Adaptive integration method and device for heterogeneous credit data

Abstract

The invention provides an adaptive integration method and device for heterogeneous credit data, and belongs to the technical field of credit data processing, and the method comprises the steps: obtaining heterogeneous credit features of all enterprises in a designated region; the heterogeneous credit features comprise structured credit features corresponding to the structured credit data of each enterprise and unstructured credit features in the unstructured credit data of each enterprise; extracting behavior information, subject information and spatio-temporal information in the heterogeneous credit features of each enterprise; and taking people, organizations and events as core entities, and obtaining credit knowledge maps corresponding to the enterprises based on the association relationship integration of the behavior information, the subject information and the spatio-temporal information corresponding to the enterprises, so as to take the credit knowledge maps corresponding to the enterprises as multi-dimensional association credit archives of the enterprises. According to the adaptation integration method and device for the heterogeneous credit data provided by the invention, the reliability and accuracy of enterprise credit evaluation can be improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of credit data processing, and more particularly relates to a heterogeneous credit data adaptive integration method and device. BACKGROUND

[0002] With the advancement of government informationization and smart city construction, credit data presents a multi-source heterogeneous characteristic, covering structured, semi-structured and unstructured data generated by multiple subjects such as government departments (such as political and legal, development and reform, market supervision), social organizations and Internet platforms.

[0003] At present, credit data with multi-source heterogeneous characteristics has barriers to intercommunication due to non-uniform standards and different formats. The existing technology generally cannot realize multi-dimensional correlation analysis of enterprise credit behavior, resulting in a large deviation between the evaluation result and the actual credit level. SUMMARY

[0004] The purpose of the present application is to provide a heterogeneous credit data adaptive integration method and device to solve the problem that the existing technology cannot realize multi-dimensional correlation analysis of enterprise credit behavior, resulting in a large deviation between the evaluation result and the actual credit level.

[0005] The first aspect of the embodiment of the present application provides a heterogeneous credit data adaptive integration method, comprising: obtaining heterogeneous credit characteristics of each enterprise in a specified area; the heterogeneous credit characteristics include structured credit characteristics corresponding to structured credit data of each enterprise, and unstructured credit characteristics in unstructured credit data of each enterprise extracted based on a pre-trained natural language processing model and / or a pre-trained visual target detection model; extracting behavior information, subject information and space-time information in the heterogeneous credit characteristics of each enterprise; integrating credit knowledge graphs corresponding to each enterprise based on the association relationship of the behavior information, the subject information and the space-time information corresponding to each enterprise, taking people, organizations and events as core entities, to take the credit knowledge graphs corresponding to each enterprise as multi-dimensional correlation credit archives of each enterprise.

[0006] In an embodiment of the present application, the unstructured credit data of each enterprise includes text data and visual data; Before obtaining the heterogeneous credit characteristics of each enterprise in the specified area, the heterogeneous credit data adaptive integration method further comprises: extracting entity information and semantic relationship in the text data based on the pre-trained natural language processing model to obtain unstructured credit characteristics, or extracting behavior information and scene text information in the visual data based on the pre-trained visual target detection model to obtain unstructured credit characteristics.

[0007] In an embodiment of the present application, after extracting the behavior information, subject information and space-time information in the heterogeneous credit features of each enterprise, the adaptive integration method of heterogeneous credit data further comprises: If the target information contains heterogeneous credit features of different modalities, the weights of each heterogeneous credit feature of different modalities are determined by using a graph attention network, the heterogeneous credit features of different modalities are converted to the same modality by using a pre-trained adversarial learning network, and the feature fusion of the heterogeneous credit features in the target information is performed based on the weights of each heterogeneous credit feature under the same modality. Wherein, the target information is behavior information, subject information or space-time information.

[0008] In an embodiment of the present application, the adaptive integration method of heterogeneous credit data further comprises: In response to a data change event triggered by a specified credit regulatory subject, the field change feature corresponding to the data change event is extracted, the data conversion rule in the credit knowledge graph is determined by using the field change feature and a pre-trained graph neural network, and the credit knowledge graph is updated based on the data conversion rule.

[0009] In an embodiment of the present application, the method further comprises: Converting the multi-dimensional associated credit archives of each enterprise in the plurality of specified regions based on the predetermined credit data standard, to perform credit evaluation on each enterprise in the plurality of specified regions based on the converted multi-dimensional associated credit archives.

[0010] In an embodiment of the present application, converting the multi-dimensional associated credit archives of each enterprise in the plurality of specified regions based on the predetermined credit data standard comprises at least one of: Converting the fields in the multi-dimensional associated credit archives of each enterprise in the plurality of specified regions based on the predetermined credit data standard; Converting the credit evaluation standards in the multi-dimensional associated credit archives of each enterprise in the plurality of specified regions based on the predetermined credit data standard.

[0011] In an embodiment of the present application, before converting the multi-dimensional associated credit archives of each enterprise in the plurality of specified regions based on the predetermined credit data standard, the adaptive integration method of heterogeneous credit data further comprises: Deploying a credit mutual recognition smart contract on a consortium chain corresponding to the plurality of specified regions, to perform the step of converting the multi-dimensional associated credit archives of each enterprise in the plurality of specified regions based on the predetermined credit data standard by using the credit mutual recognition smart contract.

[0012] The second aspect of the embodiments of the present application provides an adaptive integration device of heterogeneous credit data, comprising: The feature acquisition module is used to acquire heterogeneous credit features of various enterprises in a specified region. The heterogeneous credit features include the structured credit features corresponding to the structured credit data of various enterprises, as well as the unstructured credit features of various enterprises extracted from the unstructured credit data of various enterprises based on pre-trained natural language processing models and / or pre-trained visual object detection models. The information extraction module is used to extract behavioral information, subject information, and spatiotemporal information from the heterogeneous credit characteristics of various enterprises; The credit association module is used to integrate the credit knowledge graphs of each enterprise based on the relationship between the behavioral information, subject information and spatiotemporal information of each enterprise, with people, organizations and events as the core entities. The credit knowledge graphs of each enterprise are used as multi-dimensional credit profiles of each enterprise.

[0013] In one embodiment of this application, the unstructured credit data of each enterprise includes text data and visual data; Before acquiring the heterogeneous credit characteristics of various enterprises in a specified region, the device also includes a feature extraction module; The feature extraction module is used for: Based on pre-trained natural language processing models, entity information and semantic relationships are extracted from text data to obtain unstructured credit features, or Based on a pre-trained visual object detection model, behavioral information and scene text information are extracted from visual data to obtain unstructured credit features.

[0014] In one embodiment of this application, after extracting behavioral information, subject information, and spatiotemporal information from the heterogeneous credit features of various enterprises, the device further includes a feature fusion module; the feature fusion module is used for: If the target information contains heterogeneous credit features of different modalities, then the graph attention network is used to determine the weights of each heterogeneous credit feature of different modalities, and the pre-trained adversarial learning network is used to transform the heterogeneous credit features of different modalities into the same modality. Under the same modality, the heterogeneous credit features in the target information are fused based on the weights of each heterogeneous credit feature. The target information can be behavioral information, subject information, or spatiotemporal information.

[0015] In one embodiment of this application, the heterogeneous credit data adaptation and integration device further includes a graph update module; the graph update module is used for: In response to data change events triggered by a designated credit regulatory body, the system extracts the field change features corresponding to the data change events, uses the field change features and a pre-trained graph neural network to determine the data transformation rules in the credit knowledge graph, and updates the credit knowledge graph based on the data transformation rules.

[0016] In one embodiment of this application, the heterogeneous credit data adaptation and integration device further includes a credit evaluation module, which is used for: Based on predetermined credit data standards, the multidimensional related credit profiles of various enterprises in multiple designated regions are transformed, and the credit evaluation of various enterprises in multiple designated regions is carried out based on the transformed multidimensional related credit profiles.

[0017] In one embodiment of this application, the credit rating module is specifically used to: transform fields in the multidimensional associated credit profiles of various enterprises in multiple designated regions based on predetermined credit data standards, or Based on predetermined credit data standards, the credit evaluation standards in the multidimensional related credit files of various enterprises in multiple designated regions are converted. In one embodiment of this application, before converting the multidimensional associated credit profiles of various enterprises in multiple designated regions based on a predetermined credit data standard, the heterogeneous credit data adaptation and integration device further includes a credit mapping module; the credit mapping module is used for: Deploy credit mutual recognition smart contracts on consortium blockchains corresponding to multiple designated regions to execute steps for converting multidimensional related credit profiles of various enterprises in multiple designated regions based on predetermined credit data standards.

[0018] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the above-described method for adapting and integrating heterogeneous credit data.

[0019] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method for adapting and integrating heterogeneous credit data.

[0020] The beneficial effects of the method and apparatus for adapting and integrating heterogeneous credit data provided in this application are as follows: This application embodiment obtains heterogeneous credit features of various enterprises in a designated region, extracts behavioral information, subject information, and spatiotemporal information from these features, and thus obtains effective credit information for each enterprise. Finally, using people, organizations, and events as core entities, and integrating the relationships between the behavioral, subject, and spatiotemporal information of each enterprise, a credit knowledge graph for each enterprise is obtained. This credit knowledge graph serves as a multi-dimensional credit profile for each enterprise, effectively uncovering the relationships between various credit features and enabling multi-dimensional credit assessment of enterprises, ensuring consistency between the assessment results and actual credit levels as much as possible. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the 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.

[0022] Figure 1 A flowchart illustrating a method for adapting and integrating heterogeneous credit data according to an embodiment of this application; Figure 2 A structural block diagram of an apparatus for adapting and integrating heterogeneous credit data provided in an embodiment of this application; Figure 3 This is a schematic block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0023] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0024] To make the objectives, technical solutions, and advantages of this application clearer, the following description will be provided in conjunction with the accompanying drawings and specific embodiments.

[0025] In the construction of a social credit system, the integration and fusion of credit data across departments and regions is a core foundation for achieving the modernization of social governance. Credit data covers multiple administrative departments and may include, for example, basic information of natural persons, corporate credit records, grid inspection data, and false information on the internet, exhibiting multi-source and heterogeneous characteristics.

[0026] The following problems exist: Interoperability barriers arise from inconsistent standards and formats across multiple data sources; governance fragmentation occurs due to the lack of cross-domain data sharing mechanisms; and traditional data integration technologies struggle to meet the real-time demands of dynamic credit scenarios.

[0027] Existing technologies prioritize depth of integration but lack the ability to semantically understand and extract features from unstructured data (such as false information text and video surveillance data), making it difficult to achieve multi-dimensional correlation analysis of credit behavior. Furthermore, the lack of a mechanism to compensate for differences in cross-domain data during model training leads to a significant drop in the accuracy of credit evaluation models in cross-regional scenarios. There is an urgent need for a method for integrating and fusing credit data that is both adaptable and systematic.

[0028] In this embodiment, the designated region can be one or more. Each designated region may include one or more administrative areas, and each administrative area may include multiple government departments. Different government departments have different data interfaces, through which credit data can be collected or monitored. The credit data may include personal credit data or corporate credit data. Each data interface may be deployed with a smart contract. Cross-domain mutual recognition smart contracts are set up among the administrative areas within the designated region, enabling reliable data exchange.

[0029] In this embodiment, the essence of a knowledge graph is to weave scattered information into a logical network using "entities (nodes)" and "relationships (edges)". The core entities, as the "nodes" of the knowledge graph, primarily carry information and prevent its dispersion. Relationships, as the attribute relationships connecting entities, mainly function to construct the network.

[0030] In this embodiment, the executing entity can be a regional data monitoring center in a designated area, data monitoring stations in various administrative regions, credit monitoring and evaluation platforms in various administrative regions, cloud detection platforms, or other data processing platforms. The specific entity can be selected based on actual circumstances, and this embodiment does not impose any restrictions.

[0031] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating a method for adapting and integrating heterogeneous credit data according to an embodiment of this application. The method may include steps S101 to S103.

[0032] S101: Obtain the heterogeneous credit characteristics of various enterprises in the specified region.

[0033] Heterogeneous credit features include structured credit features corresponding to the structured credit data of each enterprise, as well as unstructured credit features extracted from the unstructured credit data of each enterprise based on pre-trained natural language processing models and / or pre-trained visual object detection models.

[0034] This application embodiment can obtain credit data of various enterprises in a specified region, and extract features from the credit data to obtain multimodal structured credit features and unstructured credit features as heterogeneous credit features.

[0035] Specifically, a company's credit data can include both structured and unstructured credit data. Structured credit data can include administrative penalty records, corporate dishonesty records, and other standard-collected credit data. Unstructured data can include inaccurate text messages, video surveillance data, and other non-standard-collected credit data.

[0036] After obtaining the structured credit data of each enterprise, feature extraction can be performed on the structured credit data to obtain the corresponding structured credit features. For example, the structured credit data can be input into a pre-trained feature extraction model to extract features and obtain the structured credit features of each enterprise.

[0037] After obtaining the unstructured credit data from various enterprises, feature extraction can be performed using either a pre-trained natural language processing model or a pre-trained visual object detection model, depending on the type of unstructured credit data. Specifically, when the unstructured credit data is text, it can be input into a pre-trained natural language processing model for feature extraction to obtain the corresponding unstructured credit features. When the unstructured credit data is video, it can be input into a pre-trained visual object detection model for feature extraction to obtain the corresponding unstructured credit features. Text data can include false information, such as misinformation in chat groups or public announcements on social media platforms. Video data can include videos or images of violations, such as illegal parking, littering, or illegal lane changes.

[0038] After obtaining the credit data of various enterprises, this application embodiment can adopt different feature extraction methods for different data types to obtain heterogeneous credit features matching each enterprise, which helps to conduct accurate analysis in the future and improves the consistency between the assessment results and the actual credit level.

[0039] S102: Extract behavioral information, subject information, and spatiotemporal information from the heterogeneous credit characteristics of each enterprise.

[0040] After obtaining the heterogeneous credit characteristics of each enterprise, the embodiments of this application can filter or classify the heterogeneous credit characteristics of each enterprise to obtain behavioral information, subject information and spatiotemporal information.

[0041] Specifically, behavioral information can represent a company's actions or events. For example, it could correspond to a company's "false advertising" or "complaints" on social media.

[0042] The subject information can represent the perpetrator of the behavior or the relevant object. For example, it can be "Company A" that engages in false advertising or "Company B" that initiates a complaint.

[0043] Spatiotemporal information can represent the time or space range in which an action occurs. For example, it can be the "time interval" of Company A's false advertising, the "release time" of a complaint text, or the "filming location" of an illegal video.

[0044] For example, in this embodiment, information location can be used to traverse various heterogeneous credit features, and behavioral information, subject information, and spatiotemporal information can be located using keywords, filtering out redundant descriptions. Then, the attribute dimensions of each located heterogeneous credit feature are further broken down to ensure the completeness of each information point, separating behavioral information, subject information, and spatiotemporal information. Finally, the subject information is integrated according to the logic of linking behavioral information with spatiotemporal information, to obtain the corresponding behavioral information, subject information, and spatiotemporal information for each enterprise.

[0045] S103: Using people, organizations, and events as core entities, and integrating the relationships between behavioral information, subject information, and spatiotemporal information of each enterprise, a credit knowledge graph corresponding to each enterprise is obtained, which serves as a multi-dimensional associated credit file for each enterprise.

[0046] In this embodiment, "person" refers to a natural person related to the enterprise, such as a legal representative, shareholder, or senior executive. "Organization" refers to an institution associated with the enterprise, such as the enterprise itself, upstream and downstream partners, or regulatory agencies. "Event" refers to a specific behavioral event related to the enterprise's credit, such as performance of obligations, default, administrative penalties, tax payment, or financing.

[0047] This application embodiment can use people, organizations, and events as nodes of a knowledge graph, and the relationships between behavioral information, subject information, and spatiotemporal information corresponding to each enterprise as edges of the knowledge graph, to integrate and obtain a credit knowledge graph corresponding to each enterprise in a specified region.

[0048] After obtaining the credit knowledge graph corresponding to each enterprise, a multi-dimensional related credit profile can be formed based on the credit knowledge graph, which can make the credit evaluation of each enterprise more three-dimensional, comprehensive and accurate.

[0049] This application embodiment constructs a three-dimensional association network of "behavior-subject-spatiotemporal" through knowledge graph, which can align structured and unstructured data into entities and realize multi-dimensional associated credit files for various enterprises.

[0050] For example, by constructing a credit knowledge graph, the administrative penalty record of "a certain enterprise's false advertising" can be associated with complaint texts on social media and on-site monitoring videos through the enterprise's unified social credit code, forming a multi-dimensional associated credit file for the enterprise.

[0051] Currently, the weak multimodal credit data fusion capability in corporate credit evaluation leads to a single dimension of credit evaluation. Consequently, credit governance can only rely on structured government data (such as administrative penalty records), and cannot integrate unstructured data such as false information texts and surveillance videos. This results in a lack of scenario-based support for credit evaluation, making it difficult to accurately depict complex credit behaviors (such as the spread of false information online and community violations). Consequently, the deviation rate between the evaluation results and the actual credit level exceeds 30%.

[0052] To address the aforementioned issues, this application's embodiments construct credit knowledge graphs for each enterprise, which can expand the credit evaluation dimensions from the traditional 5-8 to 20+, significantly improving the matching degree between enterprise evaluation results and actual credit levels.

[0053] This application embodiment obtains heterogeneous credit features of various enterprises in a designated region, extracts behavioral information, subject information, and spatiotemporal information from these features, and thus obtains effective credit information for each enterprise. Finally, using people, organizations, and events as core entities, and integrating the relationships between the behavioral, subject, and spatiotemporal information of each enterprise, a credit knowledge graph for each enterprise is obtained. This credit knowledge graph serves as a multi-dimensional credit profile for each enterprise, effectively uncovering the relationships between various credit features and enabling multi-dimensional credit assessment of enterprises, ensuring consistency between the assessment results and actual credit levels as much as possible.

[0054] In embodiments of this application, the unstructured credit data of each enterprise includes text data and visual data.

[0055] Before obtaining the heterogeneous credit features of various enterprises in a specified region, the adaptive integration method of this heterogeneous credit data also includes: extracting entity information and semantic relationships from text data based on a pre-trained natural language processing model to obtain unstructured credit features, or extracting behavioral information and scene text information from visual data based on a pre-trained visual object detection model to obtain unstructured credit features.

[0056] In this embodiment, the natural language processing model can be a bidirectional pre-trained language model based on the Transformer architecture (idirectional Encoder Representations from Transformers-base, BERT-base). This embodiment can obtain a pre-trained natural language processing model by fine-tuning the BERT-base model using credit data corpus. The credit data corpus can be corpus from the political and legal field.

[0057] Specifically, after obtaining the text data from the unstructured credit data of various enterprises, this text data can be input into a pre-trained BERT-base model to extract entity information and semantic relationships as unstructured credit features. For example, the text data can extract the dishonest entity, dissemination channel, sentiment tendency, and semantic relationships from the text data containing false information. As another example, feature vectors such as "false information type = XX" and "channel = WeChat group" can be extracted from the text data "A resident of a certain community spreads false information about XX in a WeChat group" as unstructured credit features.

[0058] In this embodiment, the visual object detection model can be a single-stage object detection algorithm (You OnlyLook Once version 8, YOLOv8). The YOLOv8 model can combine optical character recognition (OCR) to identify text information in the scene, or the YOLOv8 model can independently identify the behavioral information of the subject in the scene.

[0059] After obtaining the visual data from the unstructured credit data of various enterprises, this visual data can be input into a pre-trained YOLOv8 model for visual recognition. Combined with OCR recognition, behavioral and scene text information within the visual data can be obtained, serving as the enterprise's unstructured credit features. For example, the YOLOv8 model can identify dishonest behaviors (illegal parking, garbage dumping, etc.) in surveillance videos, and then OCR can be used to recognize scene text (house numbers, etc.) within the surveillance videos, thereby generating spatial credit tags, i.e., unstructured credit features.

[0060] Currently, most corporate credit evaluation methods lack the ability to semantically understand and extract features from unstructured data. For example, they may not be able to automatically extract key features such as "the dishonest entity, the channel of dissemination, and the scope of influence" from false information such as "a resident of a certain community posted a fake XX video on a video platform." They may also be unable to associate illegal parking behavior in surveillance videos with grid address data, resulting in credit evaluation models that can only be generated based on basic indicators (such as the number of administrative penalties) and are unable to reflect multidimensional credit behavior.

[0061] To address the aforementioned issues, this application employs the BERT and YOLOv8 models to extract textual and visual features, constructs a cross-modal association graph, and forms a multi-modal credit data fusion analysis engine. While ensuring reliable multi-dimensional evaluation of enterprises, this engine improves the accuracy of unstructured credit feature extraction to 92%, a 40% improvement compared to traditional keyword matching.

[0062] In the embodiments of this application, after extracting behavioral information, subject information, and spatiotemporal information from the heterogeneous credit features of various enterprises, the adaptive integration method for heterogeneous credit data further includes: if the target information contains heterogeneous credit features of different modalities, then using a graph attention network to determine the weights of each heterogeneous credit feature of different modalities, using a pre-trained adversarial learning network to transform the heterogeneous credit features of different modalities into the same modality, and under the same modality, performing feature fusion on the heterogeneous credit features in the target information based on the weights of each heterogeneous credit feature; wherein, the target information is behavioral information, subject information, or spatiotemporal information.

[0063] If the target information contains only one type of heterogeneous credit features, then no processing is performed.

[0064] In this embodiment, when the target information includes both textual and visual features, it can be determined that the target information contains heterogeneous credit features of different modalities. In this case, a graph attention network can be used to determine the weights of each heterogeneous credit feature of different modalities (e.g., the weight of textual features can be 40%, and the weight of visual features can be 30%). Then, a pre-trained adversarial learning network is used to align the heterogeneous credit features of different modalities, transforming them into the same modality and reducing the distribution differences of heterogeneous credit features across different modules. Finally, feature fusion is performed based on the weights of each heterogeneous credit feature to obtain a comprehensive credit evaluation vector, which facilitates enterprise rating and evaluation.

[0065] After obtaining the comprehensive credit rating vector, this comprehensive rating vector can be input into the eXtreme Gradient Boosting (XGBoost) model to score the credit rating model built based on graph attention network and adversarial learning network. The output value of the credit rating model is improved from 0.68 to 0.92, and the accuracy of identifying new types of dishonest behavior (such as online false information) is improved by 35%, further improving the accuracy of judging dishonesty / trustworthiness.

[0066] The embodiments of this application achieve alignment compensation for multimodal heterogeneous credit features through graph attention networks and adversarial learning networks, thereby improving the quality of credit evaluation data and further enhancing the consistency between evaluation results and actual credit levels.

[0067] This application's embodiments, by integrating natural language processing and graph computing technologies, overcome the limitations of existing solutions in processing unstructured data, enabling semantic understanding and multi-dimensional correlation analysis of data such as false information text and video surveillance, thereby improving the accuracy of credit rating models by more than 35%.

[0068] Currently, most credit evaluation data collection processes use the traditional Extract-Transform-Load (ETL) process, which is difficult to dynamically respond to real-time changes in data standards across government departments and cannot cope with sudden data demands from government department data interface upgrades or new credit scenarios (such as cross-regional key personnel management), resulting in low integration efficiency and high maintenance costs.

[0069] To address the aforementioned issues, this application's embodiments construct a closed-loop mechanism of dynamic perception, rule generation, and real-time mapping. This mechanism solves the problem that current ETL technologies cannot dynamically respond to changes in cross-departmental data standards, enabling real-time adaptation and efficient integration of credit data. It reduces manual adaptation costs by more than 90% and improves the efficiency of cross-departmental data flow.

[0070] Specifically, in the embodiments of this application, the method for adapting and integrating heterogeneous credit data further includes: responding to a data change event triggered by a designated credit regulatory body, extracting field change features corresponding to the data change event, using the field change features and a pre-trained graph neural network to determine data transformation rules in the credit knowledge graph, and updating the credit knowledge graph based on the data transformation rules.

[0071] In the embodiments of this application, the designated credit supervision entity can be any one of multiple government departments.

[0072] Generally speaking, the credit data monitored by credit regulatory bodies may be updated as business needs require. Data change events can indicate updated format specifications, such as adding new credit evaluation indicators or converting field types.

[0073] In this embodiment of the application, after obtaining the credit knowledge graph, a pre-trained graph neural network can be used to extract rules from the credit knowledge graph to obtain various data transformation rules of the credit knowledge graph.

[0074] In this embodiment of the application, when a data change event triggered by a government department is detected, firstly, the field change features corresponding to the data change event can be extracted through semantic parsing, such as field semantics or format specifications. Then, the field change features and a pre-trained graph neural network can be combined to determine the data transformation rules in the credit knowledge graph corresponding to the data change event. Finally, the data in the data change event is transformed using these data transformation rules, and the resulting new data transformation rules are updated in the credit knowledge graph.

[0075] When a data change event is detected, a pre-trained graph neural network can be used to match data transformation rules similar to the data change event and perform data transformation. For example, the new field "false advertising platform" is automatically associated with the "dissemination channel" entity in the credit knowledge graph, generating text classification mapping rules. As another example, when the format of the "enterprise credit record" field in the market supervision department changes from text to enumeration, a new format transformation rule can be automatically generated based on the "credit misconduct - penalty type" association in the credit knowledge graph, and then the new format transformation rule can be updated in the credit knowledge graph.

[0076] Through testing, the data conversion rule generation efficiency of this application embodiment is improved by 80%, the amount of manual intervention is reduced from 2 hours per rule to 10 minutes, and the rule accuracy rate reaches over 95%.

[0077] This application embodiment can also monitor whether data collection is being performed through a Kafka message queue. After data collection, the collected data can be format-converted based on data conversion rules (e.g., field cleaning or format conversion) in the credit knowledge graph. For example, when the "public service incentive record" data from the Development and Reform Commission flows through, the "incentive type-credit score" mapping rule generated by the credit knowledge graph is automatically invoked to complete the data standardization conversion.

[0078] The embodiments of this application can monitor the mapping success rate and abnormal data in real time through ELK Stack, which can shorten the data integration cycle from 72 hours in the traditional solution to 4 hours and reduce the data consistency error rate to below 0.05%.

[0079] Existing ETL technologies rely on static mapping rules, which cannot detect changes in data standards (such as dynamic adjustments to field types and semantic definitions) from government departments like law enforcement and market regulators in real time. This results in the need for manual reconfiguration of the conversion process when new credit data (such as records of dishonesty under XX control) is integrated, leading to low adaptation efficiency and potential data consistency issues. For example, when a market regulator adds a new indicator of "false advertising on e-commerce platforms," ​​traditional solutions require manual parsing of the new indicator's semantics and writing mapping rules, which can take weeks and may result in data integration errors due to oversights in rule writing.

[0080] The aforementioned issues, as the core bottlenecks of existing heterogeneous data integration technologies, directly lead to the need for manual redevelopment of adaptation rules when government departments upgrade their data interfaces or access data from new scenarios. This results in extended integration cycles, soaring maintenance costs, and seriously hinders real-time collaboration in cross-departmental credit governance.

[0081] To address the aforementioned issues, this application's embodiments utilize smart contracts to monitor data change events in real time and automatically generate data transformation rules based on a credit knowledge graph. This enables dynamic adaptation of the entire data collection, cleaning, and storage process, reducing the need for manual intervention by over 90% and shortening the adaptation cycle from weekly to minute-level.

[0082] Currently, there is no standardized conversion protocol for cross-regional credit data. The definitions of credit indicators (such as point calculation rules and penalty thresholds) vary significantly between different cities, and there is a lack of dynamic mapping mechanisms. When cross-regional business (such as cross-city tourism or cross-city employment) requires credit evaluation, manual coordination of the rules in the two places is necessary, which is time-consuming, labor-intensive, and prone to disputes.

[0083] The aforementioned issues lead to the formation of "credit silos," hindering cross-regional credit governance collaboration and preventing the mutual recognition of evaluation results for the same credit entity in different regions, thus increasing social governance costs. For example, the "grid credit score" in City A and the "community credit score" in City B cannot be directly interoperated due to differences in rules, limiting the implementation of cross-regional joint punishment and incentive measures.

[0084] To address the aforementioned issues, this application embodiment establishes a cross-domain conversion mechanism to break down the heterogeneous barriers between regional credit evaluation systems, enabling standardized mutual recognition of credit scores and penalty standards among multiple cities, thereby improving mutual recognition efficiency by over 80%.

[0085] Specifically, in the embodiments of this application, the method for adapting and integrating heterogeneous credit data further includes: converting the multidimensional associated credit profiles of various enterprises in multiple designated regions based on a predetermined credit data standard, so as to conduct credit evaluation of various enterprises in multiple designated regions based on the converted multidimensional associated credit profiles.

[0086] In this embodiment of the application, the conversion of multidimensional related credit profiles of various enterprises in multiple designated regions based on a predetermined credit data standard may include at least one of the following: Based on predetermined credit data standards, fields in the multidimensional related credit profiles of various enterprises in multiple designated regions are transformed.

[0087] Based on predetermined credit data standards, the credit evaluation standards in the multidimensional related credit files of various enterprises in multiple designated regions are converted.

[0088] In the embodiments of this application, the predetermined credit data standard can be a standardized metadata dictionary formed based on pre-determined credit metadata standards for multiple designated regions. This can include basic information (such as name, ID number), behavioral indicators (such as number of defaults, incentive records), and evaluation rules (such as point ranges, penalty thresholds). For example, the indicators of "grid credit score" in region A and "community credit score" in region B can be mapped to a unified standardized metadata dictionary of "social governance credit dimension." By constructing a standardized metadata dictionary, the metadata ambiguity rate can be reduced from 25% in traditional schemes to below 5%, laying the foundation for cross-regional mutual recognition.

[0089] This application embodiment can first convert the multidimensional related credit profiles of enterprises in various designated regions based on a predetermined credit data standard. Then, it can perform credit evaluation on each enterprise based on the converted multidimensional related credit profiles to ensure that the evaluation standard of an individual enterprise in different designated regions is consistent with the evaluation standard of other enterprises in that region.

[0090] In this embodiment, historical credit score data for each city can be collected in advance. A percentile mapping relationship can be calculated using an empirical distribution function, and standardization of different scoring systems can be achieved through quantile alignment. This embodiment can utilize the percentage mapping relationship to convert the credit evaluation standards in the multidimensional associated credit files of various enterprises in multiple specified regions. For example, the top 10% of credit scores in City A can be mapped to the top 15% of scores in City B, generating a dynamic conversion matrix. When querying across regions, this embodiment can automatically map the source city score to the target city standard; for example, a user's score of 85 (top 15%) in City B can be converted to a score of 90 (top 10%) in City A.

[0091] This application's embodiments can achieve dynamic conversion and mutual recognition verification of credit scores across different regions. For example, when a user in City A applies for public services in City B, their credit score can be automatically mapped to the City B standard according to its percentile, without manual intervention, improving mutual recognition efficiency by more than 80%. The cross-regional score conversion error can be controlled within ±3 points, and the mutual recognition efficiency is reduced from 2 days for manual processing to real-time response.

[0092] Existing encryption technologies struggle to balance data sharing and privacy protection needs in cross-departmental data transfers, especially when dealing with sensitive personal information (such as credit data of key personnel), which poses compliance risks and indicates an imperfect data security and privacy protection mechanism.

[0093] To address the aforementioned issues, this application embodiment reduces data interaction risks through a consortium blockchain consensus mechanism.

[0094] In the embodiments of this application, before converting the multidimensional associated credit profiles of various enterprises in multiple designated regions based on a predetermined credit data standard, the method for adapting and integrating heterogeneous credit data further includes: deploying a credit mutual recognition smart contract on the consortium blockchain corresponding to the multiple designated regions, so as to use the credit mutual recognition smart contract to execute the step of converting the multidimensional associated credit profiles of various enterprises in multiple designated regions based on a predetermined credit data standard.

[0095] In this embodiment, multiple designated regions utilize a consortium blockchain to store predetermined credit data standards and record the consensus process of mapping rules based on the consortium blockchain, ensuring the traceability of the conversion process. Simultaneously, credit mutual recognition smart contracts can be deployed on the consortium blockchains corresponding to the multiple designated regions. These smart contracts include three functional modules: data query, mapping conversion, and result verification, to automate the execution of credit data query, conversion, and verification.

[0096] For example, when a user in City A applies for public services in City B, the contract automatically retrieves credit data from both cities, converts the scoring results through a percentile mapping algorithm, and triggers the local credit incentive rules in City B without human intervention.

[0097] For example, when a user in City A applies for public services in City B, the credit databases of both cities can be accessed to obtain the source score; a percentile mapping algorithm can be automatically executed to generate the target score; the legality of the conversion can be verified through the consensus node of the consortium blockchain, and the result can be returned to the service system of City B to trigger the incentive rules.

[0098] The embodiments of this application can reduce the cross-regional business processing time from the traditional 3-5 working days to 10 minutes, with a process automation rate of 100%.

[0099] This application's embodiment employs a three-layer technical architecture of multimodal fusion, dynamic adaptation, and cross-domain mutual recognition to solve the challenges of integrating and governing cross-departmental and cross-regional credit data. Specifically, it comprises three layers: a data perception layer, a processing and fusion layer, and an application mutual recognition layer. The data perception layer primarily executes the process of dynamically capturing changes in multi-source data standards through a credit knowledge graph and generating adaptive data mapping rules. The processing and fusion layer primarily executes the process of feature extraction and association analysis of multimodal data using natural language processing and graph neural networks. The application mutual recognition layer primarily executes the standardized conversion of cross-regional credit data based on consortium blockchains and percentile mapping algorithms. This three-layer technical architecture utilizes a microservice component design, with seamless integration between layers via message queues and a data platform, ensuring efficient data flow and scalability.

[0100] This application's embodiments achieve intelligent processing of the entire process from data access to application implementation through the collaboration of a three-layer technical architecture, providing a reusable technical paradigm for cross-departmental and cross-regional social credit governance.

[0101] Corresponding to the heterogeneous credit data adaptation and integration method in the above embodiments, Figure 2 This is a structural block diagram of an apparatus for adapting and integrating heterogeneous credit data according to an embodiment of this application. For ease of explanation, only the parts relevant to the embodiment of this application are shown. References Figure 2 The heterogeneous credit data adaptation and integration device 20 includes: a feature acquisition module 201, an information extraction module 202, and a credit association module 203.

[0102] The feature acquisition module 201 is used to acquire heterogeneous credit features of various enterprises in a specified region. The heterogeneous credit features include the structured credit features corresponding to the structured credit data of various enterprises, as well as the unstructured credit features in the unstructured credit data of various enterprises extracted based on a pre-trained natural language processing model and / or a pre-trained visual object detection model. Information extraction module 202 is used to extract behavioral information, subject information and spatiotemporal information from the heterogeneous credit characteristics of various enterprises; The credit association module 203 is used to integrate the credit knowledge graphs of each enterprise based on the relationship between the behavioral information, subject information and spatiotemporal information of each enterprise, with people, organizations and events as the core entities, so as to use the credit knowledge graphs of each enterprise as multi-dimensional associated credit files of each enterprise.

[0103] In one embodiment of this application, the unstructured credit data of each enterprise includes text data and visual data; the device 20 also includes a feature extraction module; The feature extraction module is used to extract entity information and semantic relationships from text data based on a pre-trained natural language processing model to obtain unstructured credit features before obtaining heterogeneous credit features of various enterprises in a specified area; and / or to extract behavioral information and scene text information from visual data based on a pre-trained visual object detection model to obtain unstructured credit features.

[0104] In one embodiment of this application, the device 20 further includes a feature fusion module; The feature fusion module is used to extract behavioral information, subject information, and spatiotemporal information from the heterogeneous credit features of various enterprises. If the target information contains heterogeneous credit features of different modalities, a graph attention network is used to determine the weights of each heterogeneous credit feature of different modalities. A pre-trained adversarial learning network is used to transform the heterogeneous credit features of different modalities into the same modality. Under the same modality, the heterogeneous credit features in the target information are fused based on the weights of each heterogeneous credit feature. The target information can be behavioral information, subject information, or spatiotemporal information.

[0105] In one embodiment of this application, the device 20 further includes a map update module; The graph update module is used to respond to data change events triggered by a designated credit regulatory entity, extract field change features corresponding to the data change events, use the field change features and a pre-trained graph neural network to determine the data transformation rules in the credit knowledge graph, and update the credit knowledge graph based on the data transformation rules.

[0106] In one embodiment of this application, the device 20 further includes a credit rating module; The credit rating module is used to convert the multidimensional related credit profiles of various enterprises in multiple designated regions based on predetermined credit data standards, so as to conduct credit rating of various enterprises in multiple designated regions based on the converted multidimensional related credit profiles.

[0107] In one embodiment of this application, the credit rating module is specifically used to perform at least one of the following: Based on predetermined credit data standards, the fields in the multidimensional related credit profiles of various enterprises in multiple designated regions are transformed. Based on predetermined credit data standards, the credit evaluation standards in the multidimensional related credit files of various enterprises in multiple designated regions are converted.

[0108] In one embodiment of this application, the device 20 further includes: The credit mapping module is used to deploy credit mutual recognition smart contracts on the consortium blockchains corresponding to multiple designated regions before converting the multidimensional related credit profiles of various enterprises in multiple designated regions based on predetermined credit data standards. The credit mutual recognition smart contracts are used to execute the steps of converting the multidimensional related credit profiles of various enterprises in multiple designated regions based on predetermined credit data standards.

[0109] See Figure 3 , Figure 3 This is a schematic block diagram of an electronic device provided according to an embodiment of this application. Figure 3 The electronic device 300 in this embodiment may include one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memories 304 store computer programs, including program instructions. The processors 301 execute the program instructions stored in the memories 304. Specifically, the processors 301 are configured to invoke the program instructions to perform the functions of the modules in the aforementioned device embodiments, for example... Figure 2 The functions of the feature acquisition module 201, information extraction module 202, and credit association module 203 are shown.

[0110] It should be understood that, in the embodiments of this application, the processor 301 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0111] Input device 302 may include a touchpad, a fingerprint sensor (for collecting the user's fingerprint information and fingerprint orientation information), a microphone, etc., and output device 303 may include a display (LCD, etc.), a speaker, etc.

[0112] The memory 304 may include read-only memory and random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include non-volatile random access memory.

[0113] In specific implementations, the processor 301, input device 302, and output device 303 described in the embodiments of this application can execute the implementation method described in the heterogeneous credit data adaptation and integration method provided in the embodiments of this application, or they can execute the implementation method of the electronic device described in the embodiments of this application, which will not be repeated here.

[0114] In another embodiment of this application, a computer-readable storage medium is provided. This computer-readable storage medium stores a computer program, which includes program instructions. When executed by a processor, the program instructions implement all or part of the processes in the methods described above. Alternatively, the computer program can instruct related hardware to complete the process. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0115] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the foregoing embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the electronic device. Furthermore, the computer-readable storage medium can include both internal and external storage units of the electronic device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0116] This application provides a computer program product, which includes computer-executable instructions or a computer program. The computer-executable instructions or the computer program are stored in a computer-readable storage medium. The processor of an electronic device reads the computer-executable instructions from the computer-readable storage medium and executes the computer-executable instructions, causing the electronic device to perform the heterogeneous credit data adaptation and integration method described in this application.

[0117] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0118] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the electronic devices and units described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0119] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. Furthermore, the mutual couplings or direct couplings or communication connections shown or discussed can be indirect couplings or communication connections through some interfaces or units, or they can be electrical, mechanical, or other forms of connection.

[0120] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of this application, depending on actual needs.

[0121] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for adapting and integrating heterogeneous credit data, characterized in that, include: Obtain heterogeneous credit characteristics of various enterprises in a specified region; The heterogeneous credit features include the structured credit features corresponding to the structured credit data of each enterprise, and the unstructured credit features in the unstructured credit data of each enterprise extracted based on a pre-trained natural language processing model and / or a pre-trained visual object detection model. Extract behavioral information, subject information, and spatiotemporal information from the heterogeneous credit characteristics of each enterprise; Using people, organizations, and events as core entities, and integrating the relationships between behavioral information, subject information, and spatiotemporal information of each enterprise, a credit knowledge graph corresponding to each enterprise is obtained, which serves as a multidimensional associated credit file for each enterprise.

2. The method for adapting and integrating heterogeneous credit data as described in claim 1, characterized in that, The unstructured credit data of each enterprise includes text data and visual data; Before obtaining the heterogeneous credit characteristics of various enterprises in a specified region, the method further includes: The entity information and semantic relationships in the text data are extracted based on the pre-trained natural language processing model to obtain the unstructured credit features; and / or The unstructured credit features are obtained by extracting behavioral information and scene text information from the visual data based on a pre-trained visual object detection model.

3. The method for adapting and integrating heterogeneous credit data as described in claim 1, characterized in that, After extracting behavioral information, subject information, and spatiotemporal information from the heterogeneous credit characteristics of each enterprise, the method further includes: If the target information contains heterogeneous credit features of different modalities, then the weights of each heterogeneous credit feature of different modalities are determined by a graph attention network, and the heterogeneous credit features of different modalities are transformed into the same modality by a pre-trained adversarial learning network. Under the same modality, the heterogeneous credit features in the target information are fused based on the weights of each heterogeneous credit feature. The target information can be behavioral information, subject information, or spatiotemporal information.

4. The method for adapting and integrating heterogeneous credit data as described in any one of claims 1 to 3, characterized in that, The method further includes: In response to a data change event triggered by a designated credit regulatory body, the system extracts the field change features corresponding to the data change event, uses the field change features and a pre-trained graph neural network to determine the data transformation rules in the credit knowledge graph, and updates the credit knowledge graph based on the data transformation rules.

5. The method for adapting and integrating heterogeneous credit data as described in any one of claims 1 to 3, characterized in that, The method further includes: Based on predetermined credit data standards, the multidimensional related credit profiles of various enterprises in multiple designated regions are transformed, and the credit evaluation of various enterprises in multiple designated regions is carried out based on the transformed multidimensional related credit profiles.

6. The method for adapting and integrating heterogeneous credit data as described in claim 5, characterized in that, The conversion of multidimensional related credit profiles of various enterprises in multiple designated regions based on predetermined credit data standards includes at least one of the following: Based on predetermined credit data standards, fields in the multidimensional related credit profiles of various enterprises in multiple designated regions are transformed. Based on predetermined credit data standards, the credit evaluation standards in the multidimensional related credit files of various enterprises in multiple designated regions are converted.

7. The method for adapting and integrating heterogeneous credit data as described in claim 5, characterized in that, Before converting the multidimensional linked credit profiles of various enterprises in multiple designated regions based on predetermined credit data standards, the method further includes: Deploy credit mutual recognition smart contracts on the consortium blockchains corresponding to the multiple designated regions, and use the credit mutual recognition smart contracts to perform the steps of converting the multi-dimensional associated credit profiles of various enterprises in the multiple designated regions based on predetermined credit data standards.

8. A device for adapting and integrating heterogeneous credit data, characterized in that, include: The feature acquisition module is used to acquire heterogeneous credit features of various enterprises in a specified region. The heterogeneous credit features include the structured credit features corresponding to the structured credit data of each enterprise, and the unstructured credit features in the unstructured credit data of each enterprise extracted based on a pre-trained natural language processing model and / or a pre-trained visual object detection model. The information extraction module is used to extract behavioral information, subject information, and spatiotemporal information from the heterogeneous credit characteristics of the various enterprises; The credit association module is used to integrate the credit knowledge graphs of each enterprise based on the relationship between the behavioral information, subject information and spatiotemporal information of each enterprise, with people, organizations and events as core entities, so as to use the credit knowledge graphs of each enterprise as multi-dimensional associated credit files of each enterprise.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images