Multi-model application method based on sub-model fusion and multi-level routing

Abstract

The invention provides a multi-model application method based on sub-model fusion and multi-level routing, and belongs to the technical field of financial management, and the method specifically comprises the following steps: according to a data deviation coefficient of a stable data source in an available data source group, matching conditions of user data of different historical credit users, determining a priority data source in the available data source group, constructing a credit model based on sub-models of different priority data sources, determining a routing hierarchy of a stable data source of the available data source group by using a data deviation coefficient with the priority data source, and when the priority data source is abnormal, determining the routing hierarchy of the stable data source of the available data source group. And switching to the stable data source of the next routing hierarchy in the corresponding available data source group, thereby improving the stability of the credit model.

Description

Technical Field

[0001] The present invention belongs to the technical field of financial management, and in particular relates to a multi-model application method based on sub-model fusion and multi-level routing. Background Art

[0002] In order to evaluate the credit risk of users, credit institutions often combine multiple data sources to evaluate the credit risk. In the invention patent application CN202410155274.1 "A method and device for querying and deciding on an external data source", the external data source is managed according to the historical approval data and yield of the credit application, that is, the credit risk of the user is identified and processed through multiple data sources.

[0003] In the actual credit risk assessment and processing process, data from dozens of third parties are often connected at the same time. The entire cooperative modeling process, from the initial sampling, outbound, data return, modeling, interface connection to the final model launch, may take 3-6 months. However, after the actual model is launched, due to business cooperation, unstable data sources, poor actual results, etc., a certain data source has poor performance or instability and needs to be re-modeled, which affects the online strategy and makes re-development time-consuming and labor-intensive.

[0004] In response to the above technical problems, the present application specifically provides a multi-model application method based on sub-model fusion and multi-level routing. Summary of the invention

[0005] To achieve the purpose of the present invention, the present invention adopts the following technical solutions: Specifically, the present application provides a multi-model application method based on sub-model fusion and multi-level routing, which specifically includes: S1 constructs sub-models for different data sources to call abnormal data and call delayed data in history to determine the stable data source of the data source; S2: based on the deviation of user data of different historical credit users, determine the data deviation coefficients between different stable data sources, divide the stable data sources into different data source groups using the data deviation coefficients, and determine the available data source groups in the data source groups using the constituent data of the stable data sources and the data deviation coefficients between them; S3: when the correlation of the historical call abnormal data of the stable data source in the available data source group meets the requirements, determine the priority data source in the available data source group according to the data deviation coefficient with the stable data source in the available data source group and the matching of the user data of different historical credit users; S4 builds a credit model based on sub-models of different priority data sources, and uses the data deviation coefficient with the priority data source to determine the routing level of the stable data source of the available data source group. When an abnormality occurs in the priority data source, it switches to the stable data source of the next routing level in the corresponding available data source group.

[0006] The beneficial effects of the present invention are: The data deviation coefficient with the priority data source is used to determine the routing level of the stable data source of the available data source group, thereby avoiding the technical problem of low stability of the credit model due to large data deviation in the process of switching the stable data source of the credit model, ensuring the stability of the credit data processing of the credit model, and also improving the processing reliability of credit applications.

[0007] When an exception occurs in the priority data source, it switches to the stable data source of the next routing level in the corresponding available data source group, thereby avoiding the technical problem that the construction of the credit model is difficult and time-consuming due to the frequent construction of the credit model. This improves the efficiency of switching the credit model, ensures the efficiency of processing credit applications, and reduces the impact of the credit model's data source on the credit model due to call exceptions.

[0008] A further technical solution is to construct sub-models for different data sources, including: Determining the input amount of the sub-model of the data source based on the user data of the data source; The sub-model is constructed based on the input quantity of the sub-model.

[0009] A further technical solution is that the historical call exception data includes the number of abnormal responses of the data source, wherein the number of abnormal responses includes the number of response deviations, the number of no responses, and the number of response timeouts.

[0010] A further technical solution is that the call delay data includes the call processing delay of the data source at different historical call times.

[0011] A further technical solution is that the method for determining the stable data source of the data source is: Determine the number of abnormal responses of the data source based on the historical call abnormal data of the data source, and determine the proportion of the number of abnormal responses of the data source based on the proportion of the number of abnormal responses; Based on the call delay data of the data source, determine the average value of the call processing delay of the data source at different historical call times, and determine the delay response coefficient of the data source according to the ratio of the average value of the call processing delay to a preset delay threshold; The response deviation factor of the data source is determined by an average value of the delayed response coefficient and the proportion of the number of abnormal responses, and the response deviation factor is used to determine whether the data source is a stable data source.

[0012] A further technical solution is that when the response deviation factor of the data source is greater than a preset deviation factor threshold, it is determined that the data source is not a stable data source.

[0013] A further technical solution is that the routing level of the stable data source of the available data source group is determined from large to small according to the data deviation coefficient between the stable data source and the priority data source in the available data source group.

[0014] A further technical solution is that the exception in the priority data source includes an exception in the call or a call delay that does not meet the requirement.

[0015] Other features and advantages will be described in the following description. The objects and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description and drawings.

[0016] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and other features and advantages of the present invention will become more apparent by describing in detail exemplary embodiments thereof with reference to the attached drawings.

[0018] Figure 1 is a flow chart of a multi-model application method based on sub-model fusion and multi-level routing; Figure 2 is a flow chart of a method for determining a stable data source of a data source; Figure 3 is a flow chart of a method for determining a coefficient of data deviation between stable data sources; Figure 4 is a flow chart of a method for determining an available data source group in a data source group; Figure 5 is a flow chart of a method for determining a priority data source among a group of available data sources. DETAILED DESCRIPTION

[0019] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0020] In the daily collaborative modeling process, we will connect to dozens of third-party data at the same time, and select 3-6 of them for cooperation based on the modeling effect. The entire collaborative modeling process, from the initial sampling, outbound, data return, modeling, interface connection, etc. to the final model launch, may take 3-6 months. However, after the actual model is launched, due to business cooperation, unstable data sources, poor actual results, etc., a data source has poor performance and instability, and needs to be re-modeled, which affects the online strategy and re-development is time-consuming and labor-intensive.

[0021] In this application, a sub-model is developed separately based on each data source, and then the final model is obtained through multiple model fusion methods. When there is a problem with a single data source, other routes can be quickly switched to improve accuracy and adaptability. At the same time, the design of model routing can save data costs and reduce losses caused by fluctuations in a single data source.

[0022] The proportion of abnormal responses of the data source is determined by the proportion of the number of abnormal responses. The delayed response coefficient of the data source is determined according to the ratio of the average value of the call processing delay to the preset delay threshold. The response deviation factor of the data source is determined by the average value of the proportion of the delayed response times and the number of abnormal responses. The data source with a response deviation factor less than 0.1 is regarded as a stable data source.

[0023] The data deviation coefficient between stable data sources is determined based on the proportion of user data with deviations among historical credit users from different stable data sources.

[0024] The available data source group is a data source group in which the number of stable data sources is greater than 5 and the average value of data deviation coefficients between different stable data sources is less than 0.05.

[0025] When there is no date on which the call delay of the stable data source in the available data source group does not meet the requirements, it is determined that the correlation of the historical call abnormal data of the stable data source in the available data source group meets the requirements.

[0026] The priority data source is the stable data source with the smallest average value of the data deviation coefficient with other stable data sources and the smallest average value of the proportion of the number of deviation information items in the user data of historical credit users.

[0027] Example 1 like Figure 1 As shown, the present application provides a multi-model application method based on sub-model fusion and multi-level routing, which specifically includes: S1 constructs sub-models for different data sources to call abnormal data and call delayed data in history to determine the stable data source of the data source; S2: based on the deviation of user data of different historical credit users, determine the data deviation coefficients between different stable data sources, divide the stable data sources into different data source groups using the data deviation coefficients, and determine the available data source groups in the data source groups using the constituent data of the stable data sources and the data deviation coefficients between them; S3: when the correlation of the historical call abnormal data of the stable data source in the available data source group meets the requirements, determine the priority data source in the available data source group according to the data deviation coefficient with the stable data source in the available data source group and the matching of the user data of different historical credit users; S4 builds a credit model based on sub-models of different priority data sources, and uses the data deviation coefficient with the priority data source to determine the routing level of the stable data source of the available data source group. When an abnormality occurs in the priority data source, it switches to the stable data source of the next routing level in the corresponding available data source group.

[0028] Furthermore, sub-models are constructed for different data sources, including: Determining the input amount of the sub-model of the data source based on the user data of the data source; The sub-model is constructed based on the input quantity of the sub-model.

[0029] Specifically, the historical call exception data includes the number of response exceptions of the data source, wherein the number of response exceptions includes the number of response deviations, the number of no responses, and the number of response timeouts.

[0030] It should be noted that the call delay data includes the call processing delay of the data source at different historical call times.

[0031] It is understandable that if Figure 2 As shown, the method for determining the stable data source of the data source is: Determine the number of abnormal responses of the data source based on the historical call abnormal data of the data source, and determine the proportion of the number of abnormal responses of the data source based on the proportion of the number of abnormal responses; Based on the call delay data of the data source, determine the average value of the call processing delay of the data source at different historical call times, and determine the delay response coefficient of the data source according to the ratio of the average value of the call processing delay to a preset delay threshold; The response deviation factor of the data source is determined by an average value of the delayed response coefficient and the proportion of the number of abnormal responses, and the response deviation factor is used to determine whether the data source is a stable data source.

[0032] Further, when the response deviation factor of the data source is greater than a preset deviation factor threshold, it is determined that the data source is not a stable data source.

[0033] In another possible embodiment, the method for determining the stable data source of the data source is: Determine the number of abnormal responses of the data source based on the historical call abnormal data of the data source, and determine the proportion of the number of abnormal responses of the data source based on the proportion of the number of abnormal responses; Based on the call delay data of the data source, determine the average value of the call processing delay of the data source at different historical call times; The response deviation factor of the data source is determined by multiplying the average value of the call processing delay by the proportion of the number of abnormal responses, and the response deviation factor is used to determine whether the data source is a stable data source.

[0034] Specifically, Figure 3 As shown, the method for determining the data deviation coefficient between the stable data sources is: Determine the number of information items in which different stable data sources have deviations in the user data of different historical credit users based on the deviations in the user data of different historical credit users; Determining the deviated credit users among the historical credit users by using the number of deviated information items; According to the proportion of the deviation credit users in the historical credit users, the data deviation coefficients between different stable data sources are determined.

[0035] Furthermore, the deviated credit user is a historical credit user whose number of deviated information items does not meet the requirements.

[0036] It is understandable that the stable data sources are divided into different data source groups, specifically including: Stable data sources whose data deviation coefficients are within a preset deviation coefficient range are divided into the same data source group.

[0037] Specifically, Figure 4As shown, the method for determining the available data source group in the data source group is: Determine the number of stable data sources in the data source group by using the constituent data of the stable data sources in the data source group; Determine the average deviation coefficient of the data source group by taking the average value of the data deviation coefficients between the stable data sources in the data source group; A group availability factor of the data source group is determined according to the number of the stable data sources and the average deviation coefficient, and whether the data source group is an available data source group is determined based on the group availability factor.

[0038] Furthermore, the group availability factor of the data source group is a ratio of the number of stable data sources to the average deviation coefficient.

[0039] It should be noted that when the group availability factor is greater than a preset availability factor threshold, the data source group is determined to be an available data source group.

[0040] In another possible embodiment, the method for determining the available data source group in the data source group is: Determine the number of stable data sources in the data source group by using the constituent data of the stable data sources in the data source group; Based on the data deviation coefficients between the stable data sources in the data source group, determine the stable data source whose data deviation coefficient with other stable data sources is not within the preset range, and use it as the internal deviation data source; A group availability coefficient of the data source group is determined according to a ratio of the number of the stable data sources to the number of the internal deviation data sources, and whether the data source group is an available data source group is determined based on the group availability coefficient.

[0041] Further, when the group availability coefficient of the data source group is within a preset availability coefficient interval, the data source group is determined to be an available data source group.

[0042] Specifically, it is determined that the correlation of the historical call abnormal data of the stable data source meets the requirements, including: Based on the historical call exception data of the stable data sources in the available data source group, determining the distribution data of the number of abnormal responses of different stable data sources in different time periods; Based on the distribution data of the number of abnormal responses of different stable data sources in different time periods, the number of abnormal responses in different time periods is determined, and the response abnormality time periods of different stable data sources are determined by using the number of abnormal responses; According to the correlation of the response exception time periods of different stable data sources, the number of stable data sources belonging to the response exception time periods in different time periods is determined, and the number of stable data sources belonging to the response exception time periods in different time periods is used to determine whether the correlation of the historical call exception data of the stable data source meets the requirements.

[0043] Further, when the number of abnormal responses in the time period is greater than a preset abnormal response number threshold, the time period is determined to be a response abnormality time period.

[0044] A further technical solution is that when the proportion of the number of stable data sources belonging to the response abnormal period in different time periods in the available data source group does not meet the requirements, it is determined that the correlation of the historical call abnormal data of the stable data source does not meet the requirements.

[0045] It is understandable that when the correlation of the historical call abnormal data of the stable data source does not meet the requirements, there is no need to determine the priority data source of the available data source group.

[0046] In another possible embodiment, determining that the correlation of the historical call abnormal data of the stable data source meets the requirements specifically includes: Based on the historical call exception data of the stable data sources in the available data source group, determining the distribution data of the number of abnormal responses of different stable data sources in different time periods; Determine the number of abnormal responses in different time periods based on distribution data of the number of abnormal responses in different time periods from different stable data sources; Determine the sum of the number of abnormal responses of different stable data sources in different time periods and use it as the sum of abnormal numbers. Use the sum of abnormal numbers in different time periods to determine whether the correlation of historical call abnormal data of the stable data source meets the requirements.

[0047] Further, when the number of exceptions and the number of time periods within a preset number of exceptions interval are less than a preset value of the number of time periods, it is determined that the correlation of the historical call exception data of the stable data source does not meet the requirements.

[0048] In another possible embodiment, determining that the correlation of the historical call abnormal data of the stable data source meets the requirements specifically includes: Based on the historical call exception data of the stable data sources in the available data source group, determining the distribution data of the number of abnormal responses of different stable data sources in different time periods; Based on the distribution data of the number of abnormal responses of different stable data sources in different time periods, the number of abnormal responses in different time periods is determined, and the sum of the number of abnormal responses of different stable data sources in different time periods is determined, and the sum is used as the sum of abnormal numbers. When the sum of abnormal numbers in different time periods does not meet the requirements, it is determined that the correlation of the historical call abnormal data of the stable data source does not meet the requirements; When there are abnormal times and the required time period is met: When it is determined by using the number of abnormal responses that different stable data sources do not have a response abnormal period, it is determined that the correlation of the historical call abnormal data of the stable data source meets the requirements; When using the number of abnormal responses to determine whether there is a stable data source with abnormal response periods: The number of stable data sources with response exception periods is obtained, and when the proportion of the number of stable data sources with response exception periods is less than the proportion of the preset number of stable data sources, it is determined that the correlation of the historical call exception data of the stable data source does not meet the requirements; When the proportion of stable data sources that respond to abnormal periods is not less than the preset proportion of stable data sources: According to the correlation of the response abnormal time periods of different stable data sources, the number of stable data sources belonging to the response abnormal time period in different time periods is determined, and when the number of stable data sources belonging to the response abnormal time period in different time periods is determined by using the number of stable data sources belonging to the response abnormal time period, the proportion of the number of stable data sources belonging to the response abnormal time period that does not meet the requirements is greater than the preset number of time periods, it is determined that the correlation of the historical call abnormal data of the stable data source does not meet the requirements; Determine that the number of stable data sources belonging to the response abnormal period does not meet the requirement and the number of time periods is not greater than the preset number of time periods: Based on the number of abnormal responses of different stable data sources in different time periods, the response abnormality coefficients of different time periods are determined, and the response abnormality coefficients of different time periods are used to determine whether the correlation of the historical call abnormal data of the stable data source meets the requirements.

[0049] Further, when the number of time periods in which the response anomaly coefficient is within the preset anomaly coefficient range is greater than the preset number of time periods, it is determined that the correlation of the historical call anomaly data of the stable data source does not meet the requirements.

[0050] Specifically, the matching status of the user data of the historical credit users includes the distribution data of the deviation information items and the missing information items of the user data of different historical credit users of the stable data source.

[0051] It should be noted that if Figure 5 As shown, the method for determining the priority data source in the available data source group is: Determine the mean value of the data deviation coefficient of the stable data source by taking the average value of the data deviation coefficients of the stable data source in the available data source group and other stable data sources; According to the matching of the stable data source with the user data of different historical credit users, the distribution data of the deviation information items and the missing information items among different historical credit users are determined, and the distribution data are used to determine the proportion of the number of deviation information items or missing information items among the historical credit users, which is used as the proportion of information deviation users; The adaptation deviation of the stable data source is determined by using the mean value of the data deviation coefficient and the average value of the proportion of the number of information deviation users, and based on the adaptation deviation, it is determined whether the stable data source is a priority data source in the available data source group.

[0052] Furthermore, the priority data source in the available data source group is a stable data source with the smallest adaptation deviation.

[0053] Optionally, the method for determining the priority data source in the available data source group is: S41 determines the distribution data of deviation information items and missing information items among different historical credit users according to the matching conditions of the stable data source with the user data of different historical credit users, and determines the number ratio of deviation information items or missing information items among the historical credit users by using the distribution data, and determines the data matching deviation coefficient of the data source in combination with the number ratio of deviation information items and missing information items among different historical credit users; S42 determines the data deviation amount of the stable data source by using the data deviation coefficient of the stable data source in the available data source group and other stable data sources; S43 determines the adaptation deviation of the stable data source by using the data deviation and the average value of the data matching deviation coefficient, and determines whether the stable data source is a priority data source in the available data source group based on the adaptation deviation.

[0054] Optionally, the above step S41 includes the following contents: S411 determines the distribution data of deviation information items and missing information items among different historical credit users according to the matching of the stable data source with the user data of different historical credit users, and uses the distribution data to determine the ratio of the number of deviation information items or missing information items among the historical credit users, and uses it as the ratio of information deviation users. When the ratio of information deviation users of the stable data source does not meet the requirement, it is determined that the stable data source does not belong to the priority data source in the available data source group. When the ratio of information deviation users of the stable data source meets the requirement, the process proceeds to step S412. S412 determines the quantity ratio of the deviation information items and the missing information items among different historical credit users based on the distribution data of the deviation information items and the missing information items among different historical credit users, and uses it as the information deviation quantity ratio. When there are historical credit users whose information deviation quantity ratio does not meet the requirement, the process proceeds to step S413. When there are no historical credit users whose information deviation quantity ratio does not meet the requirement, the process proceeds to step S414. S413: When the proportion of historical credit users whose information deviation ratio does not meet the requirements is greater than the proportion of deviation users, it is determined that the stable data source does not belong to the priority data source in the available data source group; when the proportion of historical credit users whose information deviation ratio does not meet the requirements is not greater than the proportion of deviation users, the process proceeds to step S414; S414 uses the distribution data to determine the percentage of deviation information items or missing information items among the historical credit users, and determines the data matching deviation coefficient of the data source based on the percentage of deviation information items and missing information items among different historical credit users. When the data matching deviation coefficient of the data source does not meet the requirements, it is determined that the stable data source does not belong to the priority data source in the available data source group. When the data matching deviation coefficient of the data source meets the requirements, proceed to step S42.

[0055] Optionally, the above step S42 includes the following contents: S421: When the data matching deviation coefficient of the data source is within the preset matching deviation coefficient interval, the process proceeds to step S422; when the data matching deviation coefficient of the data source is not within the preset matching deviation coefficient interval, the process proceeds to step S424; S422: When it is determined that there is another stable data source whose data deviation coefficient does not meet the requirement among the stable data sources based on the data deviation coefficients of the stable data sources in the available data source group and other stable data sources, the process proceeds to step S423; when there is no other stable data source whose data deviation coefficient does not meet the requirement among the stable data sources, the process proceeds to step S424; S423: when the number of other stable data sources whose data deviation coefficients do not meet the requirements is greater than the threshold number of stable data sources, it is determined that the stable data source does not belong to the priority data source in the available data source group; when the number of other stable data sources whose data deviation coefficients do not meet the requirements is not greater than the threshold number of stable data sources, the process proceeds to step S424; S424 determines the data deviation of the stable data source based on the data deviation coefficient between the stable data source in the available data source group and other stable data sources. When the data deviation of the stable data source does not meet the requirement, the process proceeds to step S425. When the data deviation of the stable data source meets the requirement, the process proceeds to step S43. S425 When the data matching deviation coefficient of the data source is within the preset matching deviation coefficient range, it is determined that the stable data source does not belong to the priority data source in the available data source group; when the data matching deviation coefficient of the data source is not within the preset matching deviation coefficient range, the process proceeds to step S43.

[0056] It should be noted that the credit model is constructed based on sub-models of different priority data sources, including: The sub-models of different priority data sources are combined to obtain the credit model.

[0057] Further, the routing level of the stable data source of the available data source group is determined from large to small according to the data deviation coefficient between the stable data source and the priority data source in the available data source group.

[0058] Specifically, the exception of the priority data source includes that a call is abnormal or a call delay does not meet the requirement.

[0059] Example 2 In another possible embodiment, the method for determining the available data source group in the data source group is: Determine the number of stable data sources in the data source group by using the constituent data of the stable data sources in the data source group, and when the number of the stable data sources is less than the preset number of data sources, determine that the data source group does not belong to the available data source group; When the number of the stable data sources is not less than the preset number of data sources: Determine the average deviation coefficient of the data source group by taking the average value of the data deviation coefficients between the stable data sources in the data source group, and when the average deviation coefficient of the data source group does not meet the requirement, determine that the data source group does not belong to the available data source group; When the average deviation coefficient of the data source group meets the requirements: When the average deviation coefficient of the data source group is less than the preset coefficient threshold: then determining that the data source group belongs to an available data source group; When the average deviation coefficient of the data source group is not less than the preset coefficient threshold: Based on the data deviation coefficients between the stable data sources in the data source group, a stable data source whose data deviation coefficient with other stable data sources is not within a preset range is determined, and the stable data source is used as an internal deviation data source; when the proportion of the internal deviation data source in the stable data source does not meet the requirement, it is determined that the data source group does not belong to the available data source group; When the proportion of the internal deviation data source in the stable data source meets the requirement: taking the stable data sources except the internal deviation data source as available data sources, and determining that the data source group does not belong to the available data source group when the number of available data sources in the data source group does not meet the requirement; When the number of available data sources in the data source group meets the requirement: The group availability factor of the data source group is determined according to the number of the stable data sources and the average deviation coefficient, and combined with the proportion of the internal deviation data sources in the stable data sources, and whether the data source group is an available data source group is determined based on the group availability factor.

[0060] Example 3 Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, equipment, and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0061] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0062] The above description is only one or more embodiments of this specification and is not intended to limit this specification. For those skilled in the art, one or more embodiments of this specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of one or more embodiments of this specification shall be included in the scope of the claims of this specification.

Claims

1. A multi-model application method based on sub-model fusion and multi-level routing, characterized in that: Specifically include: Construct sub-models for different data sources to call abnormal data and call delayed data in history to determine the stable data source of the data source; Based on the deviation of user data of different historical credit users, determine the data deviation coefficients between different stable data sources, divide the stable data sources into different data source groups using the data deviation coefficients, and determine the available data source groups in the data source groups using the constituent data of the stable data sources and the data deviation coefficients between each other; When the correlation of the historical call abnormal data of the stable data source in the available data source group meets the requirements, the priority data source in the available data source group is determined according to the data deviation coefficient with the stable data source in the available data source group and the matching of the user data of different historical credit users; A credit model is constructed based on sub-models of different priority data sources. The routing level of the stable data source in the available data source group is determined by utilizing the data deviation coefficient with respect to the priority data source. When an abnormality occurs in the priority data source, the stable data source of the next routing level in the corresponding available data source group is switched.

2. The multi-model application method based on sub-model fusion and multi-level routing as claimed in claim 1, characterized in that: Construct sub-models for different data sources, including: Determining the input amount of the sub-model of the data source based on the user data of the data source; The sub-model is constructed based on the input quantity of the sub-model.

3. The multi-model application method based on sub-model fusion and multi-level routing as claimed in claim 1, characterized in that: The historical call exception data includes the number of response exceptions of the data source, wherein the number of response exceptions includes the number of response deviations, the number of no responses, and the number of response timeouts.

4. The multi-model application method based on sub-model fusion and multi-level routing as claimed in claim 1, characterized in that: The call delay data includes the call processing delay of the data source at different historical call times.

5. The multi-model application method based on sub-model fusion and multi-level routing as claimed in claim 1, characterized in that: The method for determining the stable data source of the data source is: Determine the number of abnormal responses of the data source based on the historical call abnormal data of the data source, and determine the proportion of the number of abnormal responses of the data source based on the proportion of the number of abnormal responses; Based on the call delay data of the data source, determine the average value of the call processing delay of the data source at different historical call times, and determine the delay response coefficient of the data source according to the ratio of the average value of the call processing delay to a preset delay threshold; The response deviation factor of the data source is determined by an average value of the delayed response coefficient and the proportion of the number of abnormal responses, and the response deviation factor is used to determine whether the data source is a stable data source.

6. The multi-model application method based on sub-model fusion and multi-level routing as claimed in claim 5, characterized in that: When the response deviation factor of the data source is greater than a preset deviation factor threshold, it is determined that the data source is not a stable data source.

7. The multi-model application method based on sub-model fusion and multi-level routing as claimed in claim 1, characterized in that: The method for determining the priority data source in the available data source group is: Determine the mean value of the data deviation coefficient of the stable data source by taking the average value of the data deviation coefficients of the stable data source in the available data source group and other stable data sources; According to the matching of the stable data source with the user data of different historical credit users, the distribution data of the deviation information items and the missing information items among different historical credit users are determined, and the distribution data are used to determine the proportion of the number of deviation information items or missing information items among the historical credit users, which is used as the proportion of information deviation users; The adaptation deviation of the stable data source is determined by using the mean value of the data deviation coefficient and the average value of the proportion of the number of information deviation users, and based on the adaptation deviation, it is determined whether the stable data source is a priority data source in the available data source group.

8. The multi-model application method based on sub-model fusion and multi-level routing as claimed in claim 1, characterized in that: The priority data source in the group of available data sources is a stable data source with the smallest adaptation deviation.

9. The multi-model application method based on sub-model fusion and multi-level routing as claimed in claim 1, characterized in that: The routing level of the stable data source of the available data source group is determined from large to small according to the data deviation coefficient between the stable data source and the priority data source in the available data source group.

10. The multi-model application method based on sub-model fusion and multi-level routing as claimed in claim 1, characterized in that: The exception of the priority data source includes an exception in the call or a call delay that does not meet the requirement.

Images