intermediate subject and the second intermediate subject as the correlation level between the current two reference sets.

[0059] The minimum value of the association level between the first intermediate subject and the second intermediate subject is determined as the correlation level between the current two reference sets, so as to characterize the closeness of the association relationship between the current two reference sets.

[0060] As described above, based on the correlation level between the credit anomaly prediction error corresponding to each subject to be identified and the subject to be identified, the correlation level between the first prediction error corresponding to each reference set and the reference set is determined to characterize the overall error situation of all subjects to be identified in each reference set, as well as the closeness of the correlation relationship between any two reference sets, as the basis for correcting the credit anomaly target probability of the subject to be identified.

[0061] S400: randomly select a reference set from all reference sets as a temporary set.

[0062] S500: For any reference set, if the current reference set meets the second preset condition, add the subject to be identified in the current reference set to the temporary set, delete the current reference set, and use the temporary set as the reference set.

[0063] In a specific implementation, the second preset condition is:

[0064] The correlation level between the current reference set and the temporary set is less than a preset level threshold, and the absolute value of the difference between the first prediction errors corresponding to the current reference set and the temporary set is less than a preset difference threshold.

[0065] Among them, for the temporary set, if the correlation between the reference set and the temporary set is relatively close, and the difference in the first prediction error is relatively small, the reference set and the temporary set can be merged. Specifically, the subject to be identified in the current reference set is added to the temporary set, and the current reference set is deleted, and the merged temporary set is used as the reference set and the merged object again, so as to iteratively merge all reference sets that meet the second preset condition, thereby improving the acquisition accuracy of the target set.

[0066] Among them, the specific values ​​of the preset level threshold and the preset difference threshold can be set by the implementer according to actual conditions.

[0067] In the above, the merging result between the reference set and the temporary set is judged according to the correlation level between the current reference set and the temporary set and the difference of the first prediction error. When the current reference set meets the second preset condition, the subject to be identified in the current reference set is added to the temporary set, and the current reference set is deleted, thereby reducing the number of sets and improving the efficiency of error correction. Based on the correlation between the merged sets, the error data of the abnormal probability of each subject to be identified is measured, so as to correct the credit abnormality target probability of each subject to be identified according to the error data, thereby improving the accuracy of the credit abnormality level of each subject to be identified.

[0068] S600, returning to step S300, until the number of reference sets converges, Q reference sets are obtained, and the Q reference sets are determined as Q target sets.

[0069] S700, updating the first prediction error corresponding to each target set according to the first prediction error corresponding to each target set and the correlation level between the target sets, to obtain the second prediction error corresponding to each target set.

[0070] In a specific implementation, S700 further includes the following steps:

[0071] S710: Use the first prediction error corresponding to each target set as the corresponding intermediate prediction error.

[0072] S720, obtaining second priority sorting results corresponding to the Q target sets according to the intermediate prediction error corresponding to each target set.

[0073] S730, for the g-th target set in the second priority sorting result, optimize the intermediate prediction error corresponding to the g-th target set according to the optimized prediction errors corresponding to the first g-1 target sets, the intermediate prediction errors corresponding to the g+1th to Qth target sets, and the correlation levels between the target sets, and obtain the optimized prediction error corresponding to the g-th target set, where g=1, 2, ..., Q, and Q refers to the total number of target sets.

[0074] S740, traverse Q target sets, take the optimized prediction error corresponding to each target set as the corresponding intermediate prediction error, and return to execute step S720 until the optimized prediction errors corresponding to all target sets or the number of executions meet the third preset condition, and obtain the second prediction error corresponding to each target set.

[0075] Among them, the larger the intermediate prediction error, the greater the influence of the corresponding target set. Therefore, in this embodiment, the intermediate prediction error and the corresponding second priority are set to be negatively correlated, that is, based on the intermediate prediction error of the target set with greater influence, the errors of other target sets are updated and optimized, and based on the optimized prediction error, the intermediate prediction error of the target set with greater influence is returned to optimize the intermediate prediction error, thereby giving full play to the influence of each target set in analyzing the prediction error and improving the accuracy of the second prediction error.

[0076] There are correlations between target sets, and the intermediate prediction errors corresponding to a target set will affect the size of the errors corresponding to other target sets.

[0077] Specifically, for the g-th target set in the second priority sorting result, the first g-1, g+1 to Q-th target sets are regarded as associated sets corresponding to the g-th target set. The larger the intermediate prediction error corresponding to the associated set, the greater the difference between the intermediate prediction error corresponding to the associated set and the intermediate prediction error of the g-th target set, and the smaller the correlation level between the associated set and the g-th target set, the greater the influence of the associated set on the intermediate prediction error of the g-th target set.

[0078] Furthermore, if the intermediate prediction error of the associated set is less than the intermediate prediction error corresponding to the g-th target set, then the influence of the associated set on the intermediate prediction error of the g-th target set means that the associated set will reduce the error of the g-th target set. If the intermediate prediction error of the associated set is greater than the intermediate prediction error corresponding to the g-th target set, then the influence of the associated set on the intermediate prediction error of the g-th target set means that the associated set will increase the error of the g-th target set.

[0079] In a specific implementation, the optimization prediction error corresponding to the g-th target set meets the following conditions:

[0080] W g =E g +Σ t＝1，t≠g Q (((2 / (1+exp(-B gt )))-1)×E t ×(E t -E g )), where W g It refers to the optimization prediction error corresponding to the g-th target set, E g It refers to the intermediate prediction error corresponding to the g-th target set, E t It refers to the intermediate prediction error corresponding to the t-th target set, B gt It refers to the correlation level between the g-th target set and the t-th target set, t = 1, 2, ..., g-1, g+1, ..., Q, Q refers to the total number of target sets, and exp() is an exponential function.

[0081] In the above, the correlation level and intermediate prediction error between target sets are combined to analyze the error impact between target sets, and the first prediction error of all target sets is iteratively optimized to obtain the second prediction error corresponding to each target set, thereby improving the error prediction accuracy and further improving the accuracy of obtaining the credit anomaly level.

[0082] S800, for any subject to be identified, determining a third prediction error corresponding to the current subject to be identified according to a second prediction error corresponding to a target set to which the current subject to be identified belongs.

[0083] S900, obtaining the credit anomaly level corresponding to each subject to be identified according to the third prediction error and the credit anomaly target probability corresponding to each subject to be identified.

[0084] In a specific implementation, S900 further includes the following steps:

[0085] S910: For any subject to be identified, the third prediction error and the credit anomaly target probability corresponding to the current subject to be identified are input into a preset mapping function to obtain a mapping result.

[0086] S920, determining that the mapping result is the credit anomaly correction probability of the current subject to be identified.

[0087] S930, obtaining the credit anomaly level corresponding to each subject to be identified according to the credit anomaly correction probability corresponding to each subject to be identified.

[0088] Among them, the preset mapping function is F=min(Q×(2 / (exp(-β×Q×γ)+1)), 1), wherein F refers to the credit anomaly correction probability corresponding to the subject to be identified, Q refers to the credit anomaly target probability corresponding to the subject to be identified, β refers to the preset adjustment parameter, γ refers to the third prediction error corresponding to the subject to be identified, and min() refers to the minimum value function.

[0089] The value range of the third prediction error is [-1, 1]. When the third prediction error is less than 0, it means that the target probability of credit anomaly corresponding to the subject to be identified may be larger than the actual anomaly probability. Therefore, the corrected probability of credit anomaly corresponding to the subject to be identified is smaller than the corresponding target probability of credit anomaly. When the third prediction error is equal to 0, it means that the target probability of credit anomaly corresponding to the subject to be identified may be consistent with the actual anomaly probability. Therefore, the corrected probability of credit anomaly corresponding to the subject to be identified is equal to the corresponding target probability of credit anomaly. When the third prediction error is greater than 0, it means that the target probability of credit anomaly corresponding to the subject to be identified may be smaller than the actual anomaly probability. Therefore, the corrected probability of credit anomaly corresponding to the subject to be identified is greater than the corresponding target probability of credit anomaly.

[0090] Furthermore, as the third prediction error increases, the corrected credit anomaly correction probability corresponding to the subject to be identified transitions from less than the corresponding credit anomaly target probability to greater than the corresponding credit anomaly target probability, accurately characterizing the impact of the third prediction error on the credit anomaly correction probability.

[0091] The specific value of β can be set by the implementer according to actual conditions to adjust the influence of the third prediction error on the credit anomaly correction probability. For example, in this embodiment, β=1.2 can be taken according to actual conditions, so that within the value range of [0, 1] corresponding to the credit anomaly target probability, the credit anomaly correction probability of the subject to be identified increases with the increase of the credit anomaly target probability.

[0092] The higher the probability of credit anomaly correction, the higher the corresponding credit anomaly level, which means the higher the possibility of abnormal situation occurring in the corresponding subject to be identified. This helps organizations and individuals understand the various risks they may face, so as to better avoid and control abnormalities. It can also help organizations and individuals sort and prioritize credit anomalies and rationally allocate limited resources.

[0093] Specifically, the correspondence between the credit anomaly correction probability and the credit anomaly level can be set by the implementer according to actual conditions.

[0094] In a specific implementation, the credit anomaly level corresponding to the credit anomaly correction probability in the range of [0, 0.25) is determined as the first anomaly level, the credit anomaly level corresponding to the credit anomaly correction probability in the range of [0.25, 0.5) is determined as the second anomaly level, the credit anomaly level corresponding to the credit anomaly correction probability in the range of [0.5, 0.75) is determined as the third anomaly level, and the credit anomaly level corresponding to the credit anomaly correction probability in the range of [0.75, 1] ​​is determined as the fourth anomaly level.

[0095] As described above, the corresponding credit anomaly target probability is corrected according to the third prediction error to obtain a more accurate credit anomaly correction probability, which is further mapped to obtain the corresponding credit anomaly level, thereby improving the accuracy of obtaining the credit anomaly level.

[0096] In the above, according to the correlation level between the credit anomaly prediction error corresponding to each subject to be identified and the subject to be identified, the correlation level between the first prediction error corresponding to each reference set and the reference set is determined to characterize the overall error situation of all subjects to be identified in each reference set, as well as the closeness of the correlation relationship between any two reference sets; according to the correlation level between the current reference set and the temporary set and the difference in the first prediction error, the merging result between the reference set and the temporary set is judged, and when the current reference set meets the second preset condition, the subject to be identified in the current reference set is added to the temporary set, and the current reference set is deleted, and then all reference sets are iteratively merged to obtain Q target sets, reducing The number of sets is reduced, and the efficiency of error correction is improved; the correlation level and intermediate prediction error between target sets are further combined to analyze the error impact between target sets, and the first prediction error of all target sets is iteratively optimized to obtain the second prediction error corresponding to each target set, thereby improving the accuracy of the second prediction error; for any subject to be identified, the third prediction error corresponding to the current subject to be identified is determined according to the second prediction error corresponding to the target set to which the current subject to be identified belongs, and then the corresponding credit anomaly target probability is corrected according to the third prediction error to obtain a more accurate credit anomaly correction probability, which is further mapped to obtain the corresponding credit anomaly level, thereby improving the accuracy of obtaining the credit anomaly level.

[0097] Embodiment 2

[0098] This embodiment 2 provides a method for obtaining a credit anomaly level. The method for obtaining a credit anomaly level includes: Figure 2 As shown:

[0099] S1, obtain N entities to be identified, the entity size data and initial credit anomaly probability corresponding to each entity to be identified, and the types of association relationships between the entities to be identified, where N is an integer greater than 1.

[0100] Among them, the subject size data can represent the size of the corresponding subject to be identified, and then represent the relative importance and influence between the subjects to be identified. The initial credit anomaly probability can refer to the probability of the subject to be identified having a credit anomaly, which is initially obtained after analyzing and measuring the subject size data and other related information of the subject to be identified. Among them, the initial credit anomaly probability can be obtained based on the relevant data of the credit anomaly of the subject to be identified in the historical time period, or it can be manually determined by relevant professionals.

[0101] This embodiment takes the financial and investment fields as an example for explanation. The subject size data may refer to the financial report, number of employees and other relevant operating data of the subject to be identified, which is used to characterize the economic situation of the subject to be identified. The initial credit anomaly probability may refer to the probability of the subject to be identified having a credit anomaly, which is initially obtained after analyzing and measuring the relevant operating data of the subject to be identified.

[0102] In a specific implementation, S1 further includes the following steps:

[0103] S11, obtaining the subject scale data corresponding to each subject to be identified and the trained abnormal probability recognition model, wherein the subject scale data includes the first subject scale data corresponding to each subject to be identified in a preset historical time period, and the second subject scale data corresponding to the preset current time period.

[0104] S12, inputting the subject scale data corresponding to each subject to be identified into the trained abnormal probability identification model, and obtaining the initial credit abnormality probability corresponding to each subject to be identified in the preset current time period.

[0105] The end time of the preset current time period refers to the current time point, and the start and end time of the preset historical time period and the start time of the preset current time period can be set by the implementer according to the actual situation. For example, the start time of the preset historical time period can refer to a time point 1 year from the current time point, the end time of the preset historical time period can refer to a time point 1 month from the current time point, and the start time of the preset current time period can refer to a time point 1 month from the current time point.

[0106] Correspondingly, the first subject scale data can characterize the operating conditions of the subject to be identified in a preset historical time period, and the second subject scale data can characterize the operating conditions of the subject to be identified in a preset current time period. Therefore, by extracting and analyzing the characteristics of the subject scale data through the trained abnormal probability recognition model, the changes in the operating conditions of the subject to be identified in the preset historical time period and the preset current time period can be analyzed, thereby mapping the initial credit abnormality probability corresponding to the subject to be identified in the preset current time period, which is used to characterize the probability of credit abnormality occurring in the subject to be identified in the preset current time period and the preset buffer time period after the current time point. Among them, the specific value of the preset buffer time period can be set by the implementer according to the actual situation.

[0107] Among them, the trained anomaly probability recognition model includes an encoder and a fully connected layer. The encoder is used to extract features from the subject-scale data, and the fully connected layer is used to map the extracted features to the corresponding initial credit anomaly probability.

[0108] As described above, based on the trained abnormal probability recognition model, the subject scale data is feature extracted and analyzed, and the changes in the operating conditions of the subject to be identified in the preset historical time period and the preset current time period are analyzed, so as to map the initial credit abnormality probability corresponding to the subject to be identified in the preset current time period, thereby improving the accuracy of the initial credit abnormality probability, and further improving the accuracy of the credit abnormality level of the subject to be identified.

[0109] S2: For any subject to be identified, determine other N-1 subjects to be identified as associated subjects of the current subject to be identified.

[0110] S3, determining the association level and association degree between the current subject to be identified and each corresponding associated subject according to the subject scale data corresponding to each subject to be identified and the type of association relationship between the subjects to be identified.

[0111] In a specific implementation, the types of association relationships include upstream and downstream relationships in the industrial chain, guarantee relationships, investment holding relationships, and indirect association relationships. S3 also includes the following steps:

[0112] S31, for any subject to be identified, if the type of association relationship between the other subject to be identified and the current subject to be identified is an upstream and downstream relationship in the industrial chain, a guarantee relationship or an investment holding relationship, the other subject to be identified shall be determined as a directly associated subject corresponding to the current subject to be identified.

[0113] S32, obtaining the proportion of the associated amount between the current subject to be identified and each corresponding directly associated subject.

[0114] S33, taking each subject to be identified as a node, taking the subject size data and initial credit anomaly probability corresponding to each subject to be identified as node features, connecting each subject to be identified with the corresponding directly associated subject through an edge, taking the proportion of the associated amount between each subject to be identified and the corresponding directly associated subject as the edge weight of the corresponding edge, and constructing a subject association graph.

[0115] S34, for any subject to be identified, according to the subject association map, determine the directly associated subject corresponding to the current subject to be identified as the first-level associated subject corresponding to the current subject to be identified.

[0116] S35, determining the directly associated subject corresponding to the j-th level associated subject corresponding to the current subject to be identified as the j+1-th level associated subject corresponding to the current subject to be identified, where j=1, 2, ..., M-1, and M refers to the highest level of the associated subject corresponding to the current subject to be identified.

[0117] S36, according to the first-level associated subjects to the Mth-level associated subjects corresponding to the current subject to be identified, the association level between the current subject to be identified and each corresponding associated subject is obtained, wherein the association levels include the first level, the second level, ..., the Mth level.

[0118] Among them, based on the subject association map, according to the type of association relationship between each subject to be identified, the subject to be identified is associated with other N-1 subjects to be identified, and the corresponding association level is determined to characterize the closeness of the association relationship between the corresponding subjects to be identified. Correspondingly, the smaller the association level, the closer the association relationship between the corresponding subjects to be identified.

[0119] The proportion of associated amounts can be used to represent the degree of correlation between the corresponding entities to be identified. Correspondingly, the larger the proportion of associated amounts, the higher the degree of correlation between the corresponding entities to be identified.

[0120] As described above, based on the subject size data corresponding to each subject to be identified and the type of association relationship between the subjects to be identified, the association level and degree between the current subject to be identified and each corresponding associated subject are determined to characterize the closeness of the association relationship between the corresponding subjects to be identified, providing a data basis for obtaining the credit anomaly level of the subject to be identified.

[0121] In a specific implementation, S32 further includes the following steps:

[0122] S321, for any directly associated entity corresponding to the current entity to be identified, if the type of association relationship between the current entity to be identified and the current directly associated entity is an upstream and downstream relationship in the industrial chain, the transaction amount between the current entity to be identified and the current directly associated entity, as well as the total transaction amount corresponding to the current entity to be identified, are obtained, and the ratio of the transaction amount to the total transaction amount is determined as the proportion of the association amount between the current entity to be identified and the current directly associated entity.

[0123] S322, if the type of association relationship between the current subject to be identified and the current directly associated subject is a guarantee relationship, then obtain the guarantee amount between the current subject to be identified and the current directly associated subject, as well as the total guarantee amount corresponding to the current subject to be identified, and determine the ratio of the guarantee amount to the total guarantee amount as the ratio of the association amount between the current subject to be identified and the current directly associated subject.

[0124] S323, if the type of association relationship between the current subject to be identified and the current directly associated subject is an investment holding relationship, then obtain the investment holding amount between the current subject to be identified and the current directly associated subject, and the total investment holding amount corresponding to the current subject to be identified, and determine the ratio of the investment holding amount to the total investment holding amount as the association amount ratio between the current subject to be identified and the current directly associated subject.

[0125] In a specific implementation, S3 further includes the following steps:

[0126] S37, obtaining the association weight corresponding to each association level, wherein the association weight includes the first level weight, the second level weight, ..., the Mth level weight.

[0127] S38, according to the proportion of the association amount between the current subject to be identified and the corresponding directly associated subject, and the first-level weight, the degree of association between the current subject to be identified and the corresponding first-level associated subject is obtained.

[0128] S39, according to the degree of association between the current subject to be identified and the corresponding j-1th level associated subject, the proportion of the association amount between each j-1th level associated subject and the corresponding directly associated subject, and the jth level weight, obtain the degree of association between the current subject to be identified and the corresponding j-1th level associated subject.

[0129] Among them, since the smaller the association level, the closer the association relationship between the corresponding subjects to be identified, the first-level weight, the second-level weight, ..., and the M-th-level weight are successively reduced. Specifically, the specific values ​​of the first-level weight, the second-level weight, ..., and the M-th-level weight can be set by the implementer according to actual conditions.

[0130] Therefore, for any directly associated subject corresponding to the current subject to be identified, the product of the proportion of the associated amount between the current subject to be identified and the current directly associated subject and the first-level weight is determined as the degree of association between the current subject to be identified and the current first-level associated subject.

[0131] For any directly associated subject corresponding to any j-1th level associated subject corresponding to the current subject to be identified,

[0132] The product of the degree of association between the current subject to be identified and the current j-1th level associated subject, the proportion of the association amount between the current j-1th level associated subject and the corresponding current directly associated subject, and the j-th level weight is determined as the degree of association between the current subject to be identified and the current j-1th level associated subject.

[0133] As mentioned above, combined with the association weight corresponding to the association level and the proportion of the association amount between the subjects to be identified, the degree of association between the subject to be identified and each associated subject is obtained, which is used to characterize the closeness of the association relationship between the subject to be identified and the associated subjects, and provides a data basis for obtaining the credit anomaly level of the subject to be identified.

[0134] S4, according to the association level and association degree between the current subject to be identified and each associated subject, as well as the subject size data corresponding to each associated subject, the influence degree of each associated subject on the current subject to be identified is obtained.

[0135] In a specific implementation, S4 further includes the following steps:

[0136] S41, for any associated subject corresponding to the current subject to be identified, according to the association level between the current subject to be identified and the current associated subject, obtain the influence attenuation coefficient between the current subject to be identified and the current associated subject, wherein the influence attenuation coefficient is positively correlated with the corresponding association level.

[0137] S42, based on the influence attenuation coefficient and the degree of association between the current subject to be identified and the current associated subject, as well as the subject scale data corresponding to the current associated subject, obtain the influence degree of the current associated subject on the current subject to be identified, wherein the influence degree is negatively correlated with the corresponding influence attenuation coefficient, and the influence degree is positively correlated with the corresponding degree of association and the corresponding subject scale data.

[0138] Among them, the greater the association level between the current subject to be identified and the current associated subject, the lower the closeness of the association relationship between the current subject to be identified and the current associated subject. When the initial credit anomaly probability of the current associated subject affects the anomaly probability of the current subject to be identified, it will go through multiple levels of attenuation. Correspondingly, the greater the impact attenuation coefficient between the current subject to be identified and the current associated subject.

[0139] Furthermore, the greater the influence attenuation coefficient between the current subject to be identified and the current associated subject, the smaller the degree of association, and the smaller the subject scale data corresponding to the current associated subject, the smaller the influence of the current associated subject on the current subject to be identified.

[0140] In a specific implementation, the influence attenuation coefficient between the current subject to be identified and the current associated subject meets the following conditions:

[0141] X=j / M, where X refers to the influence attenuation coefficient between the current subject to be identified and the current associated subject, j refers to the association level between the current subject to be identified and the current associated subject, and M refers to the highest level of the association level.

[0142] The degree of influence of the current associated subject on the current subject to be identified meets the following conditions:

[0143] Y=exp(-X)×(α1×D+α2×G), where Y refers to the degree of influence of the current associated subject on the current subject to be identified, D refers to the degree of association between the current subject to be identified and the current associated subject, G refers to the subject scale data corresponding to the current associated subject, exp() is an exponential function, α1 refers to the preset association degree weight, and α2 refers to the preset scale data weight.

[0144] The specific values ​​of α1 and α2 can be set by the implementer according to actual conditions.

[0145] The above-mentioned method combines the association level and degree of association between the current subject to be identified and each associated subject, as well as the subject size data corresponding to each associated subject, to characterize the influence of each associated subject on the current subject to be identified, which serves as the basis for evaluating the influence of the initial credit anomaly probability of the current associated subject on the anomaly probability of the current subject to be identified, thereby improving the accuracy of obtaining the credit anomaly level corresponding to the subject to be identified.

[0146] S5, taking the initial credit anomaly probability corresponding to each subject to be identified as the corresponding credit anomaly intermediate probability.

[0147] S6. According to the subject size data and the intermediate probability of credit anomaly corresponding to each subject to be identified, the first priority sorting results corresponding to N subjects to be identified are obtained.

[0148] In a specific implementation, S6 further includes the following steps:

[0149] S61, according to the subject size data and the initial credit anomaly probability corresponding to the subject to be identified, the optimization priority corresponding to the subject to be identified is obtained, wherein the optimization priority is negatively correlated with the corresponding subject size data and the initial credit anomaly probability.

[0150] S62, traverse N subjects to be identified, and obtain the optimization priorities corresponding to the N subjects to be identified respectively.

[0151] S63, sorting the N subjects to be identified in descending order of optimization priority, and obtaining a first priority sorting result.

[0152] Among them, the larger the subject scale data and the larger the initial credit anomaly probability, the higher the influence and importance of the corresponding subject to be identified. Therefore, in this embodiment, the optimization priority is negatively correlated with the corresponding subject scale data and the initial credit anomaly probability, that is, based on the credit anomaly intermediate probability of the subject to be identified with higher influence and importance, the anomaly probabilities of other subjects to be identified are updated and optimized, and based on the updated and optimized credit anomaly optimization probability, the credit anomaly intermediate probability of the subject to be identified with higher influence and importance is returned, thereby giving full play to the importance and influence of each subject to be identified in analyzing the anomaly probability and improving the accuracy of the credit anomaly optimization probability.

[0153] In a specific implementation, the optimization priority corresponding to the current subject to be identified meets the following conditions:

[0154] R = (exp(-U)) / V, where R refers to the optimization priority corresponding to the current subject to be identified, U refers to the subject size data corresponding to the current subject to be identified, and V refers to the initial credit anomaly probability corresponding to the current subject to be identified.

[0155] As described above, based on the subject size data and the intermediate probability of credit anomalies corresponding to each subject to be identified, the first priority sorting results corresponding to N subjects to be identified are obtained, and the order of updating and optimizing the anomaly probabilities of the subjects to be identified is determined according to the first priority sorting results, so as to give full play to the importance and influence of each subject to be identified in analyzing the anomaly probability and improve the optimization accuracy of the anomaly probability.

[0156] S7. For the ith subject to be identified in the first priority sorting result, optimize the intermediate probability of credit anomaly of the ith subject to be identified according to the optimized probability of credit anomaly corresponding to the first i-1 subjects to be identified, the intermediate probability of credit anomaly corresponding to the i+1th to Nth subjects to be identified, and the degree of influence of each associated subject on the ith subject to be identified, and obtain the optimized probability of credit anomaly corresponding to the ith subject to be identified, where i=1, 2, ..., N.

[0157] In a specific implementation, S7 further includes the following steps:

[0158] S71. For the i-th subject to be identified in the first priority sorting result, the credit anomaly reference probability of each associated subject corresponding to the i-th subject to be identified is obtained based on the credit anomaly optimization probabilities corresponding to the first i-1 subjects to be identified and the credit anomaly intermediate probabilities corresponding to the i+1th to Nth subjects to be identified.

[0159] S72, based on the reference probability of credit anomaly of each associated subject corresponding to the i-th subject to be identified, and the degree of influence of each corresponding associated subject on the i-th subject to be identified, optimize the intermediate probability of credit anomaly of the i-th subject to be identified, and obtain the optimized probability of credit anomaly corresponding to the i-th subject to be identified.

[0160] Among them, there is a correlation between the subject to be identified and the corresponding associated subject. Whether an abnormal situation occurs in the associated subject corresponding to the subject to be identified, that is, the intermediate probability of credit abnormality corresponding to the associated subject, will affect the probability of abnormality occurring in the subject to be identified.

[0161] Specifically, the greater the credit anomaly reference probability of the associated subject corresponding to the subject to be identified, the greater the difference between the credit anomaly reference probability of the associated subject and the credit anomaly reference probability corresponding to the subject to be identified, and the deeper the influence of the associated subject on the subject to be identified, the greater the influence of the associated subject on the intermediate probability of credit anomaly of the subject to be identified.

[0162] Furthermore, if the credit anomaly reference probability of the associated subject is less than the credit anomaly reference probability corresponding to the subject to be identified, the impact of the associated subject on the intermediate probability of credit anomaly of the subject to be identified means that the associated subject will reduce the probability of anomaly of the subject to be identified. If the credit anomaly reference probability of the associated subject is greater than the credit anomaly reference probability corresponding to the subject to be identified, the impact of the associated subject on the intermediate probability of credit anomaly of the subject to be identified means that the associated subject will increase the probability of anomaly of the subject to be identified.

[0163] In a specific implementation, the credit anomaly optimization probability corresponding to the i-th subject to be identified meets the following conditions:

[0164] H i =P i +Σ k＝1 N-1 (((2 / (1+exp(-Y ik )))-1)×P ik ×(P ik -P i )), where H i It refers to the credit anomaly optimization probability corresponding to the i-th subject to be identified, P i It refers to the credit anomaly reference probability corresponding to the i-th subject to be identified, P ikIt refers to the credit anomaly reference probability corresponding to the kth associated subject corresponding to the i-th subject to be identified, Y ik It refers to the degree of influence of the kth associated subject corresponding to the i-th subject to be identified on the i-th subject to be identified, and N-1 refers to the total number of associated subjects corresponding to the i-th subject to be identified.

[0165] In a specific implementation, S71 further includes the following steps:

[0166] S711, for any associated subject corresponding to the current subject to be identified, if the current associated subject belongs to the first i-1 subjects to be identified in the first priority sorting result, the credit anomaly optimization probability corresponding to the current associated subject is determined as the credit anomaly reference probability of the current associated subject.

[0167] S712: If the current associated subject belongs to the i+1th to Nth subjects to be identified in the first priority sorting result, the credit anomaly intermediate probability corresponding to the current associated subject is determined as the credit anomaly reference probability of the current associated subject.

[0168] S713, traverse all associated entities corresponding to the current subject to be identified, and obtain the credit anomaly reference probability of each associated entity corresponding to the current subject to be identified.

[0169] As described above, the order of updating and optimizing the abnormality probability of the subject to be identified is determined according to the result of the first priority sorting, so as to update and optimize the credit abnormality intermediate probability of the subject to be identified based on the credit abnormality intermediate probability of the subject to be identified, combined with the difference between the abnormality probabilities corresponding to the associated subject and the subject to be identified, as well as the degree of influence between the subjects to be identified, and based on the updated and optimized credit abnormality optimized probability, return the credit abnormality intermediate probability of the subject to be identified with a lower optimization priority, thereby improving the optimization accuracy of the abnormality probability.

[0170] S8, traverse N entities to be identified, take the credit anomaly optimization probability corresponding to each entity to be identified as the corresponding credit anomaly intermediate probability, and repeat step S6 until the credit anomaly optimization probability of all entities to be identified or the number of repetitions meets the first preset condition, and obtain the credit anomaly target probability corresponding to each entity to be identified.

[0171] Among them, the credit anomaly optimization probability or the number of repetitions of the subject to be identified meets the first preset condition, which means that the credit anomaly optimization probability of the subject to be identified converges, or the number of repetitions reaches the preset threshold, then the iteration is stopped, and the credit anomaly optimization probability corresponding to each subject to be identified at this time is determined as the final credit anomaly target probability, thereby improving the accuracy of the credit anomaly target probability.

[0172] The specific value of the preset number threshold can be set by the implementer according to actual conditions.

[0173] S9, according to the credit anomaly target probability corresponding to each subject to be identified, obtain the credit anomaly level corresponding to each subject to be identified.

[0174] Among them, the higher the credit anomaly target probability, the higher the corresponding credit anomaly level, which means that the corresponding subject to be identified is more likely to have an abnormal situation. This helps organizations and individuals understand the various risks they may face, so as to better avoid and control abnormalities. It can also help organizations and individuals sort and prioritize credit anomalies and rationally allocate limited resources.

[0175] Specifically, the correspondence between the credit anomaly target probability and the credit anomaly level can be set by the implementer according to actual conditions.

[0176] In a specific implementation, the credit anomaly level corresponding to the credit anomaly target probability in the range of [0, 0.25) is determined as the first anomaly level, the credit anomaly level corresponding to the credit anomaly target probability in the range of [0.25, 0.5) is determined as the second anomaly level, the credit anomaly level corresponding to the credit anomaly target probability in the range of [0.5, 0.75) is determined as the third anomaly level, and the credit anomaly level corresponding to the credit anomaly target probability in the range of [0.75, 1] ​​is determined as the fourth anomaly level.

[0177] In the above, according to the subject scale data corresponding to each subject to be identified and the type of association relationship between the subjects to be identified, the association level and degree between the current subject to be identified and each corresponding associated subject are determined to characterize the closeness of the association relationship between the corresponding subjects to be identified, and the association level and degree between the current subject to be identified and each associated subject, as well as the subject scale data corresponding to each associated subject, are combined to characterize the degree of influence of each associated subject on the current subject to be identified, which serves as the basis for evaluating the influence of the initial credit anomaly probability of the current associated subject on the anomaly probability of the current subject to be identified, and then according to the subject scale data corresponding to each subject to be identified and the intermediate probability of credit anomaly, the probability of the initial credit anomaly probability of the current associated subject on the abnormal probability of the current subject to be identified is obtained. The first priority sorting results corresponding to N subjects to be identified are obtained, and the order of updating and optimizing the abnormality probabilities of the subjects to be identified is determined according to the first priority sorting results, so as to update and optimize the credit abnormality intermediate probability of the subject to be identified based on the credit abnormality intermediate probability of the subject to be identified, combined with the difference between the abnormality probabilities corresponding to the associated subject and the subject to be identified, as well as the degree of influence between the subjects to be identified, and return the credit abnormality intermediate probability of the subject to be identified with a lower optimization priority based on the updated and optimized credit abnormality optimization probability, and obtain the final credit abnormality target probability by iterating the optimization steps of the above probability, thereby improving the optimization accuracy of the abnormality probability and thus improving the accuracy of the credit abnormality level.

[0178] Embodiment 3

[0179] The third embodiment provides an abnormality level acquisition device based on error correction, and the abnormality level acquisition device based on error correction includes: Figure 3 As shown:

[0180] The data acquisition module 31 is used to obtain N subjects to be identified, the credit anomaly target probability and credit anomaly prediction error corresponding to each subject to be identified, and the correlation level between the subjects to be identified, where N is an integer greater than 1.

[0181] The reference set allocation module 32 is used to allocate a corresponding reference set to each subject to be identified.

[0182] The correlation level acquisition module 33 is used to determine the correlation level between the first prediction error corresponding to each reference set and the reference set according to the correlation level between the credit anomaly prediction error corresponding to each subject to be identified and the subject to be identified.

[0183] The temporary set selection module 34 is used to randomly select a reference set from all reference sets as a temporary set.

[0184] The reference set merging module 35 is used for adding the subject to be identified in the current reference set to the temporary set for any reference set if the current reference set meets the second preset condition, and deleting the current reference set to use the temporary set as the reference set.

[0185] The target set acquisition module 36 is used to return to execute the relevant level acquisition module S300 until the number of reference sets converges to obtain Q reference sets, and determine the Q reference sets as Q target sets.

[0186] The prediction error updating module 37 is used to update the first prediction error corresponding to each target set according to the first prediction error corresponding to each target set and the correlation level between the target sets to obtain the second prediction error corresponding to each target set.

[0187] The third prediction error acquisition module 38 is used to determine, for any subject to be identified, a third prediction error corresponding to the current subject to be identified according to the second prediction error corresponding to the target set to which the current subject to be identified belongs.

[0188] The credit anomaly level acquisition module 39 is used to acquire the credit anomaly level corresponding to each subject to be identified according to the third prediction error and the credit anomaly target probability corresponding to each subject to be identified.

[0189] In a specific implementation, the data acquisition module 31 further includes:

[0190] The data acquisition submodule is used to obtain N entities to be identified, the entity size data and initial credit anomaly probability corresponding to each entity to be identified, and the types of association relationships between the entities to be identified.

[0191] The associated subject determination submodule is used to determine, for any subject to be identified, other N-1 subjects to be identified as associated subjects of the current subject to be identified.

[0192] The associated data acquisition submodule is used to determine the association level and association degree between the current subject to be identified and each corresponding associated subject according to the subject scale data corresponding to each subject to be identified and the type of association relationship between the subjects to be identified.

[0193] The influence degree acquisition submodule is used to obtain the influence degree of each associated subject on the current subject to be identified based on the association level and association degree between the current subject to be identified and each associated subject, as well as the subject scale data corresponding to each associated subject.

[0194] The intermediate probability acquisition submodule is used to take the initial credit anomaly probability corresponding to each subject to be identified as the corresponding credit anomaly intermediate probability.

[0195] The first priority sorting submodule is used to obtain the first priority sorting results corresponding to N subjects to be identified according to the number of associated subjects corresponding to each subject to be identified, the subject scale data and the intermediate probability of credit anomaly.

[0196] The probability optimization submodule is used to optimize the intermediate probability of credit anomaly of the ith subject to be identified in the first priority sorting result, based on the credit anomaly optimization probabilities corresponding to the first i-1 subjects to be identified, the credit anomaly intermediate probabilities corresponding to the i+1 to N subjects to be identified, and the degree of influence of each associated subject on the ith subject to be identified, so as to obtain the credit anomaly optimization probability corresponding to the ith subject to be identified, where i=1, 2,..., N.

[0197] The optimization iteration submodule is used to traverse N entities to be identified, take the credit anomaly optimization probability corresponding to each entity to be identified as the corresponding credit anomaly intermediate probability, and return to execute the first priority sorting submodule until the credit anomaly optimization probability or the number of repetitions of all entities to be identified meet the preset requirements, and obtain the credit anomaly target probability corresponding to each entity to be identified.

[0198] In a specific implementation, the data acquisition module 31 further includes:

[0199] The historical scale data acquisition submodule is used to acquire the historical scale data of the current subject to be identified for any subject to be identified, and the historical scale data includes the first reference subject scale data corresponding to the current subject to be identified within a preset first reference time period, and the second reference subject scale data corresponding to the preset second reference time period.

[0200] The historical abnormal probability acquisition submodule is used to input the historical scale data corresponding to the current subject to be identified into the trained abnormal probability recognition model to obtain the historical abnormal probability corresponding to the current subject to be identified within a preset second reference time period.

[0201] The credit anomaly prediction error acquisition submodule is used to obtain the credit anomaly prediction error corresponding to the current subject to be identified based on the actual anomaly value and historical anomaly probability corresponding to the preset second reference time period.

[0202] In a specific implementation, the relevant level acquisition module 33 further includes:

[0203] The first prediction error acquisition submodule is used to obtain, for any reference set, the first prediction error corresponding to the current reference set based on the credit anomaly prediction errors corresponding to all the entities to be identified in the current reference set.

[0204] In a specific implementation, the relevant level acquisition module 33 further includes:

[0205] The intermediate subject determination submodule is used to, for any two reference sets, use all the subjects to be identified contained in one of the current two reference sets as the first intermediate subjects, and use all the subjects to be identified contained in the other reference set as the second intermediate subjects.

[0206] The correlation level acquisition submodule is used to determine the minimum value of the correlation level between the first intermediate subject and the second intermediate subject as the correlation level between the current two reference sets.

[0207] In a specific implementation, the prediction error updating module 37 further includes:

[0208] The intermediate prediction error acquisition submodule is used to take the first prediction error corresponding to each target set as the corresponding intermediate prediction error.

[0209] The second priority sorting submodule is used to obtain the second priority sorting results corresponding to the Q target sets according to the intermediate prediction error corresponding to each target set.

[0210] The prediction error updating submodule is used to optimize the intermediate prediction error corresponding to the g-th target set in the second priority sorting result according to the optimized prediction errors corresponding to the first g-1 target sets, the intermediate prediction errors corresponding to the g+1th to Qth target sets, and the correlation levels between the target sets, and obtain the optimized prediction error corresponding to the g-th target set, where g=1, 2,..., Q, and Q refers to the total number of target sets.

[0211] The update iteration submodule is used to traverse Q target sets, take the optimized prediction error corresponding to each target set as the corresponding intermediate prediction error, and return to execute the second priority sorting submodule until the optimized prediction error or the number of executions corresponding to all target sets meets the third preset condition, and the second prediction error corresponding to each target set is obtained.

[0212] In a specific implementation, the credit abnormality level acquisition module 39 further includes:

[0213] The mapping submodule is used to input the third prediction error and credit anomaly target probability corresponding to the current subject to be identified into a preset mapping function for any subject to be identified, and obtain the mapping result.

[0214] The correction probability acquisition submodule is used to determine that the mapping result is the credit anomaly correction probability of the current subject to be identified.

[0215] The credit anomaly level acquisition submodule is used to obtain the credit anomaly level corresponding to each subject to be identified based on the credit anomaly correction probability corresponding to each subject to be identified.

[0216] It should be noted that the information interaction, execution process and other contents between the above modules are based on the same concept as the embodiment of the method of the present invention. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0217] Embodiment 4

[0218] Embodiment 4 of the present invention provides a non-transitory computer-readable storage medium, wherein at least one instruction or at least one program is stored in the non-transitory computer-readable storage medium, and the at least one instruction or at least one program is loaded and executed by a processor to implement the steps:

[0219] S100, obtaining N entities to be identified, the credit anomaly target probability and credit anomaly prediction error corresponding to each entity to be identified, and the correlation level between the entities to be identified, where N is an integer greater than 1.

[0220] S200, respectively assigning a corresponding reference set to each subject to be identified.

[0221] S300, determining the correlation level between the first prediction error corresponding to each reference set and the reference set according to the correlation level between the credit anomaly prediction error corresponding to each subject to be identified and the subject to be identified.

[0222] S400: randomly select a reference set from all reference sets as a temporary set.

[0223] S500: For any reference set, if the current reference set meets the second preset condition, add the subject to be identified in the current reference set to the temporary set, delete the current reference set, and use the temporary set as the reference set.

[0224] S600, returning to step S300, until the number of reference sets converges, Q reference sets are obtained, and the Q reference sets are determined as Q target sets.

[0225] S700, updating the first prediction error corresponding to each target set according to the first prediction error corresponding to each target set and the correlation level between the target sets, to obtain the second prediction error corresponding to each target set.

[0226] S800, for any subject to be identified, determining a third prediction error corresponding to the current subject to be identified according to a second prediction error corresponding to a target set to which the current subject to be identified belongs.

[0227] S900, obtaining the credit anomaly level corresponding to each subject to be identified according to the third prediction error and the credit anomaly target probability corresponding to each subject to be identified.

[0228] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronization link DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0229] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0230] Embodiment 5

[0231] A fifth embodiment of the present invention provides an electronic device, which includes a processor and the non-transitory computer-readable storage medium in the fourth embodiment of the present invention.

[0232] The above are only preferred embodiments of the present invention, and are not intended to limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not intended to limit the present invention. Any technician familiar with the profession can make some changes or modifications to equivalent embodiments of equivalent changes using the technical contents disclosed above without departing from the scope of the technical solution of the present invention. However, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A method for obtaining anomaly levels based on error correction, characterized in that: The abnormality level acquisition method based on error correction comprises the following steps: S100, obtaining N entities to be identified, the credit anomaly target probability and credit anomaly prediction error corresponding to each entity to be identified, and the correlation level between the entities to be identified, where N is an integer greater than 1; S200, respectively assigning a corresponding reference set to each subject to be identified; S300, determining the correlation level between the first prediction error corresponding to each reference set and the reference set according to the correlation level between the credit anomaly prediction error corresponding to each subject to be identified and the subject to be identified; S400, randomly selecting a reference set from all reference sets as a temporary set; S500, for any reference set, if the current reference set meets the second preset condition, add the subject to be identified in the current reference set to the temporary set, delete the current reference set, and use the temporary set as the reference set; S600, returning to step S300, until the number of reference sets converges, Q reference sets are obtained, and the Q reference sets are determined as Q target sets; S700, updating the first prediction error corresponding to each target set according to the first prediction error corresponding to each target set and the correlation level between the target sets, to obtain a second prediction error corresponding to each target set; S800, for any subject to be identified, determining a third prediction error corresponding to the current subject to be identified according to a second prediction error corresponding to a target set to which the current subject to be identified belongs; S900, obtaining the credit anomaly level corresponding to each subject to be identified according to the third prediction error and the credit anomaly target probability corresponding to each subject to be identified.

2. The method for obtaining anomaly level based on error correction according to claim 1, characterized in that: S100 also includes the following steps: S1, obtain N entities to be identified, the entity size data and initial credit anomaly probability corresponding to each entity to be identified, and the types of association relationships between the entities to be identified; S2, for any subject to be identified, determine the other N-1 subjects to be identified as associated subjects of the current subject to be identified; S3, determining the association level and degree of association between the current subject to be identified and each corresponding associated subject according to the subject scale data corresponding to each subject to be identified and the type of association relationship between the subjects to be identified; S4, according to the association level and association degree between the current subject to be identified and each associated subject, as well as the subject size data corresponding to each associated subject, obtain the influence degree of each associated subject on the current subject to be identified; S5, taking the initial credit anomaly probability corresponding to each subject to be identified as the corresponding credit anomaly intermediate probability; S6, according to the number of associated entities, entity scale data and credit anomaly intermediate probability corresponding to each entity to be identified, obtain the first priority ranking results corresponding to N entities to be identified; S7, for the i-th subject to be identified in the first priority sorting result, according to the credit anomaly optimization probabilities corresponding to the first i-1 subjects to be identified, the credit anomaly intermediate probabilities corresponding to the i+1th to Nth subjects to be identified, and the influence of each associated subject on the i-th subject to be identified, optimize the credit anomaly intermediate probability of the i-th subject to be identified, and obtain the credit anomaly optimization probability corresponding to the i-th subject to be identified, where i=1, 2, ..., N; S8, traverse N entities to be identified, take the credit anomaly optimization probability corresponding to each entity to be identified as the corresponding credit anomaly intermediate probability, and return to execute step S6 until the credit anomaly optimization probability or the number of repetitions of all entities to be identified meet the preset requirements, and obtain the credit anomaly target probability corresponding to each entity to be identified.

3. The method for obtaining anomaly level based on error correction according to claim 2, characterized in that: S100 also includes the following steps: S110, for any subject to be identified, obtaining historical scale data of the subject to be identified, wherein the historical scale data includes first reference subject scale data corresponding to the subject to be identified within a preset first reference time period, and second reference subject scale data corresponding to the subject to be identified within a preset second reference time period; S120, inputting the historical scale data corresponding to the current subject to be identified into the trained abnormal probability recognition model, and obtaining the historical abnormal probability corresponding to the current subject to be identified within the preset second reference time period; S130, obtaining a credit anomaly prediction error corresponding to the current subject to be identified based on the actual anomaly value corresponding to the preset second reference time period and the historical anomaly probability.

4. The method for obtaining anomaly level based on error correction according to claim 1, characterized in that: The second preset condition is: The correlation level between the current reference set and the temporary set is less than a preset level threshold, and the absolute value of the difference between the first prediction errors corresponding to the current reference set and the temporary set is less than a preset difference threshold.

5. The method for obtaining anomaly level based on error correction according to claim 1, characterized in that: S300 also includes the following steps: S310, for any reference set, according to the credit anomaly prediction errors corresponding to all the entities to be identified in the current reference set, obtain the first prediction error corresponding to the current reference set.

6. The method for obtaining anomaly level based on error correction according to claim 1, characterized in that: S300 also includes the following steps: S320, for any two reference sets, taking all the subjects to be identified included in one of the two reference sets as first intermediate subjects, and taking all the subjects to be identified included in the other reference set as second intermediate subjects; S330: Determine the minimum value of the association level between the first intermediate subject and the second intermediate subject as the correlation level between the current two reference sets.

7. The method for obtaining anomaly level based on error correction according to claim 1, characterized in that: S700 also includes the following steps: S710, taking the first prediction error corresponding to each target set as the corresponding intermediate prediction error; S720, obtaining second priority sorting results corresponding to the Q target sets according to the intermediate prediction error corresponding to each target set; S730, for the g-th target set in the second priority sorting result, according to the optimized prediction errors corresponding to the first g-1 target sets, the intermediate prediction errors corresponding to the g+1th to Qth target sets, and the correlation levels between the target sets, optimize the intermediate prediction errors corresponding to the g-th target set, and obtain the optimized prediction errors corresponding to the g-th target set, where g=1, 2, ..., Q, and Q refers to the total number of target sets; S740, traverse Q target sets, take the optimized prediction error corresponding to each target set as the corresponding intermediate prediction error, and return to execute step S720 until the optimized prediction errors corresponding to all target sets or the number of executions meet the third preset condition, and obtain the second prediction error corresponding to each target set.

8. The method for obtaining anomaly level based on error correction according to claim 1, characterized in that: S900 also includes the following steps: S910, for any subject to be identified, input the third prediction error and the credit anomaly target probability corresponding to the current subject to be identified into a preset mapping function to obtain a mapping result; S920, determining that the mapping result is a credit anomaly correction probability of the current subject to be identified; S930, obtaining the credit anomaly level corresponding to each subject to be identified according to the credit anomaly correction probability corresponding to each subject to be identified.

9. An abnormality level acquisition device based on error correction, characterized in that: The abnormality level acquisition device based on error correction includes: A data acquisition module is used to acquire N subjects to be identified, the credit anomaly target probability and credit anomaly prediction error corresponding to each subject to be identified, and the correlation level between the subjects to be identified, where N is an integer greater than 1; A reference set allocation module, used to allocate a corresponding reference set to each subject to be identified; A correlation level acquisition module, used to determine the correlation level between the first prediction error corresponding to each reference set and the reference set according to the correlation level between the credit anomaly prediction error corresponding to each subject to be identified and the subject to be identified; A temporary set selection module, used for randomly selecting a reference set from all reference sets as a temporary set; A reference set merging module, for any reference set, if the current reference set meets the second preset condition, adding the subject to be identified in the current reference set to the temporary set, deleting the current reference set, and using the temporary set as the reference set; a target set acquisition module, configured to return to execute the related level acquisition module until the number of reference sets converges to obtain Q reference sets, and determine the Q reference sets as Q target sets; A prediction error updating module, used for updating the first prediction error corresponding to each target set according to the first prediction error corresponding to each target set and the correlation level between the target sets, to obtain a second prediction error corresponding to each target set; A third prediction error acquisition module is used to determine, for any subject to be identified, a third prediction error corresponding to the current subject to be identified according to the second prediction error corresponding to the target set to which the current subject to be identified belongs; The credit anomaly level acquisition module is used to acquire the credit anomaly level corresponding to each subject to be identified based on the third prediction error and credit anomaly target probability corresponding to each subject to be identified.

10. A non-transitory computer-readable storage medium, wherein at least one instruction or at least one program is stored in the non-transitory computer-readable storage medium, characterized in that: The at least one instruction or the at least one program is loaded and executed by the processor to implement the abnormality level acquisition method based on error correction as described in any one of claims 1-8.

11. An electronic device, characterized in that: The invention comprises a processor and the non-transitory computer-readable storage medium as claimed in claim 10.

Images