A method for solving the inconsistency between offline and online pedestrian features

Abstract

This invention provides a method for solving the inconsistency problem between offline and online pedestrian characteristics, belonging to the field of data processing technology. Specifically, it includes: determining the update processing scheme for the compared risk users based on the updated matching users and the risk change feature types of the compared risk users; and determining the change results of different update processing schemes for the compared risk users based on the update data of the comparison results of offline and online models in the risk change feature types, thereby improving the efficiency and reliability of the consistency verification process between offline and online models.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, and in particular relates to a method for solving the problem of inconsistency between online and offline pedestrian characteristics. Background Technology

[0002] Human characteristics (such as user identity, behavioral attributes, credit records, etc.) are core data for offline modeling and online real-time risk control decision-making. They are frequently used in the risk control field when conducting user risk assessment. Specific technical solutions are given in CN202111410257.0 "A method for mining credit report features" and CN202110547959.7 "A method and device for processing credit authorization queries".

[0003] The current industry generally adopts a dual-link independent development model for the People's Bank of China (PBOC), namely, two systems: offline and real-time. This model leads to the following key problems:

[0004] 1) Severe inefficiency. Two independent development processes mean that two waves of developers, one offline and one real-time, are needed to conduct two tests, which greatly consumes manpower and time. Moreover, the testing phase not only needs to be verified in the test environment, but also relies heavily on T+1 comparison in production. Once a problem is found, the repair process is lengthy, which seriously affects the project's progress and business timeliness.

[0005] 2) Quality is difficult to guarantee. Two separate development processes are highly prone to inconsistency issues. Firstly, differing understandings of pedestrian feature requirements in different scenarios lead to inconsistent development directions. Secondly, varying language proficiency among developers further exacerbates the differences in implementation. Finally, the limited number of test cases for pedestrian reporting scenarios in the testing environment makes it difficult to comprehensively detect potential problems, ultimately resulting in inconsistencies in pedestrian feature data between offline and online scenarios, seriously threatening the accuracy and reliability of risk control decisions.

[0006] To address the aforementioned technical problems, this application merges the offline and online systems into a single system. To ensure consistency between the calculation results of the offline and online systems when pedestrian characteristics change in the online system, the update direction of different users' pedestrian feature combinations is limited. Updates are only performed when certain features change, ensuring that different users have a large number of reference users with consistent feature distributions when updating features in the offline system. This effectively verifies the consistency of different types of feature combinations, guaranteeing the consistency of pedestrian feature processing between the offline and online systems. Specifically, this application provides a method to solve the problem of inconsistency between offline and online pedestrian features. Summary of the Invention

[0007] To achieve the objectives of this invention, the following technical solution is adopted:

[0008] Specifically, this application provides a method for resolving the inconsistency between online and offline pedestrian characteristics, which includes:

[0009] S1 constructs an integrated feature calculation model based on pedestrian characteristics. Based on the distribution data of pedestrian characteristics of users in the offline system, it determines the reference users for feature changes of users. Based on the distribution of reference users for feature changes under the feature change type, it determines the comparison risk users and the risk change feature types among the users.

[0010] S2 determines the comparison risk users that are updated only in specific risk change feature types based on the comparison results of the risk change feature types matched by the comparison risk users and the offline and online models of the feature change reference users, and uses them as the updated matching users.

[0011] S3 determines the update processing scheme for the compared risk users based on the updated matching users in the risk change feature type and the risk change feature type matched by the compared risk users. Based on the update data of the comparison results of offline and online models in the risk change feature type, the change result of the update processing scheme for the compared risk users is determined.

[0012] The beneficial effects of this invention are as follows:

[0013] Based on the comparison results of offline and online models of risk change feature types matched with risk users and reference users for feature changes, risk users are identified that are updated only in specific risk change feature types. This avoids random update processing for different risk users. Due to the small number of reference users, the consistency of certain combinations of human characteristics between offline and online models cannot be effectively verified. By limiting the specific risk change feature types, the update processing direction of risk users is restricted, ensuring that more reference users are retained in certain combinations of human characteristics, thereby improving the efficiency of consistency verification between offline and online models.

[0014] Based on the updated data of the comparison results between offline and online models in the risk change feature types, the changes in the update processing schemes for different comparison risk users are determined. This not only takes into account the changes in the consistency and reliability of the risk change feature types caused by the updated data of the comparison results between offline and online models, but also combines the updated data of the matching risk change feature types of the comparison risk users. This enables the identification of comparison risk users with more serious changes in the matching risk change feature types, ensuring the consistency and update reliability of the offline system and the People's Bank of China's online system.

[0015] Furthermore, the integrated feature calculation model is a calculation model obtained by jointly constructing an offline model and an online model of pedestrian features, that is, the offline model and the online model use the same pedestrian feature calculation model.

[0016] Furthermore, the user's feature change reference user is a user in the offline system who has the same human characteristics as the user but also has human characteristics that the user does not have. It can be understood that the user has human characteristics, and the feature change reference user has the same human characteristics, but has 1 to 4 human characteristics that the user does not have.

[0017] Furthermore, the method for determining risky users through comparison is as follows:

[0018] Based on the reference users for feature changes, determine the reference users for feature changes of the user in the absence of human characteristics;

[0019] Based on the number of reference users whose characteristics change under different missing human characteristics, determine whether the user is a comparison risk user.

[0020] Furthermore, the method for determining the change results of the update processing scheme for the risky users is as follows:

[0021] Using the updated data of the comparison results of offline and online models in different risk change characteristic types, the number of users whose comparison results of offline and online models are consistent in different risk change characteristic types is determined and used as the number of consistent users. Based on the number of consistent users and the number of inconsistent users, the updated change characteristic type in different risk change characteristic types is determined.

[0022] Based on the updated data in the online system of the People's Bank of China for comparing risk users, determine the number of missing features in the offline system of the People's Bank of China for comparing risk users, and use them as the number of updated features;

[0023] Based on the number of update change feature types in different risk change types and the number of update features of the compared risk users, the change result of the update processing scheme for the compared risk users is determined.

[0024] Other features and advantages will be set forth in the following description, and the objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0025] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0026] The above and other features and advantages of the present invention will become more apparent from a detailed description of exemplary embodiments thereof with reference to the accompanying drawings.

[0027] Figure 1 This is a flowchart of a method for resolving the inconsistency between pedestrian characteristics and offline / online features.

[0028] Figure 2 This is a flowchart illustrating the method for identifying risky users through user comparison.

[0029] Figure 3 This is a flowchart of the method for updating and matching users;

[0030] Figure 4 This is a flowchart illustrating the method for determining the update handling plan for risky users. Detailed Implementation

[0031] 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 with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0032] In the field of risk control, user characteristics (such as user identity, behavioral attributes, and credit history) are the core data for offline modeling and online real-time risk control decisions. Currently, the industry generally adopts a dual-link independent development model (offline and online), which leads to the following key problems:

[0033] 1) Severe inefficiency. Two independent development processes mean that two waves of developers, one offline and one real-time, are needed to conduct two tests, which greatly consumes manpower and time. Moreover, the testing phase not only needs to be verified in the test environment, but also relies heavily on T+1 comparison in production. Once a problem is found, the repair process is lengthy, which seriously affects the project's progress and business timeliness.

[0034] 2) Quality is difficult to guarantee. Two separate development processes are highly prone to inconsistency issues. Firstly, differing understandings of pedestrian feature requirements in different scenarios lead to inconsistent development directions. Secondly, varying language proficiency among developers further exacerbates the differences in implementation. Finally, the limited number of test cases for pedestrian reporting scenarios in the testing environment makes it difficult to comprehensively detect potential problems, ultimately resulting in inconsistencies in pedestrian feature data between offline and online scenarios, seriously threatening the accuracy and reliability of risk control decisions.

[0035] Based on this, this application determines, according to the distribution data of users' pedestrian characteristics, that when changes occur in the online pedestrian system, some users can only perform changes in a predetermined direction. This ensures that other users have a sufficient number of reference users with consistent feature distributions to perform joint verification when updating the offline pedestrian characteristics system. This improves the efficiency of verifying the consistency between the offline and online models during pedestrian characteristic updates, including the following:

[0036] 1. Core logic: By defining data from the same source and adapting to heterogeneous frameworks, it achieves "one set of logic and two-state execution".

[0037] Same source definition: Abstract pedestrian feature calculation business domain model, supporting offline batch processing and online stream processing dynamic parsing.

[0038] Framework Adaptation Layer: Design an intermediate adaptation layer to automatically switch computing modes according to the scenario. For example, in online scenarios, convert HiveQL to H2SQL to meet real-time requirements.

[0039] Technological Breakthrough: Resolving differences in syntax, functions, and execution between heterogeneous frameworks, enabling "one-time development, seamless online and offline operation" for pedestrian feature processing, providing businesses with highly consistent, low-latency, and low-cost feature computing capabilities.

[0040] 2. Pedestrian Feature Replay Engine

[0041] Accurate reproduction: Based on historical data, the feature calculation process is traced back, supporting comparison and verification in offline and online environments.

[0042] Time-series replay: Records the original data and computation context (such as timestamps and dependencies), and uses the timestamp-driven engine to replay the data in the order of events, ensuring complete consistency with the real-time computation logic.

[0043] Difference localization: The built-in difference analysis module automatically identifies deviations from the online calculation results, that is, the degree of consistency between the offline model and the online model in the calculation and processing, to assist in rapid repair.

[0044] 3. Online and Offline Consistency Alignment Mechanism for Pedestrians

[0045] By employing intelligent algorithms and data analysis, it monitors and corrects for discrepancies between online and offline data across time and space in real time. This overcomes complex issues such as data transmission delays and format differences, ensuring that feature data remains consistent throughout its entire lifecycle and laying a solid foundation for the risk control system.

[0046] Uniqueness of the solution and its industry value

[0047] Technological integration and innovation: Unlike the traditional "dual-link development" model, this solution achieves the unification of online and offline computing logic for the first time through the domain-driven approach of pedestrian feature data and the design of the framework adaptation layer, reducing development costs by more than 60%.

[0048] Dynamic interpretability: The feature replay engine not only solves the consistency problem, but also provides root cause analysis of deviations, enhancing the auditability of the risk control system and meeting financial regulatory requirements.

[0049] This solution can be widely used in fields such as financial risk control and anti-fraud, effectively solving the model drift problem caused by inconsistencies between features online and offline.

[0050] Example 1

[0051] like Figure 1 As shown, this application provides a method for solving the problem of inconsistency between online and offline pedestrian characteristics, specifically including:

[0052] S1 constructs an integrated feature calculation model based on pedestrian characteristics. Based on the distribution data of pedestrian characteristics of users in the offline system, it determines the reference users for feature changes of users. Based on the distribution of reference users for feature changes under the feature change type, it determines the comparison risk users and the risk change feature types among the users.

[0053] Furthermore, the integrated feature calculation model is a calculation model jointly constructed from offline and online models of pedestrian features. That is, the offline and online models use the same pedestrian feature calculation model, but the data sources of the pedestrian features obtained by the offline and online models are different. The offline model obtains the features from the offline system, and the online model obtains them from the online system of the People's Bank of China. Since the data structure and type of the online and offline systems of the People's Bank of China are different, their data processing types are also different. Therefore, during the update process, that is, when updating the data from the offline system to the online system, in order to verify the consistency of the integrated feature calculation model, it is necessary to reserve some users from updating. This allows other users to effectively perform consistency verification when updating. When there is a problem with the consistency verification result, it is possible to determine whether the problem lies in the feature calculation of the integrated feature calculation model, that is, whether there is a problem with the consistency of the feature calculation or with the reading results of the data port.

[0054] Furthermore, the user's feature change reference user is a user in the offline system who has the same human characteristics as the user but also has human characteristics that the user does not have. It can be understood that the user has human characteristics, and the feature change reference user has the same human characteristics, but has 1 to 4 human characteristics that the user does not have.

[0055] Specifically, such as Figure 2 As shown, the method for determining risky users through comparison is as follows:

[0056] Based on the reference users for feature changes, determine the reference users for feature changes of the user in the absence of human characteristics;

[0057] Based on the number of reference users whose characteristics change under different missing human characteristics, determine whether the user is a comparison risk user.

[0058] It should be noted that the feature variation type is the missing human body feature.

[0059] It should be noted that the reference user for feature changes under the missing pedestrian features is the same as the user except that the missing pedestrian features are not present.

[0060] It is understood that when a user has multiple missing human characteristics where the number of reference users for feature changes does not meet the requirements, the user is determined to be a comparison risk user. In one possible embodiment, when a user has more than three missing human characteristics where the number of reference users for feature changes is less than ten, the user is determined to be a comparison risk user.

[0061] It should be noted that when the user is not a risk user, the user's offline system will be updated in real time after the People's Bank of China's online system is updated. After the update, the consistency between the calculation results of the features of the offline model and the calculation results of the features of the online model will be verified.

[0062] In another embodiment, the method for determining the risky users by comparison is as follows:

[0063] Based on the reference users for feature changes, determine the reference users for feature changes of the user in the absence of human characteristics;

[0064] The mean number of reference users for feature changes under different missing pedestrian features is determined based on the number of reference users for feature changes under different missing pedestrian features.

[0065] Based on the average number of reference users whose feature changes under different missing human characteristics, it is determined whether the user is a comparison risk user.

[0066] It is understandable that when the average number of reference users for feature changes under different missing human characteristics is less than 8, the user is determined to be a comparison risk user.

[0067] S2 determines the comparison risk users that are updated only in specific risk change feature types based on the comparison results of the risk change feature types matched by the comparison risk users and the offline and online models of the feature change reference users, and uses them as the updated matching users.

[0068] Furthermore, the risk change characteristic type is a missing human characteristic where the number of reference users for risk change does not meet the requirements.

[0069] Specifically, such as Figure 3 As shown, the method for determining the updated matching user is as follows:

[0070] Based on the risk change feature types matched by the risk comparison users, determine the number of risk change feature types matched by the risk comparison users;

[0071] Based on the comparison results of offline and online models of feature change reference users with different risk change characteristic types, the number of feature change reference users with inconsistent comparison results is determined and they are regarded as the number of inconsistent users.

[0072] Based on the number of inconsistent users with different risk change characteristics, determine whether the compared risk users are update matching users.

[0073] It is understood that when the number of risk change feature types of the compared risk users does not meet the requirements, that is, when the number of risk change feature types of the compared risk users is greater than the preset feature type number threshold, in one possible embodiment, when the number of risk change feature types of the compared risk users is more than 6, it is determined that the compared risk users do not belong to the update matching users.

[0074] Furthermore, when the number of risk change feature types of the compared risk users meets the requirements, that is, when the number of risk change feature types of the compared risk users is not greater than the preset feature type number threshold, it is also necessary to further determine whether the difference between the number of users with the most inconsistent risk change feature type and the number of users with other risk change feature types is greater than the preset deviation threshold. If so, it is determined that the compared risk user belongs to the update matching user; if not, it is determined that the compared risk user does not belong to the update matching user.

[0075] In one possible embodiment, when the difference between the number of users with the most inconsistent risk change feature type and the number of users with other risk change feature types is greater than 2, the user being compared is determined to be an updated matching user.

[0076] It is understood that the specific risk change characteristic type is the risk change characteristic type with the largest number of inconsistent users compared to the risk users.

[0077] S3 determines the update processing scheme for the compared risk users based on the updated matching users in the risk change feature types and the risk change feature types matched by the compared risk users;

[0078] Specifically, such as Figure 4 As shown, the method for determining the update processing scheme for comparing risky users is as follows:

[0079] Based on the risk change characteristic types matched with the risk users mentioned above, determine the number of matching characteristic types;

[0080] Based on the updated matching users corresponding to different matching feature types, determine the number of updated matching users for each matching feature type.

[0081] Based on the number of updated matching users and the number of matching feature types, an update processing scheme for the comparison of risky users is determined.

[0082] It should be noted that the updated matching users corresponding to the matching feature type are comparison risk users whose specific risk change feature type is the matching feature type.

[0083] It is understood that, based on the number of updated matching users and the number of matching feature types, the update processing scheme for the comparison of high-risk users is determined, specifically including:

[0084] Based on the number of matching feature types, determine the number of matching feature types corresponding to the update processing of the compared risk users;

[0085] Based on the number of matching feature types corresponding to the update processing of the risk users, and according to the number of updated matching users corresponding to different matching feature types from small to large, the update processing scheme for the risk users is determined.

[0086] It should be noted that the number of matching feature types corresponding to the update processing of risky users is determined by the product of the number of matching feature types and a preset ratio factor. In one possible embodiment, it is determined by the product of the number of matching feature types of risky users and 0.3.

[0087] Furthermore, by using the number of corresponding updated matching users from small to large, the number of matching feature types corresponding to the updated processing of risky users is selected.

[0088] S4 determines the changes in the update processing plan for users with compared risks based on the updated data of the comparison results between offline and online models in the risk change characteristic type.

[0089] Furthermore, the method for determining the change results of the update processing scheme for the risky users is as follows:

[0090] Using the updated data of the comparison results of offline and online models in different risk change characteristic types, the number of users whose comparison results of offline and online models are consistent in different risk change characteristic types is determined and used as the number of consistent users. Based on the number of consistent users and the number of inconsistent users, the updated change characteristic type in different risk change characteristic types is determined.

[0091] Based on the updated data in the online system of the People's Bank of China for comparing risk users, determine the number of missing features in the offline system of the People's Bank of China for comparing risk users, and use them as the number of updated features;

[0092] Based on the number of update change feature types in different risk change types and the number of update features of the compared risk users, the change result of the update processing scheme for the compared risk users is determined.

[0093] It should be noted that the updated change feature type is the risk change feature type in which the difference between the number of consistent users and the number of inconsistent users meets the requirements. In one possible embodiment, if the difference between the number of consistent users and the number of inconsistent users is greater than 15, it indicates that the offline model and the online model of the risk change feature type have a high degree of consistency in calculation. Therefore, the risk change feature type is determined to be the updated change feature type.

[0094] It is understandable that when the number of updated change feature types in the risk change type is too large, in one possible embodiment, that is, when the proportion of the number of updated change feature types in the risk change feature type is greater than 0.25, many risk change feature types no longer belong to the risk change feature type. In this case, the update processing scheme for all comparison risk users is changed, that is, the risk change feature type and the update processing scheme for comparison risk users are re-determined.

[0095] Furthermore, when the number of updated change feature types in the risk change type is not excessive, it is necessary to further determine the composition of the updated change feature types in the matching feature types of the risk user. When there are no updated change feature types in the matching feature types of the risk user, it is determined that there is no need to update the update processing scheme of the risk user.

[0096] Additionally, it can be understood that when there are updated or changed feature types in the matching feature types of the risk-following users, if the proportion of the updated or changed feature types in the matching feature types meets the requirements, i.e., greater than 0.7 in one possible embodiment, then the offline system for the risk-following users will be updated immediately to ensure the consistency between the offline system and the People's Bank of China's online system.

[0097] If the proportion of updated and changed feature types in the matching feature types does not meet the requirements, it is necessary to further determine the number of updated features for the comparison risk users. For comparison risk users whose number of updated features exceeds the preset feature number threshold (i.e., more than 4 updated features), the inconsistency between their online and offline systems is relatively high. Therefore, the update processing scheme for the comparison risk users needs to be modified to maintain the consistency between the offline system and the online system of the People's Bank of China. For comparison risk users whose number of updated features does not exceed the preset feature number threshold, there is no need to modify the update processing scheme for comparison risk users.

[0098] Optionally, the method for determining the change results of the update processing scheme for comparing risky users is as follows:

[0099] Based on the updated data of the comparison results of offline and online models in different risk change characteristic types, the number of users whose comparison results of offline and online models are consistent in different risk change characteristic types is determined and used as the number of consistent users. Based on the number of consistent users and the number of inconsistent users, the updated change characteristic type in different risk change characteristic types is determined. The updated affected users are determined based on the composition of the updated change characteristic type of users with different comparison risks.

[0100] Users with comparison risks of updated feature types are identified as potential update targets. Based on the updated data in the People's Bank of China's online system for these potential update targets, the number of features missing in the People's Bank of China's offline system for these potential update targets is determined and used as the number of updated features.

[0101] Based on the constituent data of the update change feature types and the number of update features of different potential update objects, the change result of the update processing scheme of the potential update objects is determined.

[0102] Specifically, if the number of updated change feature types in the risk change type is too large, in one possible embodiment, that is, when the proportion of the number of updated change feature types in the risk change feature type is greater than 0.25, then many risk change feature types no longer belong to the risk change feature type. In this case, the update processing scheme for all comparison risk users is changed, that is, the risk change feature type and the update processing scheme for comparison risk users are re-determined.

[0103] Additionally, it should be noted that if the number of updated change feature types in the risk change types is too large, it is necessary to determine the users affected by the update based on the composition of the updated change feature types. Specifically, if the proportion of the updated change feature type in the matching feature types meets the requirements, the users affected by the update are considered as the users affected by the update. In one possible embodiment, this proportion is greater than 0.7. In this case, the offline system of the users affected by the update is immediately updated to ensure the consistency between the offline system and the People's Bank of China's online system. When the number of users affected by the update is large, that is, greater than the preset threshold for the number of affected users, since all users affected by the update need to be updated, it is necessary to re-determine the update processing scheme for the risk change feature types and the users affected by the update. In other cases, proceed to the next step.

[0104] It is understandable that when the number of potential update objects exceeds the preset threshold for the number of potential update objects, it is necessary to re-determine the risk change characteristic type and the update processing scheme for the risk users.

[0105] It should be noted that when the number of potential update objects is not greater than the preset threshold for the number of potential update objects, if the number of update features of the potential update objects does not meet the requirements in the above steps, that is, if the number of update features is more than 4, then it is directly determined that the offline system of the potential update objects needs to be updated. The degree of inconsistency between the People's Bank of China's online system and the offline system is relatively high. Therefore, on this basis, the update processing scheme for the comparison risk users needs to be modified, that is, update processing needs to be performed to maintain the consistency between the offline system and the People's Bank of China's online system.

[0106] Furthermore, for users at risk of not having updated or changed feature types, it is directly determined that there is no need to perform the update processing scheme for the users at risk of change.

[0107] Optionally, the update requirement value of the potential update object can be determined by the average of the ratio of the proportion of different potential update object update change feature types in the matching feature types and the ratio of the number of updated features to the number of pedestrian features in the original offline system of the potential update object, or by the sum of the ratio of the proportion of different potential update object update change feature types in the matching feature types and the ratio of the number of updated features to the number of pedestrian features in the original offline system of the potential update object. When the update requirement value is greater than the preset update requirement threshold, the update processing scheme for the comparison risk user needs to be modified, that is, an update processing is performed to maintain the consistency between the offline system and the pedestrian online system.

[0108] Furthermore, when the number of potentially affected users requiring immediate update processing is determined, and when the number of potentially affected users requiring immediate update processing is large, i.e. greater than the preset update user number threshold, it is necessary to re-determine the risk change characteristic type and the update processing plan for the risk users.

[0109] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0110] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0111] The above description is merely one or more embodiments of this specification and is not intended to limit this specification. Various modifications and variations can be made to the one or more embodiments of this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of one or more embodiments of this specification should be included within the scope of the claims of this specification.

Claims

1. A method for resolving the inconsistency between online and offline pedestrian characteristics, characterized in that, Specifically, it includes: Based on pedestrian characteristics, an integrated feature calculation model is constructed. Based on the distribution data of pedestrian characteristics of users in the offline system, reference users for feature changes are determined. Based on the distribution of reference users for feature changes under feature change types, comparison risk users and risk change feature types in the user group are determined. Based on the comparison results of the risk change feature types matched with the risk users and the offline and online models of the feature change reference users, the risk users who are updated only in specific risk change feature types are identified and used as the updated matching users. Based on the updated matching users corresponding to the risk change feature types and the risk change feature types matched by the compared risk users, the update processing scheme for the compared risk users is determined. Based on the update data of the comparison results of offline and online models in the risk change feature types, the change result of the update processing scheme for the compared risk users is determined. The method for determining the updated matching user is as follows: Based on the risk change feature types matched by the risk comparison users, determine the number of risk change feature types matched by the risk comparison users; Based on the comparison results of offline and online models of feature change reference users with different risk change characteristic types, the number of feature change reference users with inconsistent comparison results is determined and they are regarded as the number of inconsistent users. Based on the number of inconsistent users with different risk change characteristics, determine whether the compared risk users are update matching users.

2. The method for resolving the inconsistency between online and offline pedestrian characteristics as described in claim 1, characterized in that, The integrated feature calculation model is a calculation model obtained by jointly constructing offline and online models of pedestrian features.

3. The method for solving the problem of inconsistency between online and offline pedestrian features as described in claim 1, characterized in that... The user's feature change reference user is a user in the offline system who has the same human characteristics as the user and also has human characteristics that the user does not have.

4. The method for resolving the inconsistency between online and offline pedestrian characteristics as described in claim 1, characterized in that, The method for determining risky users through comparison is as follows: Based on the reference users for feature changes, determine the reference users for feature changes of the user in the absence of human characteristics; Based on the number of reference users whose characteristics change under different missing human characteristics, determine whether the user is a comparison risk user.

5. The method for solving the problem of inconsistency between pedestrian characteristics online and offline as described in claim 4, characterized in that... The feature variation type is the missing human body feature.

6. The method for solving the problem of inconsistency between online and offline pedestrian features as described in claim 4, characterized in that... If a user has multiple missing human characteristics, such as a lack of reference users whose number of characteristic variations does not meet the requirements, then the user is identified as a comparison risk user.

7. The method for resolving the inconsistency between online and offline pedestrian characteristics as described in claim 1, characterized in that, The method for determining the update processing scheme for comparing risky users is as follows: Based on the risk change characteristic types matched with the risk users mentioned above, determine the number of matching characteristic types; Based on the updated matching users corresponding to different matching feature types, determine the number of updated matching users for each matching feature type. Based on the number of updated matching users and the number of matching feature types, an update processing scheme for the comparison of risky users is determined.

8. The method for resolving the inconsistency between online and offline pedestrian characteristics as described in claim 7, characterized in that, The updated matching user corresponding to the matching feature type is a comparison risk user whose specific risk change feature type is the matching feature type.

9. The method for resolving the inconsistency between online and offline pedestrian features as described in claim 7, characterized in that, Based on the number of updated matching users and the number of matching feature types, an update processing plan for the comparison risk users is determined, specifically including: Based on the number of matching feature types, determine the number of matching feature types corresponding to the update processing of the compared risk users; Based on the number of matching feature types corresponding to the update processing of the risk users, and according to the number of updated matching users corresponding to different matching feature types from small to large, the update processing scheme for the risk users is determined.

Images