A method and related equipment for completing compliance risk field information
By integrating multi-source data and using an OCR engine for recognition and processing, the problem of low efficiency and timeliness in the quality management of customer file data in financial institutions has been solved. It has achieved efficient completion and automatic repair of compliance risk fields, meeting anti-money laundering regulatory requirements.
Patent Information
- Application Number
- CN202611202483.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-10
- Publication Date
- 2026-11-17
AI Technical Summary
In existing technologies, financial institutions face problems such as low efficiency, delayed timeliness, and low completion and repair rates in customer file data quality management, which cannot meet the timely identification and reporting requirements of anti-money laundering supervision.
By integrating multi-source data, missing compliance risk fields are identified, the field types are determined, and the OCR engine is invoked for recognition processing when the threshold for the number of OCR interface calls is met, thus achieving automatic completion.
It improved the efficiency of integrity checks and the rate of automatic completion and repair of compliance risk fields, met the regulatory requirements for timely identification and reporting, and enhanced the timeliness and accuracy of data quality management.
Smart Images

Figure CN123115108A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of data processing technology, and in particular to a method and related equipment for completing compliance risk field information. Background Technology
[0002] Anti-money laundering is a legal obligation for financial institutions. According to regulatory requirements, financial institutions are required to collect and maintain multiple core identity information entries in customer files, such as name, type of identification document, document number, document validity period, gender, contact information, address, occupation, and nationality. The completeness and accuracy of these fields constitute the basic dimensions of customer file data quality management.
[0003] In related technologies, the quality management of customer file data suffers from problems such as low efficiency, delayed timeliness, and low completion and repair rate. Summary of the Invention
[0004] To address the problems of existing technologies, this application provides a method and related equipment for completing compliance risk field information. The technical solution is as follows: On the one hand, a method for completing compliance risk field information is provided, the method including: In response to a field completion task for at least one missing compliance risk field in a target object file record, multi-source data fusion processing is performed on each missing compliance risk field based on data from multiple data sources to obtain a multi-source data fusion result; If, based on the results of the multi-source data fusion, it is determined that there are still fields to be completed, the field type of the fields to be completed is determined; the at least one missing compliance risk field includes the fields to be completed. When the field type is a target field type, a target OCR interface call count threshold that matches the priority of the target field type is determined based on the OCR interface call configuration information; the target field type is a key field or a text field; the OCR interface call configuration information indicates the OCR interface call count threshold corresponding to different priorities. If the number of OCR interface calls within the current time window is less than the target OCR interface call threshold, then the preset OCR interface is called, and the OCR engine associated with the preset OCR interface is used to recognize and process the target object document image corresponding to the target object file record. Based on the results of the identification process, the fields to be completed are supplemented with information in the target object file record.
[0005] On the other hand, a compliance risk field information completion device is provided, the device comprising: The multi-source data fusion module is used to respond to a field completion task for at least one missing compliance risk field in the target object file record, and to perform multi-source data fusion processing on each missing compliance risk field based on data from multiple data sources to obtain a multi-source data fusion result; The field type determination module is used to determine the field type of the field to be completed when it is determined, based on the multi-source data fusion result, that there are still fields to be completed; the at least one missing compliance risk field includes the field to be completed. The OCR quota determination module is used to determine, based on the OCR interface call configuration information, a target OCR interface call count threshold that matches the priority of the target field type when the field type is a target field type; the target field type is a key field or a text field; the OCR interface call configuration information indicates the OCR interface call count threshold corresponding to different priorities. The OCR interface call module is used to call a preset OCR interface if the number of OCR interface calls within the current time window is less than the target OCR interface call threshold, and then use the OCR engine associated with the preset OCR interface to recognize and process the target object document image corresponding to the target object file record. The information completion module is used to complete the information of the field to be completed in the target object file record based on the result of the recognition process.
[0006] On the other hand, an electronic device is provided, including a processor and a memory, wherein the memory stores at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor to implement the compliance risk field information completion method of any of the above aspects.
[0007] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction or at least one program is stored therein, the at least one instruction or the at least one program being loaded and executed by a processor to implement the compliance risk field information completion method as described above.
[0008] On the other hand, a computer program product is provided, which includes a computer program that, when executed by a processor, implements the compliance risk field information completion method of any of the above aspects.
[0009] This application embodiment, in response to a field completion task for at least one missing compliance risk field in a target object file record, performs multi-source data fusion processing on each missing compliance risk field based on data from multiple data sources to obtain a multi-source data fusion result. If, based on the multi-source data fusion result, it is determined that a field still needs to be completed, the field type of the field to be completed is determined. If the field type is a target field type, a target OCR interface call count threshold matching the priority of the target field type is determined based on OCR interface call configuration information. If the number of OCR interface calls within the current time window does not exceed the target OCR interface call count threshold, a preset OCR interface is called. The OCR engine associated with the preset OCR interface performs recognition processing on the target object document image corresponding to the target object file record. Based on the recognition processing result, information is completed for the field to be completed in the target object file record. This achieves efficient compliance risk field integrity checking and intelligent automatic completion and repair, improving the automatic completion and repair rate and timeliness, and meeting the regulatory requirement of "timely identification and timely reporting." Attached Figure Description
[0010] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0011] Figure 1 This is a flowchart illustrating a method for completing compliance risk field information provided in an embodiment of this application; Figure 2 This is a flowchart illustrating another method for completing compliance risk field information provided in this application embodiment; Figure 3 This is a flowchart illustrating another method for completing compliance risk field information provided in this application embodiment; Figure 4 This is a flowchart illustrating another method for completing compliance risk field information provided in this application embodiment; Figure 5 This is a schematic diagram of the system architecture provided in this application embodiment for implementing the method for completing compliance risk field information; Figure 6 This is a specific example diagram of the compliance risk field information completion method provided in the embodiments of this application; Figure 7 This is a specific example diagram of the OCR intelligent routing decision provided in the embodiments of this application; Figure 8 This is a feedback closed-loop flowchart based on manual review provided in the embodiments of this application; Figure 9 This is a structural block diagram of a compliance risk field information completion device provided in an embodiment of this application; Figure 10 This is a hardware structure block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation
[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.
[0013] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.
[0014] In the embodiments of this application, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.
[0015] It is understood that in the specific embodiments of this application, data such as user information are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.
[0016] In some implementation methods, the quality management of customer profile data can include the following methods: (1) Manual sampling method: Business personnel randomly select a certain percentage (usually 1%-5%) of customer profiles and visually check whether the ID card information submitted by the customer is consistent with the information entered in the system. (2) Rule engine verification method: Use rule engines such as Drools to configure format verification rules, such as verifying the mobile phone number format (^1[3-9]\d{9}$) and verifying "the ID card number cannot be empty" and "the mobile phone number must be 11 digits". (3) Batch export analysis method: Use ETL tools to export customer data from the system to HDFS distributed storage or Excel files on a regular basis (e.g., monthly) for offline analysis by compliance personnel. (4) Single data source inspection method: Only check the field integrity from the customer's main table, ignoring the data differences that may exist for the same customer in different business channels (e.g., online APP, offline outlets, third-party cooperation platforms). (5) Lack of intelligent repair capability and reliance on manual data repair: For fields that have been found to be missing, the repair methods only include notifying the customer to supplement or manual repair. This results in the following technical defects: (1) Low inspection efficiency, strong subjectivity, and inability to implement management: For financial institutions with tens of millions of customers, a sampling rate of 1% to 5% means that at least 95% of customers are not inspected. Taking 10 million customers as an example, even if we calculate based on the high workload of each person inspecting 100 files per day, a full inspection would require about 200 years of manual work, which is completely impractical in actual operation. Moreover, different business personnel may have different judgment standards for the same file, resulting in a lack of consistency and traceability in the inspection results. Manual sampling is usually carried out quarterly or annually, which cannot meet the "timeliness" principle required by regulators, and the risk exposure window is long. (2) Can only detect syntactic errors: The rule engine can only verify whether the data format conforms to the preset pattern, and cannot identify data missing or logical contradictions at the semantic level. For example, if a field value is "unknown" or "not available", although it meets the string format requirements, it is actually invalid data. And the inability to resolve cross-channel data conflicts: When the information filled in by the same customer in different channels is inconsistent (e.g., the occupation is "engineer" on the APP and "freelancer" on another channel), the rule engine lacks a credibility assessment mechanism and cannot determine which value is more accurate. It usually directly refuses to submit or requires manual intervention. (3) Delay in timeliness: The ETL extraction, transformation and loading process for processing 10 million customer data takes 6-8 hours. With the addition of manual analysis time, the complete cycle from data export to problem discovery is usually 2-4 weeks. This means that when data quality problems are discovered, the relevant customer may have already completed several high-risk transactions, which cannot meet the regulatory requirement of "timely identification and timely reporting". And the low efficiency of manual processing: Compliance personnel need to manually compare data in Excel through functions, pivot tables and other operations. Each person can only process a few hundred records per day at most. (4) The data conflict resolution strategy is simple and crude: When multiple values are detected in the same field, the existing system usually adopts the "last updated first" strategy, that is, directly using the value with the most recent update time to cover other values. This strategy ignores the authority and historical accuracy of the data source, which may lead to high-quality data being covered by low-quality data. For example: data manually verified by tellers at offline branches may be overwritten by data filled in by users on the online APP. (5) Customer supplementation or manual data correction is not only lengthy and has low customer cooperation, but also has a low success rate. It also completely ignores the possibility of automatic correction using existing resources (such as the ID card image information retained when opening an account).
[0017] In view of this, this application provides a method for completing compliance risk field information. This method, in response to a field completion task for at least one missing compliance risk field in a target object's file record, performs multi-source data fusion processing on each missing compliance risk field based on data from multiple data sources to obtain a multi-source data fusion result. If, based on the multi-source data fusion result, it is determined that there are still fields to be completed, the field type of the field to be completed is determined. If the field type is a target field type, a target OCR interface call count threshold matching the priority of the target field type is determined based on OCR interface call configuration information. If the number of OCR interface calls within the current time window does not exceed the target OCR interface call count threshold, a preset OCR interface is called. The OCR engine associated with the preset OCR interface performs recognition processing on the target object's document image corresponding to the target object's file record. Based on the recognition processing result, the information to be completed for the field to be completed is performed in the target object's file record. This achieves efficient compliance risk field integrity checking and intelligent automatic completion and repair, improving the automatic completion and repair rate and timeliness, and meeting the regulatory requirement of "timely identification and timely reporting."
[0018] Please see Figure 1 The diagram illustrates a flowchart of a method for completing compliance risk field information according to an embodiment of this application. This method can be applied to a server, which can be a server in a compliance risk data quality management system used to provide compliance risk field information completion services. In a specific application scenario, the compliance risk data quality management system can be an anti-money laundering data quality management system for financial institutions.
[0019] It should be noted that the compliance risk field in this application embodiment refers to preset data attributes used to identify, assess, or monitor the compliance risks of business objects, including but not limited to fields such as identity identifier, document type, document number, document validity period, gender, contact information (personal mobile phone number), address (registered address), occupation, and nationality. In this application embodiment, the compliance risk field being an anti-money laundering field will be used as an example for specific illustrative explanation.
[0020] It should be noted that this specification provides the operational steps of the methods described in the embodiments or flowcharts, but based on conventional or non-inventive labor, more or fewer operational steps may be included. The order of steps listed in the embodiments is merely one possible execution order among many and does not represent the only execution order. In actual system or product execution, the methods shown in the embodiments or drawings can be executed sequentially or in parallel (e.g., in a parallel processor or multi-threaded processing environment). Specifically, as shown... Figure 1 As shown, the method may include: S101, in response to a field completion task for at least one missing compliance risk field in a target object file record, performs multi-source data fusion processing on each of the missing compliance risk fields based on data from multiple data sources to obtain a multi-source data fusion result.
[0021] The target object file record is one of the current record sets of objects. The target object file record includes the archived master data of the target object. The target object can be a business object served in financial business, or it can be referred to as a customer.
[0022] In some implementations, such as Figure 2 As shown, before step S101, the method may further include: S201, in response to the scheduled task trigger instruction, obtain the set of object file records to be processed based on the scheduled task type.
[0023] In practical implementation, the XXL-JOB distributed task scheduling platform can be integrated. This platform can periodically trigger scanning tasks based on time periods and business channels, retrieving a set of object archive records to be processed according to the scheduled task type. The scheduled task types can include full scan, incremental scan, and targeted scan, which can be configured according to actual needs. Unlike traditional single-machine scheduled tasks, the XXL-JOB distributed task scheduling platform provides cluster scheduling, fragmented broadcasting, and failure retries, thus supporting efficient scanning of tens of millions of customer data points.
[0024] For the full scan mode, which can be triggered at 2:00 AM on Sunday, a sharded broadcast mechanism is used to distribute all customer IDs by hash modulo to multiple executor nodes for parallel processing. For example, deploying 10 executor nodes, each responsible for scanning approximately 1 million customer data points, can reduce the overall processing time from over 6 hours on a single machine to less than 40 minutes.
[0025] For incremental scanning mode, for example, it can be triggered at 4:00 AM daily to scan only customer records that changed the previous day. The system obtains a list of newly added and modified customer IDs from the database table's update time, which serves as the input range for the incremental scan. This mode keeps the daily scan data volume below tens of thousands, achieving near real-time data quality monitoring.
[0026] For targeted scanning mode, compliance personnel can manually trigger it, creating temporary scan tasks based on business needs, such as channel source, customer type, and time range. For example, when regulators require a targeted investigation of the data quality of "newly opened accounts through third-party cooperative channels within the past 30 days," targeted scanning can be started immediately through configuration parameters.
[0027] S203, based on the preset compliance risk field rule information, perform field integrity detection on each object file record in the object file record set to obtain the field integrity detection result; the field integrity detection result indicates that at least one target object file record has missing compliance risk fields.
[0028] S205, Based on the field integrity detection results, generate a field completion task corresponding to each target object file record.
[0029] The preset compliance risk field rules information is used to define multiple compliance risk fields and their verification rules for compliance risk regulatory requirements. For example, it is used to define nine core identity information fields and their verification rules for anti-money laundering regulatory requirements, specifically including identity identifier, document type, document number, document validity period, gender, contact information (personal mobile phone number), address (registered address), occupation, and nationality. The preset compliance risk field rules information in this embodiment can be dynamically configured and adjusted by business personnel through the management interface, taking effect without modifying the code.
[0030] In specific implementation, the pre-defined compliance risk field rules information configures basic field attributes and a set of validation rules for each field. The basic field attributes for each field are defined as follows: Field Name: such as "Name", "ID Number", "ID Validity Period", etc.; Data Type: string, date, enumeration value, etc.; Required: marked as mandatory or conditionally required according to regulatory requirements; Weighting Coefficient: used to calculate the overall completeness score, with key fields (such as ID Number) having a higher weight (e.g., set to 15), and secondary fields (such as ethnicity) having a relatively lower weight (e.g., set to 5).
[0031] Each field's validation rule set can include multiple layers of validation rules, specifically: non-empty validation rules, format validation rules, logical validation rules, dictionary matching rules, and checksum verification rules. Specifically, non-empty validation rules check whether the field is an empty value, an empty string, or a placeholder (such as "unknown" or "none"); format validation rules validate the data format using regular expressions, such as requiring mobile phone numbers to conform to the pattern ^1[3-9]\d{9}$; logical validation rules validate the rationality of field values, such as requiring the validity period of a certificate to be later than the issuance date; dictionary matching rules compare field values with a standard dictionary, such as requiring occupations to conform to the relevant occupational classification standard code; and checksum verification rules validate the correctness of the 18th checksum for ID card numbers according to regulatory standards.
[0032] In this embodiment, the preset compliance risk field rule information may include field derivation relationship definitions, which are used to define the derivation relationships between fields to avoid unnecessary subsequent OCR (Optical Character Recognition) interface calls. For example, from the ID card number, the following can be derived: gender (odd 17th digit for male, even 17th digit for female), date of birth (7th-14th digits), and household registration area code (1st-6th digits). From the document type, the following can be derived: nationality.
[0033] When performing field integrity checks on each object file record in the object file record set based on preset compliance risk field rules, each object file record can first undergo preprocessing. This preprocessing includes field standardization and cleaning, such as standardizing null value fields and uniformly marking NULL, empty strings, "-", "unknown", etc., as "missing status". At the same time, field values containing leading or trailing spaces or special characters are cleaned.
[0034] For each preprocessed object file record, a field-by-field validation is performed according to the set of validation rules for each field in the preset compliance risk field rule information. Specifically, for each field, non-empty validation (based on the corresponding non-empty validation rule), format validation (based on the corresponding format validation rule), and logical validation (based on the corresponding logical validation rule) are performed in sequence. If any validation rule fails, the field is determined to be a "missing compliance risk field" and included in the scope to be supplemented.
[0035] In some optional embodiments, missing compliance risk fields can be further classified based on the corresponding validation results: when the validation result indicates that the field value is empty or a placeholder, it is determined to be a completely missing field, which may need to be supplemented from an external data source; when the validation result indicates that the field has a value but does not conform to the validation rules, it is determined to be a format error field, which needs to be corrected or re-collected; when the validation result indicates that the field value passes the format validation but has a logical contradiction with other fields (such as the birth date and the ID number being inconsistent), it is determined to be a suspected error field, which needs to be further verified.
[0036] In this embodiment of the application, for each target object file record with missing compliance risk fields, a field completion task is generated for each target object file record based on the missing compliance risk fields corresponding to each target object file record. The field completion task is used to instruct the completion of field information for at least one missing compliance risk field in the corresponding target object file record.
[0037] The above implementation method realizes automated scheduling and differentiated execution of tasks. It can quickly and accurately identify target object file records with missing compliance risk fields, and automatically generate corresponding field completion tasks for each target object file record with missing fields. It can flexibly adapt to different business scenarios and improve the coverage, consistency and efficiency of the inspection.
[0038] In some implementations, step S101, when performing multi-source data fusion processing on each of the missing compliance risk fields based on data from multiple data sources to obtain the multi-source data fusion result, may include: For each of the missing compliance risk fields, candidate values corresponding to the missing compliance risk fields are searched from the data of the multiple data sources respectively; For a missing compliance risk field for which multiple candidate values are found, a comprehensive confidence score is determined for each candidate value based on the time decay factor of each candidate value and the historical accuracy of the data source to which the candidate value belongs on the missing compliance risk field; based on the candidate value with the highest comprehensive confidence score, it is determined whether the missing compliance risk field is a field to be completed. For a missing compliance risk field for which no candidate value was found, the missing field is determined to be a field to be completed; wherein, the multi-source data fusion result indicates the field to be completed among the at least one missing compliance risk field.
[0039] Specifically, multiple data sources can be set based on actual application scenarios, including third-party cooperation channels (such as customer data from cooperation channels), online channels (such as registration and modification of data via mobile terminal APP or online banking), and offline outlets (such as data entered or verified by tellers).
[0040] In this embodiment of the application, for a missing compliance risk field in which multiple candidate values are found in multiple data sources, for each candidate value, the time decay factor of the candidate value and the historical accuracy of the data source to which the candidate value belongs on the missing compliance field are calculated, and then the time decay factor and the historical accuracy are multiplied to obtain the comprehensive confidence score corresponding to the candidate value.
[0041] The time decay factor characterizes the freshness of data, and the freshness of data directly affects its reliability. Newer candidate values should receive a larger time decay factor. For example, the exponential decay function can be used to calculate the time decay factor, as shown in the following formula (1): (1) in, This represents the time decay factor, which is dimensionless and ranges from (0,1]. This represents the time decay rate, a hyperparameter that can be set by the business logic. This represents the time difference, which is the time difference between the current moment and the last time the candidate value was updated.
[0042] The time decay factor is calculated using the above formula. , The larger the value, the fresher and more trustworthy the data (i.e., candidate values); conversely, the smaller the value, the more trustworthy it is. The smaller the value, the older the data (i.e., candidate values); however, due to the formula above... In mathematics, zero is never possible. Therefore, old data does not become completely invalid, but its reliability gradually decreases.
[0043] Historical accuracy is used to characterize the accuracy performance of each business channel in past data collection processes. In this embodiment, a data statistics table can be maintained to record the historical accuracy of each data source on each compliance risk field. For example, the calculation of historical accuracy may be based on at least one of the following methods or any combination thereof:
[0044] Comparison with subsequent authoritative supporting documents provided by the customer: When the customer subsequently provides authoritative identity verification documents (such as the original ID card) through counter, remote video, or other channels, the authentic field information contained in the document is compared item by item with the corresponding field values in the historical file. If the two match, it is recorded as a correct data collection; if they do not match, it is recorded as an incorrect data collection.
[0045] Cross-validation with authoritative third-party data sources: By connecting to an external authoritative data source interface (such as an ID card information verification interface from an authoritative department), the stored customer profile field values can be cross-compared with the standard values returned by the authoritative third-party data source. If the field values match those of the authoritative data source, they are considered correct; otherwise, they are considered incorrect.
[0046] Confidence statistics of OCR recognition results: Data sources with stable confidence distribution and high values have correspondingly higher historical accuracy.
[0047] Manual sampling verification results: Quality inspectors or compliance auditors conduct sampling verification of the collected customer file data. Based on the accuracy rate of the sampling verification, combined with the sampling ratio and confidence interval, the historical accuracy rate of the data source on each field is estimated.
[0048] By combining one or more of the above methods, the historical accuracy data of each data source in each compliance risk field in the data statistics table can be continuously updated.
[0049] In this embodiment of the application, for multiple candidate values of the same missing compliance risk field, after calculating the comprehensive confidence score of each candidate value, the candidate value corresponding to the highest comprehensive confidence score is used to determine whether the missing compliance risk field is a field to be completed.
[0050] The above implementation method makes full use of data from multiple data sources, avoiding the waste of manpower caused by directly triggering the completion task due to the lack of a single data source. At the same time, by conducting multi-dimensional quantitative evaluation of candidate values, it avoids misjudgment caused by a single dimension, and solves the data conflict problem through comprehensive confidence, significantly improving the accuracy and stability of candidate value selection and avoiding unnecessary completion operations.
[0051] In some implementations, determining whether the missing compliance risk field is a field to be completed is based on the candidate value with the highest overall confidence score, including: If the highest overall confidence score is less than a preset score threshold, the corresponding missing compliance risk field is determined to be a field to be completed. If the highest overall confidence score is greater than or equal to the preset score threshold, and the score difference between the highest overall confidence score and the second highest overall confidence score is less than the score difference threshold, then the target candidate value corresponding to the missing compliance risk field is determined based on the authority of the data source corresponding to the candidate value of the highest overall confidence score and the authority of the data source corresponding to the candidate value of the second highest overall confidence score. A cross-field logical consistency check is performed on each of the missing compliance risk fields with target candidate values, and the missing compliance risk fields that indicate logical contradictions in the check results are identified as fields to be completed.
[0052] Specifically, the preset score threshold can be set based on practical experience, for example, 0.5 (equivalent to 50% after normalization). The highest overall confidence score is compared with this preset score threshold. If the highest overall confidence score is lower than the preset score threshold, it indicates that the overall confidence scores corresponding to all candidate values are less than the preset score threshold. In this case, it can be determined that all candidate values are unreliable, and the missing compliance risk field is identified as a field to be completed, thus triggering subsequent corresponding completion processing. Conversely, if the highest overall confidence score is greater than or equal to the preset score threshold, it is further determined that the highest overall confidence score is "close to confidence". At this time, the authority of the data source can be compared, that is, based on the authority of the data source corresponding to the candidate value with the highest overall confidence score and the authority of the data source corresponding to the candidate value with the second highest overall confidence score, the target candidate value corresponding to the missing compliance risk field is determined.
[0053] Specifically, if there is a significant difference in authority between the data source corresponding to the candidate value with the highest overall confidence score and the data source corresponding to the candidate value with the second highest overall confidence score, then the candidate value from the data source with the higher authority is prioritized as the target candidate value for the missing compliance risk field. If the authority of the two data sources is relatively close, the field type of the missing compliance risk field can be considered to determine whether it is a field to be completed. If the field type is a non-critical field (such as ethnicity), the most recently updated candidate value can be selected as the target candidate value for the missing compliance risk field; if the field type is a critical field, then the missing compliance risk field is determined to be a field to be completed.
[0054] In this embodiment, for each missing compliance risk field, if the optimal target candidate value is selected, a cross-field logical consistency check is further performed. Specifically, this check may include: whether the gender matches the 17th digit of the ID card number; whether the date of birth matches the 7th to 14th digits of the ID card number; and whether the validity period of the document is reasonable (e.g., the validity period of the ID card cannot exceed 20 years). If a logical contradiction is found, the missing compliance risk field with the logical contradiction is identified as a field to be completed; otherwise, if there is no logical contradiction, the corresponding target candidate value is used as the field value of the corresponding missing compliance risk field.
[0055] The above implementation method ensures that fields that still fail to meet the minimum credibility standard after multi-source data fusion can be identified in a timely manner and included in the completion process, thereby guaranteeing the overall credibility of anti-money laundering customer identification data. By introducing the authority of the data source, it avoids the situation where low-authority source data is mistakenly selected due to slight differences in confidence scores, improving the reliability and business compliance of multi-source data fusion results. At the same time, it also reflects a multi-level, progressive decision-making logic in handling data conflicts, improving the decision-making accuracy of the anti-money laundering data quality management system, and thus improving the accuracy of field information completion.
[0056] S103, if it is determined from the multi-source data fusion result that there are still fields to be completed, determine the field type of the fields to be completed; the at least one missing compliance risk field includes the fields to be completed.
[0057] In this embodiment, multiple compliance risk fields are categorized into key fields, derivable fields, text fields, and default value fields based on their identification characteristics. Specifically, key fields require extremely high accuracy, as errors can lead to serious compliance risks; these fields may include identity identifiers (such as name), ID number, and ID validity period. Derivable fields can be directly extracted from ID numbers without OCR recognition; these fields may include gender, date of birth, and place of residence. Text fields are free text, which is more difficult to recognize and is easily affected by layout and font; these fields may include address, occupation, and ethnicity. Default value fields may include nationality.
[0058] In this embodiment of the application, if it is determined that there are still fields to be completed based on the results of the multi-source data fusion, the field type of the field to be completed is determined, that is, which field type it is among key field, derivable field, text field and default value field.
[0059] If the field type of the field to be completed is a key field or a text field, then it is the target field type, and steps S105 to S109 can be executed. Otherwise, if it is not a key field or a text field, but a deducible field or a default value field, then it is a non-target field type, and steps S111 can be executed for non-target field types.
[0060] S105, when the field type is a target field type, determine the target OCR interface call count threshold that matches the priority of the target field type based on the OCR interface call configuration information; the target field type is a key field or a text field; the OCR interface call configuration information indicates the OCR interface call count threshold corresponding to different priorities.
[0061] In this embodiment, priorities are set for different types of field completion tasks, and differentiated OCR interface call thresholds are set for different priorities. Specifically, key fields are set to high priority, text fields to medium priority, and deducible or default value fields to low priority. High priority is associated with a high OCR interface call threshold, medium priority with a medium OCR interface call threshold, and low priority with a low (e.g., set to 0 to skip OCR interface calls) OCR interface call threshold. In other words, the OCR interface call configuration information in this embodiment can include OCR interface call thresholds corresponding to different priorities, with the thresholds decreasing sequentially from high to low priority.
[0062] In practice, a maximum OCR interface call threshold can be preset. This threshold is configured based on the API (interface) call frequency limit agreed upon with the OCR service provider. The OCR interface call thresholds for different priorities can then be multiplied by a coefficient less than 1 from this maximum threshold, with higher priorities having larger coefficients. For example, the OCR interface call threshold associated with high priority could be the maximum threshold multiplied by a coefficient of 0.8; while the OCR interface call threshold associated with medium priority could be the maximum threshold multiplied by a coefficient of 0.5.
[0063] If the field to be completed is a key field, it can be determined as a high priority field. Then, the OCR interface call count threshold corresponding to the high priority field can be obtained from the OCR interface call configuration information as the target OCR interface call count threshold field. If it is a text field, it can be determined as a medium priority field. Then, the OCR interface call count threshold corresponding to the medium priority field can be obtained from the OCR interface call configuration information as the target OCR interface call count threshold field.
[0064] S107. If the number of OCR interface calls within the current time window is less than the target OCR interface call threshold, then the preset OCR interface is called, and the OCR engine associated with the preset OCR interface is used to recognize and process the target object document image corresponding to the target object file record.
[0065] The target object's ID card image includes fields to be completed. The default OCR interface uses a RESTful API interface, and the parameters carried in the call request may include: (1) Image data: Base64 encoded JPEG or PNG format image; (2) Recognition type: such as "ID card recognition" or "general text recognition"; (3) Returned fields: specify the fields to be extracted.
[0066] In this embodiment, an OCR interface call counter can be maintained to record the number of OCR interface calls initiated within the current time window using a sliding window method. The preset time length corresponding to the sliding window can be set based on actual needs, such as 1 hour. The OCR interface call counter can obtain the number of OCR interface calls within the current time window and compare this number with the target OCR interface call threshold determined in step S105. If the number of OCR interface calls within the current time window is less than the target OCR interface call threshold, then calling the preset OCR interface is allowed. In this case, a routing status indicating direct OCR engine call, such as DIRECT_OCR, can be returned, and the preset OCR interface is then called. The OCR engine associated with the preset OCR interface performs synchronous recognition processing on the target object document image corresponding to the target object file record to obtain the OCR recognition result. Conversely, if the number of OCR interface calls within the current time window is greater than or equal to the target OCR interface call threshold, it indicates that the current system OCR call load is high, and calls can be disallowed.
[0067] In this embodiment of the application, when calling the preset OCR interface, for high-priority key fields, the OCR interface corresponding to the high-precision OCR engine is selected first, and for medium-priority text fields, the OCR interface corresponding to the standard or enhanced OCR engine is selected according to the image quality.
[0068] Understandably, when the preset OCR interface is called, the OCR interface call counter increments by one within the current time window.
[0069] By using a multi-level threshold-based OCR interface call quota management mechanism, the waste and congestion of system resources (call frequency, number of concurrent connections, response latency) caused by invalid or low-value calls to the OCR engine API can be avoided, thereby improving the utilization rate of system resources.
[0070] In some implementations, such as Figure 3 As shown, prior to step S105, the method may further include: S301, based on the target object's document image, the field identifier of the field to be completed, and the current engine version of the OCR engine, a query cache key is generated using a preset hash algorithm.
[0071] S303, based on the query cache key, search for the target cache value that matches the query cache key in the OCR recognition result cache; the OCR recognition result cache is used to store the OCR recognition value of the field in the form of key-value pairs.
[0072] S305, if the target cached value is found, the field to be completed is supplemented in the target object file record based on the OCR recognition value indicated by the target cached value.
[0073] S307, if the target cache value is not found, execute step S105 to determine the target OCR interface call count threshold that matches the priority of the target field type based on the OCR interface call configuration information.
[0074] Specifically, the preset hash algorithm can be the MD5 hash algorithm. The field identifier for the field to be completed can be the field name, such as "name" or "id_number". The target document image can be an ID card image. When generating the query cache key, a perceptual hash can be calculated on the ID card image first to generate a 64-bit image hash value to resist slight image transformations (such as scaling or cropping). Then, based on the image hash value, the field name, and the current OCR engine version, the MD5 hash algorithm is used to generate the query cache key. Introducing the current OCR engine version avoids using old cached results after engine upgrades, which helps improve the accuracy of field completion.
[0075] In this embodiment, the OCR recognition result cache stores the OCR recognition values of fields in key-value pairs. The key is generated using a preset hash algorithm based on the target object's document image, the field identifier of the field to be completed, and the engine version of the OCR engine used in the corresponding recognition result. The corresponding value includes the OCR recognition value. For example, the OCR recognition result cache can use a Redis cluster. The following is an example of the OCR recognition result cache structure: Key: ocr_cache:{cache_key} Value: { "recognized_value": "Zhang San", "confidence": 0.96, "create_time": "2024-01-15 10:30:00", "hit_count": 5 } Wherein, "recognized_value" is the OCR recognition value; "confidence" is the recognition confidence level; "create_time" is the creation time; and "hit_count" indicates the number of times the OCR recognition result is hit in the cache.
[0076] In this embodiment of the application, after calculating the query cache key, a target cache value matching the query cache key is searched in the OCR recognition result cache. That is, whether the query cache key matches the OCR recognition result cache. If it matches, the matched cache value is determined as the target cache value, and the OCR recognition value (e.g., recognized_value is Zhang San) is obtained from the target cache value. This OCR recognition value is then used as the field value of the corresponding field to be completed, and information is completed in the target object file record. If it does not match, step S105 is executed.
[0077] Optionally, if a match is found, the recognition confidence score is obtained from the target cache value. It is then determined whether the recognition confidence score is greater than a preset recognition confidence score threshold (set as needed, such as 0.9). If it is greater, the OCR recognition value in the target cache value is used as the field value of the corresponding field to be completed, and the information is completed in the target object file record, thereby further improving accuracy.
[0078] In practical applications, if the OCR recognition result cache is hit and the OCR recognition value in the target cache value is returned, the INCR command can be called to increment the cache hit count (hit_count) by 1 and return the routing status indicating the OCR recognition result cache, such as USE_CACHE.
[0079] In some optional embodiments, a cache expiration policy can be set for the OCR recognition result cache, which may include: a default cache validity period, such as 30 days; a high-frequency access cache (hit count > 10), extending the cache validity period to 90 days; a low confidence cache (confidence < 0.85), caching for only 24 hours, which can wait for the OCR engine to be optimized and re-recognized; and active invalidation, clearing all OCR caches of the customer when the customer actively updates the ID card image.
[0080] In some optional embodiments, a cache preheating mechanism may also be included. For batch tasks, the cache is queried before execution to calculate the cache hit rate. If the cache hit rate is >60%, the batch tasks are executed directly. If the cache hit rate is <30%, cache preheating is started 1 hour in advance to identify frequently accessed images in advance.
[0081] The above implementation method, through a caching and reuse mechanism, can ensure accuracy while avoiding repeated recognition of the same document image, which helps to improve the efficiency of field completion.
[0082] In some implementations, such as Figure 4 As shown, the method may further include: S401, if the number of OCR interface calls within the current time window is greater than or equal to the target OCR interface call count threshold, then the call request to the preset OCR interface will be encapsulated into a priority task that matches the priority of the target field type.
[0083] S403, store the priority task into the asynchronous priority queue.
[0084] S405, in response to the OCR interface call counter falling back to a preset safety value, the priority task in the asynchronous priority queue is executed; the OCR interface call counter is used to record the number of OCR interface calls within the current time window in a sliding window manner.
[0085] Specifically, if the number of OCR interface calls within the current time window is greater than or equal to the target OCR interface call threshold, it indicates that the system load is too high. In this case, the OCR engine will no longer be called synchronously, but a degradation process will be automatically triggered. This involves encapsulating the current call request to the preset OCR interface into a priority task of the corresponding priority, and storing this priority task in an asynchronous priority queue, such as a sorted set (ZsortedSet) in Redis. The task will wait for the OCR interface call counter to fall back to a preset safety value before being dequeued and processed, i.e., the priority task in the asynchronous priority queue will be executed. The preset safety value can be set based on practical experience.
[0086] In practical applications, after executing a priority task in the asynchronous priority queue, the routing status of the asynchronous queue can be returned, such as ASYNC_QUEUE.
[0087] The above implementation transforms real-time OCR call pressure into asynchronous, peak-shaving system load, effectively avoiding instantaneous peaks in OCR interface call requests and reducing sudden pressure on downstream OCR services. While ensuring that the OCR interface is not over-called and avoiding triggering service provider rate limiting, it maximizes the timely completion of high-priority field recognition tasks, thereby improving the stability and robustness of the entire system.
[0088] In some implementations, the priority of the key field is higher than that of the text field, and the threshold for the number of OCR interface calls corresponding to the key field is greater than the threshold for the number of OCR interface calls corresponding to the text field. In this case, storing the priority task in the asynchronous priority queue may include: if the target field type is a key field, inserting the priority task at the head of the asynchronous priority queue; if the target field type is a text field, inserting the priority task in the middle of the asynchronous priority queue. It is understood that tasks for other fields can be placed at the tail of the asynchronous priority queue. Thus, priority tasks for high-priority fields at the head of the asynchronous priority queue can be processed first after OCR interface resources are released, while priority tasks for medium-priority fields in the middle are processed sequentially. With limited OCR interface call resources, differentiated responses to fields of different importance are achieved, improving the matching degree between resource allocation and business urgency.
[0089] Optionally, for tasks entering the asynchronous priority queue, a separate asynchronous consumption thread periodically checks the decline of the API call counter. When the count value falls below the safety threshold, the task is retrieved according to the queue priority and the OCR call is executed.
[0090] This application embodiment achieves refined management of OCR interface call resources through the above-mentioned hierarchical scheduling mechanism: when the system load is low, high and medium priority tasks can receive real-time responses; and when the system load is high, priority is given to ensuring the high-precision recognition needs of key fields, while non-urgent tasks are postponed, thereby achieving the optimal balance between overall completion efficiency and accuracy within the limited API call quota.
[0091] S109, Based on the result of the recognition process, information is completed for the field to be completed in the target object file record.
[0092] In this embodiment of the application, the recognition processing results include the recognition results of multiple compliance risk fields and the recognition confidence level corresponding to each compliance risk field. It can be understood that the OCR recognition results include the recognition results of fields to be completed.
[0093] In specific implementation, the results returned by the OCR engine are the results of the recognition process, or the OCR recognition results, which include: (1) Recognized text: the recognition results of each field; (2) Confidence: the recognition confidence of each field (between 0 and 1); (3) Location information: the coordinate position of the recognized text in the image, which can be used for visualization; (4) Image quality score: the evaluation of image quality by the OCR engine.
[0094] In some implementations, based on the result of the identification process, information completion is performed on the field to be completed in the target object file record, including: Post-processing verification is performed based on the identification results of multiple compliance risk fields to obtain post-processing verification results. The post-processing verification includes verification code verification, dictionary matching verification, date logic verification, and cross-field consistency verification. Based on the post-processing verification results, a verification pass coefficient is determined. Based on the verification pass coefficient, the identification confidence of each compliance risk field is corrected to obtain the corrected identification confidence of each compliance risk field. If the corrected identification confidence of each compliance risk field meets the preset confidence threshold condition, information is completed for the fields to be completed in the target object file record based on the identification processing results.
[0095] In this embodiment of the application, in order to improve the accuracy of OCR recognition, a four-layer post-processing verification mechanism is provided, including verification code verification, dictionary matching verification, date logic verification, and cross-field consistency verification.
[0096] For ID card numbers, a check digit verification is performed according to authoritative standards. The verification algorithm includes the following steps: (1) Multiply the first 17 digits of the ID card number by the corresponding weighting factors: [7, 9, 10, 5, 8, 4, 2, 1, 6, 3, 7, 9, 10, 5, 8, 4, 2]; (2) Sum the products; (3) Take the remainder modulo 11 of the sum; (4) Look up the check digit in a table based on the remainder: [1, 0, X, 9, 8, 7, 6, 5, 4, 3, 2]; (5) Compare whether the calculated check code is consistent with the 18th digit of the ID card number. If the check code is inconsistent, the recognition result is determined to be incorrect, and the following measures are taken: a, if the recognition confidence is >0.9, mark it as "awaiting manual review"; b, if the confidence is <0.9, try to call the enhanced OCR to re-recognize; c, if the recognition fails twice in a row, mark it as "recognition failure" and transfer it to manual review.
[0097] For text fields such as name, address, occupation, and ethnicity, dictionary matching verification is performed against a standard dictionary database. The name dictionary matching process is as follows: (1) Check whether the name is composed of common Chinese characters (excluding rare characters and special symbols); (2) Check whether the name length is between 2 and 10 characters; (3) Check whether the name conforms to common naming rules (such as compound surnames "Ouyang" and "Sima"); (4) Calculate the edit distance between the name and the common name database. If the difference is too large, it is marked as abnormal. The occupation dictionary matching process is as follows: (1) Compare the recognition results with the authoritative standard occupation name database; (2) Use a fuzzy matching algorithm (such as Levenshtein distance) to find the closest standard occupation code; (3) If the matching degree with the closest standard occupation code is <0.7, it is marked as "occupation classification is unclear" and manual confirmation is required. The ethnic dictionary matching process is as follows: (1) compare the identification results with the standard name list of 56 ethnic groups; (2) support common alias mapping (such as "Han" -> "Han nationality"); (3) if a match cannot be found, mark it as "ethnic identification abnormal".
[0098] For date fields such as document validity period and date of birth, date logic verification is performed. The specific date logic verification includes: (1) the issuance date must be earlier than the expiry date; (2) the relationship between the validity period of the ID card and the holder's age: 5 years for those under 16 years old, 10 years for those 16-25 years old, 20 years for those 26-45 years old, and long-term validity for those 46 years old and above; (3) if the identified validity period does not conform to the rule in (2) above, it is marked as "validity period abnormal"; (4) check whether the document has expired. If it has expired, the customer is prompted to replace the document, including date of birth verification, the date of birth cannot be later than the current date, the date of birth must be consistent with the 7th to 14th digits of the ID card number, and the age corresponding to the date of birth must be within a reasonable range (0-120 years old).
[0099] Cross-field consistency verification is used to check the logical consistency between multiple fields, including: (1) whether the gender is consistent with the 17th digit of the ID card number; (2) whether the household registration area code is consistent with the first 6 digits of the ID card number; (3) whether other fields should not have values when the name is empty (to avoid partial recognition success).
[0100] After completing the above post-processing verification, the verification pass coefficient is determined based on the post-processing verification results. The specific verification pass coefficient can be set as follows: if all verifications pass, the verification pass coefficient = 1.0, maintaining the original confidence level; if a minor verification fails (such as dictionary matching degree 0.6-0.7), the verification pass coefficient = 0.8; if a major verification fails (such as check code error), the verification pass coefficient = 0.5; if multiple verifications fail, the verification pass coefficient = 0.3, marked as "low confidence".
[0101] The recognition confidence level can be corrected using the following formula: Corrected recognition confidence level = Recognition confidence level x Verification pass coefficient.
[0102] In this embodiment, the preset confidence threshold can be set based on practical experience, for example, it can be set to be greater than or equal to 0.85. After obtaining the corrected recognition confidence of each compliance risk field, it can be determined whether the corrected recognition confidence of the field to be completed is greater than or equal to 0.85. If the result is yes, the preset confidence threshold is met, and information can be completed in the target object file record based on the OCR recognition result of the field to be completed.
[0103] In some implementations, the method may further include: For compliance risk fields whose corrected confidence level does not meet the preset confidence threshold, a manual review of the compliance risk field is triggered to obtain the manual review result corresponding to the compliance risk field. The manual review results corresponding to the compliance risk field are compared with the candidate values of the compliance risk field in each of the data sources, and the historical accuracy of each of the data sources in the compliance risk field is updated based on the comparison results.
[0104] Specifically, if the above judgment result is negative, that is, the confidence level of the corrected field to be completed is less than 0.85, then it is determined that the preset confidence threshold condition is not met. In this case, the field to be completed is identified as the field to be composited and pushed to the manual review node for manual review to obtain the manual review result corresponding to the field to be composited. The manual review result includes the correct field value corresponding to the field to be reviewed determined by the reviewer (which can also be called the final manual adoption value). Then, the manual composite result of the field to be composited is compared with the candidate value of the field to be composited in each data source, and the historical accuracy of each data source in the field to be composited is updated based on the comparison result to achieve continuous improvement of data source quality.
[0105] In practice, the final manually adopted value can be compared item by item with the original candidate values provided by each data source for that field to determine whether each data source's candidate value has been adopted or not. If the candidate value of a data source is completely consistent with the final manually adopted value, it is determined to be adopted; if the candidate value of a data source is inconsistent with the final manually adopted value and has not been referenced by the human, it is determined to be unadopted. Optionally, if the human combines or modifies multiple candidate values, it can also be determined to be partially adopted.
[0106] For the adopted data source, the following formula (2) can be used to dynamically update the historical accuracy of the data source on the corresponding field to be reviewed based on the comparison results: (2) For data sources that are not adopted, the following formula (3) can be used to dynamically update the historical accuracy of the data source on the corresponding field to be reviewed based on the comparison results: (3) in, This indicates the updated historical accuracy. This indicates the historical accuracy rate before the update; This represents the learning rate, which is set to 0.05 in this implementation example to ensure smooth updates and avoid drastic fluctuations in historical accuracy caused by a single misjudgment.
[0107] The updated historical accuracy will be written into the "Data Source Reliability Statistics Table" as the input for the "Historical Accuracy" item in subsequent multi-source data fusion processing.
[0108] Optionally, after obtaining the manual review results, the OCR cache confidence level can also be updated. Specifically, a preset number of manual review records (e.g., the last 100 times) can be counted. If the adoption rate of a certain OCR engine version + field combination is lower than a threshold (e.g., less than 80%), the recognition confidence level of that combination in the OCR recognition result cache will be lowered by a preset amount (configurable as needed, such as 0.05). This will encourage the selection of other data sources or trigger re-recognition during subsequent processing, which is beneficial to improving the accuracy of field completion.
[0109] The above implementation method triggers manual review when the confidence level is insufficient after correction, and updates the data source quality based on the manual review results. This makes the historical accuracy of the data source continuously updated and more accurate with the manual review results, thereby making the subsequent multi-source data fusion results more accurate and reliable. It realizes the dynamic evolution of data source quality and intelligent decision support, which is conducive to improving the accuracy and compliance of anti-money laundering customer identification data.
[0110] In some optional embodiments, when a valid field value not covered by the dictionary is detected during manual review, such as emerging occupational names, new ID document types, newly added place names, ethnic standard names, and regional codes, the value is automatically entered into the corresponding dictionary, such as an occupational dictionary, ID document type dictionary, place name database, ethnic database, and regional code database. The associated dictionaries, such as occupational classification mapping tables, ethnic standard name lists, and regional code mapping tables, are updated simultaneously. This enables the dynamic evolution of the dictionary database, avoids recognition errors caused by outdated dictionaries, and ensures data compliance and business adaptability.
[0111] Optionally, the OCR recognition result can be marked as "awaiting manual review" and transferred to the manual review queue in the following situations: the confidence level after correction is less than 0.85; the verification code fails; the image quality score is less than 100; two consecutive OCR recognition results are inconsistent; or the customer complains or actively applies for information changes.
[0112] Optionally, the manual review node can provide several auxiliary functions to improve review efficiency and accuracy, which may include: (1) Original image display: Highlighting the position of the recognized field on the original image of the OCR recognition so that the reviewer can locate it intuitively; (2) Recognition result comparison: Displaying multiple OCR recognition results of the same field side by side; (3) Recommended value recommendation: Recommending the optimal candidate value based on the fusion of multi-source data in a prominent position on the manual review interface; (4) Quick operation: Providing one-click adoption of recommended values and manual modification (support, simplifying the operation process). The above can improve data accuracy and review efficiency.
[0113] In this embodiment, the object file record can be located in the customer information master table. Therefore, based on the result of the identification process, the information to be completed in the target object file record can be completed by writing the target field value obtained from the identification process into the corresponding missing field of the corresponding record in the customer information master table. Optionally, when writing to the customer information master table, an optimistic locking and version control data writing strategy can be adopted to ensure the atomicity and traceability of data updates. Specifically, a version number field (version) is added to the customer information master table to identify the updated version of each record. During each update operation, the following process can be executed: (1) Read the current customer record and obtain the version number v1; (2) Execute the data update and specify version=v1 in the WHERE condition so that the update is only allowed when the version number is v1; (3) If the update is successful, the version number is automatically incremented to v2; (4) If the update fails, the latest record is automatically reread to obtain the new version number v2, and the update operation is retried. Through the above optimistic locking mechanism, the data overwrite problem caused by concurrent updates is avoided.
[0114] Optionally, for batch completion tasks, such as completing thousands of customer files at once, the following optimization strategies are adopted to improve efficiency and stability: (1) Batch submission: such as 1,000 records per batch to avoid long transactions occupying database resources; (2) Asynchronous writing: multiple batches of writing operations are executed in parallel through a thread pool; (3) Retry on failure: for records that fail to write, a maximum of 3 retries are made, and if they still fail, they are recorded in the exception table; (4) Transaction rollback: when an exception occurs in a single batch, the entire batch is rolled back to ensure data consistency.
[0115] In some optional implementations, to meet regulatory audit requirements, historical version records are retained while updating the master customer information table. Specifically, a customer information history table (t_customer_info_history) can be created to record detailed information for each data change. The reason for each data change is noted to facilitate subsequent statistical analysis; the reasons can include: OCR auto-completion, multi-source data fusion, manual modification, customer self-update, and system error correction. Optionally, it can also support tracing back to historical versions of customer information by time point, such as querying all field values of a customer at a specified time point, comparing field differences between two time points, and generating data change audit reports.
[0116] S111, if the field type is a deducible field, the field to be completed is supplemented in the target object file record based on the value of the field to be completed extracted from the target object's document number; if the field type is a default value field, the field to be completed is supplemented in the target object file record based on the default fill value corresponding to the target object's document type.
[0117] Specifically, if the field type is an inferable field (such as gender, date of birth, or region), the rule extraction template is directly called, returning a routing status indicating rule extraction, such as RULE_EXTRACTION. This allows the field value to be derived from the ID card number to complete the corresponding field information. If the field type is a default value field (such as nationality), the default value filling template is directly called, returning a routing status indicating the use of default values, such as USE_DEFAULT. This allows the default filling value corresponding to the ID card type to be used as the field value to complete the corresponding field information.
[0118] Please see Figure 5 The diagram shown is a system architecture diagram for implementing a method to complete compliance risk field information, as provided in an embodiment of this application. Figure 5 The system includes a user entry point, a data acquisition and scheduling layer, a data quality assessment layer, an intelligent repair decision layer, an execution and post-processing layer, and a persistent monitoring layer.
[0119] User access points include targeted scanning triggered by the compliance management console, customer apps / branch counters, and third-party partner platforms.
[0120] The first layer of data acquisition and scheduling includes: XXL - JOB Distributed Scheduling Center (used to implement full scan, incremental scan, targeted scan, segmented broadcast, failure retry, and cluster scheduling); online channel adapter (collects APP / webpage registration and data modification); offline outlet adapter (collects counter entry / verification data); and third-party platform adapter (collects customer data from cooperative channels).
[0121] The second layer of data quality assessment includes: a core field rule base (definitions of nine core fields, rules for non-empty / format / logical / dictionary / checksum, and field color relationship configuration); a field quality assessment engine (data preprocessing, multi-level field-by-field validation, and missing field classification); and a list of tasks to be completed (customer ID + a list of missing fields sorted by priority).
[0122] The third layer of intelligent repair decision-making includes: a multi-source data fusion engine (cross-channel data aggregation, credibility assessment of data source authority + time decay + historical accuracy, conflict detection and optimal value selection); and an OCR intelligent routing engine (field type analysis, cache reuse check, API quota management, and degradation strategy).
[0123] The fourth layer of execution and post-processing includes: a manual review module (marking low-confidence fields, highlighting original images, recommending suggested values, one-click adoption and manual modification), a post-processing verification module (verification code verification, dictionary matching, date logic verification, cross-field consistency check), an OCR engine call module (high-precision OCR, standard OCR, remote API call, result cache writing), and an asynchronous queue consumption module (priority queue, quota counter fallback monitoring, timed dequeue consumption).
[0124] The fifth layer of persistent monitoring includes: the main customer information table (optimistic locking + version control writing, storage of nine core fields), the customer information history table (snapshots before and after changes, classification of change reasons, and support for version backtracking), the Redis cache cluster (OCR result cache pool, API quota counter, asynchronous priority queue), and the data quality monitoring panel (field completeness rate, OCR recognition accuracy rate, automatic repair success rate, and manual review rate).
[0125] Please see Figure 6 The diagram shown is a specific example of the compliance risk field information completion method provided in this application embodiment. Figure 6 As shown, it may include: S1 integrates the XXL-JOB distributed task scheduling platform, which triggers full or incremental data scanning tasks on a regular basis according to time periods and business channels.
[0126] S2, based on a pre-defined anti-money laundering field rule base, assesses the completeness of each customer's fields and identifies missing fields.
[0127] S3. For fields with multiple candidate values, construct a data credibility assessment model, comprehensively consider the authority of the data source, the time decay factor, and historical accuracy, calculate the confidence score of each candidate value, and select the value with the highest confidence.
[0128] S4. For fields that still need to be completed, the OCR intelligent routing engine is activated, and the OCR recognition strategy is dynamically selected based on four dimensions: field type, cache reuse, and API call resource optimization.
[0129] S5 calls the selected OCR engine for recognition and performs post-processing verification through check code verification, dictionary matching, date logic verification, and cross-field consistency checks.
[0130] In step S6, when the corrected confidence level is greater than or equal to 0.85, optimistic locking and version control are used to write the completed data into the main customer information table, and a snapshot of the change is saved to the historical table, supporting version rollback and auditing. When the corrected confidence level is less than 0.85, manual review is triggered, and the result of the manual review is associated with step S3 to form a feedback loop, updating the historical accuracy of the data source and updating the OCR confidence benchmark.
[0131] Please see Figure 7 The diagram illustrates a specific example of OCR intelligent routing decision-making provided in this application. Specifically, upon receiving a field completion task from S3 that is still missing after fusion, the field type is determined. If it is a deducible field, the routing status is set to "RULE_EXTRACTION" to indicate that gender / date of birth / region is deduced from the ID card number. If it is a default value field, the routing status is set to "USE_DEFAULT" to indicate that the nationality is used as the default field. If it is a key field or a text field, the system further determines whether the cache has been hit and whether the confidence level is greater than 0.9. If the cache has been hit and the confidence level is greater than 0.9, the routing status is set to "USE_CACHE" to indicate that the cached identification value is returned and the cache hit count is incremented by 1. If a cache miss occurs or the confidence level is less than 0.9, the field priority is further determined. If it is a high-priority field (such as a critical field), the API call counter is checked to see if it is lower than the higher quota threshold (HIGH_PRIORITY_QUOTA) associated with the high-priority task. If it is lower, the quota is available, the routing status is set to "DIRECT_OCR" to indicate the call to the high-precision OCR engine for synchronous recognition, and the call counter is incremented by 1. If it is not lower, the quota has been reached, and the routing status is set to "ASYNC_QUEUE" to indicate that the field is placed at the head of the asynchronous queue for priority dequeue processing. If it is a medium-priority field (text field), the API call counter is checked to see if it is lower than the medium quota threshold (MEDIUM_PRIORITY_QUOTA) associated with the medium-priority task. If it is lower, the quota is available, the routing status is set to "DIRECT_OCR" to indicate the call to the standard OCR engine for synchronous recognition, and the call counter is incremented by 1. If it is not lower, the quota has been reached, and the routing status is set to "ASYNC_QUEUE" to indicate that the field is placed in the middle of the asynchronous queue to wait for the counter to fall back. Optionally, if a low priority is set (derivable field / default value field), then when a low priority is determined, the routing state is directly set to "ASYNC_QUEUE" to indicate that the task is added to the end of the asynchronous queue, skipping the API quota check. The asynchronous queue consumption thread periodically checks the counter's fallback status and dequeues tasks according to priority to trigger the OCR call. After the above routing decision is completed, the task object can be output to step S5.
[0132] By automating full-scale inspections, coverage was increased from 5% to 100%, significantly improving efficiency. Through multi-channel data fusion, missing fields can be directly retrieved from other sources; OCR intelligent routing automatically completes remaining fields, greatly improving the overall automatic repair rate and significantly reducing manual intervention. This avoids wasting significant manpower and resources on manual repairs by avoiding unnecessary OCR calls through an intelligent routing mechanism, significantly reducing call costs compared to full-scale OCR solutions. Supporting near real-time monitoring and repair at T+1 level meets regulatory timeliness requirements; compared to the traditional T+30 batch processing method, the risk exposure window is significantly shortened, and scalability is excellent.
[0133] Please see Figure 8 The diagram shown is a feedback closed-loop flowchart based on manual review provided in an embodiment of this application. Figure 8 The following are included: Manual review complete: Reviewers make a final judgment on fields marked "Pending manual review" to determine the correct field values. Next, the review results are collected: The system automatically collects the customer ID and corresponding field name, OCR recognition results, multi-source fusion candidate values and system suggested values, the final manually adopted value, the review operation time, and the operator's employee number. The system then compares the manually adopted value with candidate values from each data source to determine the source of discrepancies. If candidate values from any data source are adopted or not adopted, the historical accuracy of the data source is dynamically updated; if OCR recognition results are frequently corrected, the OCR cache confidence benchmark is adjusted; if new valid field values not covered by the dictionary exist, the dictionary is automatically expanded. The updated results are then persisted: The above three types of update results are written to their respective storage, including the data source confidence statistics table (updating historical accuracy fields), the OCR cache configuration table (updating confidence benchmark parameters), and the standard dictionary table (adding new dictionary entries). Next, the feedback takes effect, influencing the next routing decision. Specifically: the multi-source fusion engine reads the updated historical accuracy of the data source, thus affecting the confidence calculation; the OCR intelligent routing engine reads the updated confidence benchmark, thus affecting engine selection and caching strategies; the post-processing verification module uses the expanded dictionary to improve the matching success rate and reduce the amount of manual review. This forms a persistent and complete closed loop. As the system runs, it continuously improves its automated processing capabilities and accuracy, gradually reducing its reliance on manual intervention.
[0134] Corresponding to the compliance risk field information completion methods provided in the above embodiments, this application also provides a compliance risk field information completion device. Since the compliance risk field information completion device provided in this application corresponds to the compliance risk field information completion methods provided in the above embodiments, the implementation methods of the aforementioned compliance risk field information completion methods are also applicable to the compliance risk field information completion device provided in this embodiment, and will not be described in detail in this embodiment.
[0135] Please see Figure 9 The diagram shows a structural schematic of a compliance risk field information completion device provided in an embodiment of this application. This device has the function of implementing the compliance risk field information completion method described in the above method embodiment. This function can be implemented by hardware or by hardware executing corresponding software. Figure 9 As shown, the compliance risk field information completion device 900 may include: The multi-source data fusion module 910 is used to respond to a field completion task for at least one missing compliance risk field in the target object file record, and to perform multi-source data fusion processing on each of the missing compliance risk fields based on data from multiple data sources to obtain a multi-source data fusion result; The field type determination module 920 is used to determine the field type of the field to be completed when it is determined, based on the multi-source data fusion result, that there are still fields to be completed; the at least one missing compliance risk field includes the field to be completed; OCR quota determination module 930 is used to determine a target OCR interface call count threshold that matches the priority of the target field type based on OCR interface call configuration information when the field type is a target field type; the target field type is a key field or a text field; the OCR interface call configuration information indicates the OCR interface call count threshold corresponding to different priorities. OCR interface calling module 940 is used to call a preset OCR interface if the number of OCR interface calls in the current time window is less than the target OCR interface call threshold, and to recognize and process the target object document image corresponding to the target object file record through the OCR engine associated with the preset OCR interface. The information completion module 950 is used to complete the information of the field to be completed in the target object file record based on the result of the recognition process.
[0136] In some embodiments, the device further includes a cache query module, which is used to: generate a query cache key using a preset hash algorithm based on the target object document image, the field identifier of the field to be completed, and the current engine version of the OCR engine; search for a target cache value matching the query cache key in the OCR recognition result cache based on the query cache key; the OCR recognition result cache is used to store the OCR recognition value of the field in the form of key-value pairs; if the target cache value is found, complete the information of the field to be completed in the target object file record based on the OCR recognition value indicated by the target cache value; if the target cache value is not found, perform the step of determining the target OCR interface call count threshold that matches the priority of the target field type based on the OCR interface call configuration information.
[0137] In some embodiments, the apparatus further includes an asynchronous module, the asynchronous module being used for: If the number of OCR interface calls within the current time window is greater than or equal to the target OCR interface call threshold, then the call request to the preset OCR interface is encapsulated into a priority task that matches the priority of the target field type; the priority task is stored in an asynchronous priority queue; in response to the OCR interface call counter falling back to a preset safety value, the priority task in the asynchronous priority queue is executed; the OCR interface call counter is used to record the number of OCR interface calls within the current time window in a sliding window manner.
[0138] In some implementations, the priority of the key field is higher than that of the text field, and the threshold for the number of OCR interface calls corresponding to the key field is greater than the threshold for the number of OCR interface calls corresponding to the text field; when storing the priority task in the asynchronous priority queue, the asynchronous module has the following functions: if the target field type is a key field, insert the priority task at the head of the asynchronous priority queue; if the target field type is a text field, insert the priority task in the middle of the asynchronous priority queue.
[0139] In some implementations, the multi-source data fusion module 910 is specifically configured to: for each missing compliance risk field, search for candidate values corresponding to the missing compliance risk field from the data of the multiple data sources respectively; for a missing compliance risk field for which multiple candidate values are found, determine a comprehensive confidence score for each candidate value based on the time decay factor of each candidate value and the historical accuracy of the data source to which the candidate value belongs on the missing compliance risk field; determine whether the missing compliance risk field is a field to be completed based on the candidate value with the highest comprehensive confidence score; for a missing compliance risk field for which no candidate value is found, determine that the missing field is a field to be completed; wherein, the multi-source data fusion result indicates the field to be completed among the at least one missing compliance risk field.
[0140] In some implementations, when the multi-source data fusion module 910 determines whether a missing compliance risk field is a field to be completed based on the candidate value of the highest comprehensive confidence score, it is specifically used to: determine the corresponding missing compliance risk field as a field to be completed if the highest comprehensive confidence score is less than a preset score threshold; if the highest comprehensive confidence score is greater than or equal to the preset score threshold, and the score difference between the highest comprehensive confidence score and the second highest comprehensive confidence score is less than a score difference threshold, then determine the target candidate value corresponding to the missing compliance risk field based on the authority of the data source corresponding to the candidate value of the highest comprehensive confidence score and the authority of the data source corresponding to the candidate value of the second highest comprehensive confidence score; perform a cross-field logical consistency check on each missing compliance risk field with a target candidate value, and determine the missing compliance risk field with logical contradictions indicated by the check results as a field to be completed.
[0141] In some implementations, the result of the identification process includes the identification results of multiple compliance risk fields and the identification confidence level corresponding to each compliance risk field; the information completion module 950 is specifically used for: performing post-processing verification based on the identification results of the multiple compliance risk fields to obtain post-processing verification results; the post-processing verification includes verification code verification, dictionary matching verification, date logic verification, and cross-field consistency verification; determining a verification pass coefficient based on the post-processing verification results; correcting the identification confidence level of each compliance risk field based on the verification pass coefficient to obtain a corrected identification confidence level corresponding to each compliance risk field; and, if the corrected identification confidence level corresponding to each compliance risk field meets a preset confidence threshold condition, performing information completion on the field to be completed in the target object file record based on the result of the identification process.
[0142] In some implementations, the information completion module 950 is further configured to: trigger manual review of the compliance risk field for a compliance risk field whose corrected identification confidence does not meet the preset confidence threshold condition, and obtain the manual review result corresponding to the compliance risk field; compare the manual review result corresponding to the compliance risk field with the candidate value of the compliance risk field in each of the data sources, and update the historical accuracy of each of the data sources in the compliance risk field based on the comparison result.
[0143] In some implementations, the non-target field type includes deducible fields and default value fields. The information completion module 950 is further configured to: when the field type is a deducible field, complete the information of the field to be completed in the target object file record based on the value of the field to be completed extracted from the target object's document number; when the field type is a default value field, complete the information of the field to be completed in the target object file record based on the default fill value corresponding to the target object's document type.
[0144] In some embodiments, the apparatus further includes a field completion task generation module, which is used to, in response to a scheduled task triggering instruction, obtain a set of object file records to be processed based on the scheduled task type; perform field integrity checks on each object file record in the set of object file records based on preset compliance risk field rule information, and obtain a field integrity check result; the field integrity check result indicates that at least one target object file record has a missing compliance risk field; and generate a field completion task corresponding to each target object file record based on the field integrity check result.
[0145] It should be noted that the device provided in the above embodiments is only illustrated by the division of the above functional modules when realizing its functions. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. For example, Figure 9 The functional modules shown are as follows. Furthermore, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and their specific implementation processes are detailed in the method embodiments, and will not be repeated here.
[0146] This application provides an electronic device, which includes a processor and a memory. The memory stores at least one instruction or at least one program. The at least one instruction or at least one program is loaded and executed by the processor to implement any of the compliance risk field information completion methods provided in the above method embodiments.
[0147] The methods and embodiments provided in this application can be executed in a computer terminal, server, or similar computing device; that is, the aforementioned electronic device may include a computer terminal, server, or similar computing device. Taking running on a server as an example... Figure 10 This is a hardware structure block diagram of a server that runs a compliance risk field information completion method according to an embodiment of the present invention, such as... Figure 10 As shown, the server 1000 can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 1010 (CPUs 1010 may include, but are not limited to, microprocessors (MCUs) or programmable logic devices (FPGAs), a memory 1030 for storing data, and one or more storage media 1020 (e.g., one or more mass storage devices) for storing application programs 1023 or data 1022. The memory 1030 and storage media 1020 may be temporary or persistent storage. The program stored in the storage media 1020 may include one or more modules, each module may include a series of instruction operations on the server. Furthermore, the CPU 1010 may be configured to communicate with the storage media 1020 and execute the series of instruction operations in the storage media 1020 on the server 1000. Server 1000 may also include one or more power supplies 1060, one or more wired or wireless network interfaces 1050, one or more input / output interfaces 1040, and / or one or more operating systems 1021, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.
[0148] The input / output interface 1040 can be used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of server 1000. In one example, the input / output interface 1040 includes a network interface controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the input / output interface 1040 may be a radio frequency (RF) module used for wireless communication with the Internet.
[0149] Those skilled in the art will understand that Figure 10 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, server 1000 may also include... Figure 10 The more or fewer components shown, or having the same Figure 10The different configurations shown.
[0150] Embodiments of this application also provide a computer-readable storage medium, which can be disposed in an electronic device to store at least one instruction or at least one program related to implementing a supplementary control method for resource applications. The at least one instruction or the at least one program is loaded and executed by the processor to implement any of the compliance risk field information completion methods provided in the above-described method embodiments.
[0151] The embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements any of the compliance risk field information completion methods provided in the above method embodiments.
[0152] Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.
[0153] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired result. Additionally, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0154] 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 apparatus embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.
[0155] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.
[0156] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.
Claims
1. A method for completing compliance risk field information, characterized in that, The method includes: In response to a field completion task for at least one missing compliance risk field in a target object file record, multi-source data fusion processing is performed on each missing compliance risk field based on data from multiple data sources to obtain a multi-source data fusion result; If, based on the results of the multi-source data fusion, it is determined that there are still fields to be completed, the field type of the fields to be completed is determined; the at least one missing compliance risk field includes the fields to be completed. When the field type is a target field type, a target OCR interface call count threshold that matches the priority of the target field type is determined based on the OCR interface call configuration information; the target field type is a key field or a text field; the OCR interface call configuration information indicates the OCR interface call count threshold corresponding to different priorities. If the number of OCR interface calls within the current time window is less than the target OCR interface call threshold, then the preset OCR interface is called, and the OCR engine associated with the preset OCR interface is used to recognize and process the target object document image corresponding to the target object file record. Based on the results of the identification process, the fields to be completed are supplemented with information in the target object file record.
2. The method according to claim 1, characterized in that, Before determining the target OCR interface call count threshold that matches the priority of the target field type based on the OCR interface call configuration information, the method further includes: Based on the target object's document image, the field identifier of the field to be completed, and the current engine version of the OCR engine, a query cache key is generated using a preset hash algorithm; Based on the query cache key, the target cache value that matches the query cache key is searched in the OCR recognition result cache; the OCR recognition result cache is used to store the OCR recognition value of the field in the form of key-value pairs; If the target cached value is found, the information of the field to be completed is completed in the target object file record based on the OCR recognition value indicated by the target cached value; If the target cache value is not found, the step of determining the target OCR interface call count threshold that matches the priority of the target field type based on the OCR interface call configuration information is performed.
3. The method according to claim 1 or 2, characterized in that, The method further includes: If the number of OCR interface calls within the current time window is greater than or equal to the target OCR interface call threshold, then the call request for the preset OCR interface will be encapsulated into a priority task that matches the priority of the target field type. Store the priority task in an asynchronous priority queue; In response to the OCR interface call counter falling back to a preset safety value, the priority task in the asynchronous priority queue is executed; the OCR interface call counter is used to record the number of OCR interface calls within the current time window in a sliding window manner.
4. The method according to claim 3, characterized in that, The priority of the key field is higher than that of the text field, and the threshold for the number of OCR interface calls corresponding to the key field is greater than the threshold for the number of OCR interface calls corresponding to the text field; storing the priority task in the asynchronous priority queue includes: If the target field type is a key field, then the priority task is inserted at the head of the asynchronous priority queue; If the target field type is a text field, then the priority task is inserted in the middle position of the asynchronous priority queue.
5. The method according to claim 1, characterized in that, The multi-source data fusion process is performed on each of the missing compliance risk fields based on data from multiple data sources to obtain the multi-source data fusion results, including: For each of the missing compliance risk fields, candidate values corresponding to the missing compliance risk fields are searched from the data of the multiple data sources respectively; For a missing compliance risk field for which multiple candidate values are found, a comprehensive confidence score is determined for each candidate value based on the time decay factor of each candidate value and the historical accuracy of the data source to which the candidate value belongs on the missing compliance risk field; based on the candidate value with the highest comprehensive confidence score, it is determined whether the missing compliance risk field is a field to be completed. For missing compliance risk fields where no candidate values were found, the missing field is determined to be a field to be completed; The multi-source data fusion result indicates the field to be completed in the at least one missing compliance risk field.
6. The method according to claim 5, characterized in that, The determination of whether the missing compliance risk field is a field to be completed is based on the candidate value with the highest comprehensive confidence score, including: If the highest overall confidence score is less than a preset score threshold, the corresponding missing compliance risk field is determined to be a field to be completed. If the highest overall confidence score is greater than or equal to the preset score threshold, and the score difference between the highest overall confidence score and the second highest overall confidence score is less than the score difference threshold, then the target candidate value corresponding to the missing compliance risk field is determined based on the authority of the data source corresponding to the candidate value of the highest overall confidence score and the authority of the data source corresponding to the candidate value of the second highest overall confidence score. A cross-field logical consistency check is performed on each of the missing compliance risk fields with target candidate values, and the missing compliance risk fields that indicate logical contradictions in the check results are identified as fields to be completed.
7. The method according to claim 1, characterized in that, The result of the identification process includes the identification results of multiple compliance risk fields and the identification confidence level corresponding to each compliance risk field; based on the result of the identification process, the information completion of the fields to be completed in the target object file record includes: Post-processing verification is performed based on the identification results of the multiple compliance risk fields to obtain the post-processing verification results; the post-processing verification includes verification code verification, dictionary matching verification, date logic verification, and cross-field consistency verification. Based on the post-processing verification results, the verification pass coefficient is determined; Based on the verification pass coefficient, the identification confidence of each compliance risk field is corrected to obtain the corrected identification confidence of each compliance risk field; If the confidence level of the corrected identification for each of the aforementioned compliance risk fields meets the preset confidence threshold, the information for the fields to be completed is supplemented in the target object file record based on the result of the identification process.
8. The method according to claim 7, characterized in that, The method further includes: For compliance risk fields whose corrected confidence level does not meet the preset confidence threshold, a manual review of the compliance risk field is triggered to obtain the manual review result corresponding to the compliance risk field. The manual review results corresponding to the compliance risk field are compared with the candidate values of the compliance risk field in each of the data sources, and the historical accuracy of each of the data sources in the compliance risk field is updated based on the comparison results.
9. The method according to claim 1, characterized in that, Non-target field types include inferable fields and default value fields, and the method further includes: When the field type is a derivable field, the information of the field to be completed is completed in the target object's file record based on the value of the field to be completed extracted from the target object's ID number; When the field type is a default value field, the information of the field to be filled is completed in the target object file record based on the default fill value corresponding to the target object document type.
10. The method according to claim 1, characterized in that, Before performing multi-source data fusion processing on each missing compliance risk field based on data from multiple data sources in response to a field completion task for at least one missing compliance risk field in a target object file record, the method further includes: In response to a scheduled task trigger command, retrieve the set of object file records to be processed based on the scheduled task type; Based on the preset compliance risk field rule information, a field integrity check is performed on each object file record in the object file record set to obtain a field integrity check result; the field integrity check result indicates that at least one target object file record has a missing compliance risk field; Based on the field integrity detection results, a field completion task is generated for each target object file record.
11. A device for completing compliance risk field information, characterized in that, The device includes: The multi-source data fusion module is used to respond to a field completion task for at least one missing compliance risk field in the target object file record, and to perform multi-source data fusion processing on each missing compliance risk field based on data from multiple data sources to obtain a multi-source data fusion result; The field type determination module is used to determine the field type of the field to be completed when it is determined, based on the multi-source data fusion result, that there are still fields to be completed; the at least one missing compliance risk field includes the field to be completed. The OCR quota determination module is used to determine, based on the OCR interface call configuration information, a target OCR interface call count threshold that matches the priority of the target field type when the field type is a target field type; the target field type is a key field or a text field; the OCR interface call configuration information indicates the OCR interface call count threshold corresponding to different priorities. The OCR interface call module is used to call a preset OCR interface if the number of OCR interface calls within the current time window is less than the target OCR interface call threshold, and then use the OCR engine associated with the preset OCR interface to recognize and process the target object document image corresponding to the target object file record. The information completion module is used to complete the information of the field to be completed in the target object file record based on the result of the recognition process.
12. An electronic device, characterized in that, The method includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the compliance risk field information completion method as described in any one of claims 1 to 10.
13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one instruction or at least one program segment, which is loaded and executed by a processor to implement the compliance risk field information completion method as described in any one of claims 1 to 10.
