Business approval method, device and equipment and storage medium

Abstract

The invention provides a service approval method and device, equipment and a storage medium, and relates to the technical field of service management. According to the specific technical scheme, the service type of a to-be-approved service and service associated data of an application object associated with the to-be-approved service are obtained, wherein the service associated data comprise attribute information of the application object, resource condition information of the application object and behavior information of the application object; and determining at least one target data fusion method from preset data fusion methods based on the service type and a preset service type and data fusion method mapping rule. And processing the service associated data through each target data fusion method to obtain a data fusion result corresponding to each target data fusion method. And performing service approval on the to-be-approved service based on the service type and each data fusion result to obtain a service approval result of the to-be-approved service. And the business approval efficiency and accuracy are improved.

Description

Technical Field

[0001] This application relates to the field of business management technology, and in particular to a business approval method, apparatus, device, and storage medium. Background Technology

[0002] Business approval is applied to the processing of various business access and resource allocation requests. Building a sound business approval mechanism can improve the standardization and security of business operations.

[0003] Currently, business approvals are typically based on predetermined rules or analytical models. Specifically, this business approval mechanism determines whether to approve or reject requests based on statically set threshold conditions, blacklists / whitelists, or classification models.

[0004] However, the aforementioned methods based on predetermined rules or fixed models suffer from pattern solidification. When faced with new patterns and features that continuously emerge in business scenarios, the mechanism's generalization ability is insufficient, leading to a decrease in its discrimination accuracy and making it difficult to guarantee the accuracy and efficiency of business approval. Summary of the Invention

[0005] This application provides a business approval method, apparatus, device, and storage medium. By employing multiple data fusion algorithms to process multi-source data, different data fusion results are generated to support the final approval decision. This addresses the problems of insufficient generalization ability, low discrimination accuracy, and low decision-making efficiency caused by the rigidity of related business approval methods.

[0006] To achieve the above objectives, the embodiments of this application adopt the following technical solutions: Firstly, a business approval method is provided, which includes: First, the business type of the pending approval business and the associated business-related data of the applicants are obtained. This associated data includes the applicants' attribute information, resource status information, and behavioral information. Second, based on the business type and preset mapping rules between business types and data fusion methods, at least one target data fusion method is determined from the preset data fusion methods. These preset data fusion methods include: business rule-based data fusion methods, statistical method-based data fusion methods, graph database-based data fusion methods, and machine learning algorithm-based data fusion methods. Then, the business-related data is processed using each target data fusion method to obtain the corresponding data fusion result. Finally, based on the business type and each data fusion result, the pending approval business is approved to obtain the approval result.

[0007] The business approval method provided in this application establishes a matching mechanism between business types and data fusion methods, thereby achieving intelligent approval processes and improving the accuracy and efficiency of business approvals. Specifically, this solution first establishes a comprehensive decision-making information foundation by acquiring business types and multi-dimensional business-related data. Subsequently, based on preset mapping rules, differentiated data fusion method combinations are determined for different business types, ensuring the adaptability of data processing to business characteristics. Furthermore, by executing multiple fusion algorithms, the respective advantages of rule engines, statistical analysis, graph computing, and machine learning methods can be fully utilized. Finally, the multi-source fusion results are comprehensively evaluated based on the business type to form the business approval result.

[0008] Compared to the static rules or fixed models used in related mechanisms, the dynamic combination mechanism of the data fusion method in this solution effectively improves the generalization ability to cope with different business scenarios. Cross-validation and in-depth mining of multi-dimensional data can improve the accuracy of business approval decisions, while the structured processing flow optimizes the overall approval efficiency while ensuring the quality of decisions.

[0009] In one possible implementation of the first aspect, each target data fusion method processes business-related data to obtain a data fusion result corresponding to each target data fusion method. This includes: sequentially executing the target data fusion methods to process the business-related data according to the data fusion order among the target data fusion methods, thereby obtaining a data fusion result corresponding to each target data fusion method. The input data of a target data fusion method that is later in the data fusion order includes at least the data fusion result output by its preceding target data fusion method.

[0010] Target data fusion methods include graph database-based data fusion methods and / or machine learning algorithm-based data fusion methods.

[0011] The graph database-based data fusion method includes the following steps: Based on business-related data, identify interaction objects that have resource interaction relationships with the applicant. Construct a resource interaction relationship graph using the applicant and interaction objects as nodes. Analyze the topological structure of the applicant in the resource interaction relationship graph to obtain the resource exchange relationship characteristic indicators of the applicant, which serve as the data fusion result of the graph database-based data fusion method. The resource exchange relationship characteristic indicators characterize the stability and influence range of the applicant's resource interaction network.

[0012] The data fusion method based on machine learning algorithms includes the following steps: Generating a feature vector based on at least one of attribute information, resource status information, and behavioral information, and / or the data fusion result obtained through other data fusion methods. Processing the business type and feature vector using a preset business approval model yields a business security score for the business to be approved, which serves as the data fusion result of the machine learning-based data fusion method. The business type is used to instruct the preset business approval model to adjust the weight allocation for security assessment.

[0013] It should be understood that by establishing a sequential execution mechanism for data fusion methods, a hierarchical feature extraction and decision support process is constructed. This sequential processing approach allows subsequent methods to conduct further analysis based on the in-depth results of preceding methods, resulting in progressive information refinement. In particular, by mining network topology features using graph database methods and then combining them with machine learning methods for comprehensive scoring and prediction, the depth of feature mining is ensured, and the intelligence level of the final decision-making is improved, effectively enhancing the ability to process complex business data.

[0014] In another possible implementation of the first aspect, business approval is performed on the business to be approved based on the business type and each data fusion result to obtain the business approval result for the business to be approved. This includes: performing conflict detection on each data fusion result based on preset result conflict detection rules to obtain conflict detection results. The result conflict detection rules include contradictory combinations between different data fusion results. If at least one data fusion result combination is a contradictory combination in the conflict detection results, a conflict report is generated. A conflict confirmation result is obtained in response to manual input based on the conflict report to determine the valid data fusion result. The valid data fusion result includes data fusion results that did not trigger conflict rules and data fusion results that are confirmed as valid and conflicting in the conflict confirmation results. Based on the business type and the valid data fusion results, the business approval result for the business to be approved is obtained.

[0015] It should be understood that a verification and coordination system for the results of multi-method fusion has been established through a result conflict detection mechanism and a manual confirmation process. This solution effectively solves the problem of inconsistencies that may occur in the fusion of multi-source data. Through human-machine collaborative decision-making, it ensures reliability in complex business scenarios while maintaining the efficiency and standardization of the approval process.

[0016] In another possible implementation of the first aspect, the business approval result for the pending business is obtained based on the business type and the effective data fusion results. This includes: determining the weight value corresponding to each effective data fusion result based on the business type and a preset mapping rule set between business types and fusion result weights; weighting and summing the weight values ​​of each effective data fusion result to obtain the business evaluation score for the pending business; and determining the business approval result as passed if the business evaluation score is greater than a preset evaluation threshold.

[0017] It should be understood that a precise decision-making fusion is achieved through a business-type adaptive weight allocation mechanism. This solution dynamically adjusts the contribution of each data fusion result to the final decision based on the characteristics of different business types, ensuring a precise match between approval standards and business needs. This differentiated weight allocation strategy not only reflects adaptability to different business scenarios but also guarantees the objectivity and consistency of the decision-making process through quantitative weighted summation.

[0018] In another possible implementation of the first aspect, the method further includes: determining the business approval result as unsuccessful if the business evaluation score is less than or equal to a preset evaluation threshold; determining the weighted evaluation value of each valid fusion result based on its corresponding weight value if the business approval result is unsuccessful; comparing the weighted evaluation value of each valid fusion result with its corresponding preset single-item evaluation threshold to determine the substandard data fusion results whose weighted evaluation values ​​are less than the corresponding preset evaluation threshold; and generating a business approval result report based on the substandard data fusion results and related business data.

[0019] It should be understood that by establishing a mechanism for identifying and generating reports on non-compliance results, a transparent approval service is provided. This solution not only provides approval conclusions but also accurately identifies the key factors leading to approval failures, offering applicants clear directions for improvement. This refined results analysis enhances the transparency of the approval process and increases the added value of the service through targeted improvement suggestions, achieving a functional upgrade from simple approval to intelligent service.

[0020] In another possible implementation of the first aspect, the attribute information of the applicant includes: basic information of the applicant and credibility information of the applicant. The resource status information of the applicant includes: resource inventory information, available resource information, resource input information, and resource output information of the applicant. The behavioral information of the applicant includes: resource interaction event records, business request records, and business browsing records of the applicant.

[0021] It should be understood that by dividing the information of applicants into three dimensions—attributes, operations, and behaviors—a structured multi-source data system has been established. This solution addresses the problems of single data dimensions and insufficient information utilization in traditional approval processes, providing a complete data foundation for subsequent multi-method fusion analysis and ensuring the accuracy and comprehensiveness of approval decisions from a data perspective.

[0022] In another possible implementation of the first aspect, the method further includes: acquiring approval efficiency data for each of the applicant's multiple evaluated business processes. The approval efficiency data for each evaluated business process includes: approval time, number of manual conflict confirmations, and deviation of business evaluation results. The deviation of business evaluation results measures the degree of difference between the business evaluation results of the evaluated business and the applicant's actual business behavior. Pattern summarization is performed on the approval efficiency data for each of the multiple evaluated business processes to determine the applicant's approval strategy pattern. Based on the approval strategy pattern, a personalized data fusion strategy is generated for the applicant. The personalized data fusion strategy defines the execution priority and combination of various data fusion methods in subsequent evaluation processes for the applicant.

[0023] It should be understood that by summarizing patterns based on historical approval efficiency data, personalized data fusion strategies are generated for specific applicants. This solution realizes a shift in approval strategies from a "one-size-fits-all" approach to precise customization. By dynamically optimizing the combination of methods and execution priorities, it significantly improves the processing efficiency of subsequent business while ensuring approval quality, and enhances the self-learning and continuous optimization capabilities of the approval mechanism for this application.

[0024] Secondly, a business approval device is provided, the device comprising: The acquisition module is used to acquire the business type of the business to be approved and the business association data of the application object associated with the business to be approved; the business association data of the application object includes: the attribute information of the application object, the resource status information of the application object, and the behavior information of the application object; The processing module is used to determine at least one target data fusion method from a preset set of data fusion methods based on the business type and preset business type and data fusion method mapping rules; the preset data fusion methods include: a data fusion method based on business rules, a data fusion method based on statistical methods, a data fusion method based on graph databases, and a data fusion method based on machine learning algorithms; process the business-related data through each of the target data fusion methods to obtain the data fusion result corresponding to each target data fusion method; and perform business approval on the business to be approved based on the business type and each of the data fusion results to obtain the business approval result of the business to be approved.

[0025] Thirdly, a business approval device is provided, the method comprising: a memory and at least one processor. The memory is communicatively connected to the processor. The memory is used to store computer program code, the computer program code including computer instructions. When the processor executes the computer instructions, it causes the business approval device to perform the method as described in the first aspect and any possible implementation thereof.

[0026] Fourthly, a computer-readable storage medium is provided that stores computer instructions. When executed by a processor, the computer instructions are used to implement the method as described in the first aspect and any possible implementation thereof.

[0027] Fifthly, a computer program product is provided that, when running on a computer or executed by a computer's processor, implements the method described in the first aspect and any possible design thereof. The computer may be the business approval device described in the third aspect and any possible implementation thereof.

[0028] It is understood that the beneficial effects that the business approval device described in the second aspect, the business approval equipment described in the third aspect, the computer-readable storage medium described in the fourth aspect, and the computer program product described in the fifth aspect can achieve can be referred to the beneficial effects in the first aspect and any possible implementation thereof, which will not be repeated here. Attached Figure Description

[0029] Figure 1 This application provides a schematic diagram of a data processing flow for a computing device. Figure 2 A flowchart illustrating a business approval method provided in an embodiment of this application; Figure 3 A flowchart illustrating a data fusion method based on a graph database provided in an embodiment of this application; Figure 4 A flowchart illustrating a data fusion method based on a machine learning algorithm provided in an embodiment of this application; Figure 5 A flowchart illustrating another business approval method provided in this application embodiment; Figure 6 A flowchart illustrating a method for determining business approval results provided in an embodiment of this application; Figure 7 A flowchart illustrating another method for determining business approval results provided in this application embodiment; Figure 8 A flowchart illustrating a personalized data fusion strategy generation method provided in this application embodiment; Figure 9 This is a schematic diagram of the structure of a business approval device provided in an embodiment of this application; Figure 10 This is a schematic diagram of the structure of a business approval device provided in an embodiment of this application. Detailed Implementation

[0030] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this embodiment, unless otherwise stated, "a plurality of" means two or more.

[0031] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0032] The technical solutions provided in this application, including the collection, storage, use, processing, transmission, provision, and disclosure of financial data or user data, comply with relevant laws and regulations and do not violate public order and good morals.

[0033] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, they do not mean that the applicant has used or necessarily used the solution.

[0034] The relevant business approval mechanism has significant technical bottlenecks when dealing with complex and ever-changing business scenarios. Specifically, the mechanism is usually based on a predefined rule system or a fixed decision-making model. This rigid approval model, when faced with new business models or unconventional business requests, suffers from insufficient data feature extraction and a single decision-making dimension, making it difficult to effectively mine deep correlation information in business data. This results in insufficient ability to identify complex business models, and the accuracy and reliability of approval results cannot be guaranteed.

[0035] Therefore, how to improve the accuracy of business approval mechanisms when facing complex business scenarios is an urgent problem to be solved.

[0036] To address the aforementioned technical challenges, this application proposes an intelligent approval scheme based on multi-method collaboration. The core idea of ​​this scheme is to optimize the approval process by establishing an intelligent matching mechanism between business characteristics and processing methods. Specifically, this includes: automatically selecting appropriate combinations of analysis methods based on the business type; improving decision-making quality through collaborative processing of multi-dimensional data; and employing an intelligent result integration mechanism to ensure the reliability of the final approval conclusion. This scheme effectively balances approval quality and processing efficiency while enhancing adaptability.

[0037] The business approval method provided in this application can be applied to computing devices. Specifically, the computing device can be a single server or a server cluster composed of multiple servers, or a computer, or a processor or processing chip in a server or computer, etc. This application does not limit the specific device form of the computing device.

[0038] Meanwhile, the business approval method provided in this application embodiment can be applied to different business scenarios.

[0039] For example, in the financial services sector, the types of business pending approval may include: credit line approval, insurance product underwriting, and cross-border transaction compliance review.

[0040] In this scenario, the applicant's business-related data can include: the applicant's credit history, asset information, etc. Through this solution's multi-method collaborative approval process, potential security issues such as credit risk and insufficient repayment ability in pending applications can be effectively identified.

[0041] For example, in the field of supply chain management, the types of business items pending approval may include: supplier access review, purchase order risk assessment, and logistics service provider screening.

[0042] In this scenario, business-related data may include enterprise qualification certificates, historical on-time delivery rates, and product quality inspection reports; through the intelligent approval of this solution, the stability, performance capability, and potential operational risks of suppliers can be comprehensively assessed.

[0043] Therefore, the embodiments of this application do not limit the specific application fields.

[0044] In an exemplary embodiment, such as Figure 1 The diagram shown is a data processing flow diagram of a computing device provided in an embodiment of this application, specifically including the following: Data acquisition stage: Collect business types and business-related data of applicants from multiple data sources for the business to be approved, and standardize the format and handle missing values ​​for the collected attribute information, resource status information and behavioral information.

[0045] Method matching step: Based on the acquired business type, determine the target data fusion method (combination) suitable for the current business scenario and business type from the available data fusion method set.

[0046] Data fusion process: Execute one or more of the following data fusion methods in the data fusion order: Rule-based reasoning fusion process: Perform logical judgments on data based on predefined business rules; Statistical analysis fusion process: Extract multi-dimensional statistical features from operational and behavioral data; Graph computing fusion process: Construct resource interaction graphs and analyze network topology features; Machine learning fusion process: Perform intelligent security score prediction based on feature vectors.

[0047] In the results integration phase: conflict detection is performed on the data fusion results output by each fusion processing method. When contradictions are found, a manual conflict confirmation process is initiated to determine the valid data fusion result. Based on the preset approval rules and the valid data fusion result, a business approval result is generated.

[0048] Strategy optimization phase: Continuously monitor the approval efficiency data of each of the applicant's multiple evaluated business processes, including efficiency data indicators during the approval process, such as approval time, arbitration trigger frequency, and result deviation. By analyzing the approval efficiency data, adjust the data fusion strategy for the applicant.

[0049] For a detailed description of the data processing steps described above, please refer to the specific implementation examples below, which will not be described in detail here.

[0050] like Figure 2 As shown in the embodiment of this application, a business approval method, when applied to the aforementioned computing device, specifically includes the following: S101. Obtain the business types of the pending approval business and the business association data of the application objects associated with the pending approval business.

[0051] The business-related data of the applicant includes: the applicant's attribute information, the applicant's resource status information, and the applicant's behavior information.

[0052] Specifically, pending approval transactions refer to business requests that are reviewed and evaluated by computing devices, resulting in a decision to approve or disapprove. Business types can be categorized by labels or identifiers based on their business attributes, security characteristics, or approval processes. Business-related data refers to data structures related to the currently pending approval transactions and their applicants, which can be used to support decision-making.

[0053] In complex business approval scenarios, this application can build a more comprehensive decision-making context environment for computing devices by systematically collecting business-related data, so that the computing devices have more comprehensive information and thus overcome the defects of insufficient approval accuracy caused by the partiality and ambiguity of input information.

[0054] In some embodiments, the attribute information of the applicant includes: basic information of the applicant and credibility information of the applicant. The resource status information of the applicant includes: resource inventory information, available resource information, resource input information, and resource output information of the applicant. The behavioral information of the applicant includes: resource interaction event records, business request records, and business browsing records of the applicant.

[0055] Specifically, credibility information refers to the evaluation parameters formed by information such as the applicant's historical performance reputation, which can objectively reflect the applicant's performance capability in past business dealings and provide quantifiable confidence for current approval decisions.

[0056] Available resource information refers to the information on available resources remaining after deducting unusable resources from the applicant's existing resource stock. It can accurately reflect the applicant's resource guarantee capacity to fulfill its contractual obligations when relying on its own resources.

[0057] For example, in the financial sector, this resource could be available credit limits in an account, credit card limits, or limits in a tangible entity. Alternatively, in the supply chain management sector, this resource could be available warehousing capacity, deployable transportation capacity, or available raw material inventory.

[0058] Resource interaction event logs refer to the resource input and output logs involved in the request, which may include: resource transfer time, resource quota, resource transfer object, etc. By analyzing these serialized behaviors, the resource flow patterns of the requester and potential abnormal resource interaction behaviors can be effectively reflected.

[0059] Business browsing history refers to the browsing logs of an applicant before submitting a business application, including the browsing duration for different business types and the order in which business details were viewed. Mining these browsing records can help determine the maturity of the business request, the prudence of the decision-making process, and the applicant's understanding of key information.

[0060] In some embodiments, the process by which a computing device obtains the business category of a business to be approved can be represented as follows: First, obtain the description information of the business to be applied for, and use a lightweight text classification model (a bidirectional encoder representation model based on a transformer) to analyze the description information in real time, and automatically output the business category to which the business to be approved belongs.

[0061] It should be understood that this approach is particularly suitable for handling entirely new or undefined business types, and can enhance the adaptability of computing devices to complex business scenarios.

[0062] In some embodiments, the computing device can introduce a data availability assessment mechanism for acquiring business-related data. Specifically, before fetching business-related data, the computing device first checks the accessibility and expected latency of each data source. If a critical data source is temporarily unavailable, the data acquisition strategy can be dynamically adjusted. For example, cached historical data can be used as a substitute, or subsequent degradation processing can be performed based on a subset of available data, thereby ensuring that the computing device can maintain a certain level of service capability even if some dependencies fail.

[0063] One possible implementation is that the computing device can receive raw, pending business requests through its deployed business request access interface (such as an API gateway interface or message queue). Subsequently, the metadata or predefined fields of the business request are parsed to determine its business type. Simultaneously or subsequently, the computing device initiates a data aggregation process. This process, based on the determined business type, pulls relevant business-related data from both internal and external data sources within the computing device. All this data, after data cleaning and standardization, is assembled into a structured data object (business-related data).

[0064] For example, the business-related data can be a JSON object or a protocol buffer format object.

[0065] S102. Based on the business type and the preset business type and data fusion method mapping rules, determine at least one target data fusion method from the preset data fusion methods.

[0066] The preset data fusion methods include: data fusion methods based on business rules, data fusion methods based on statistical methods, data fusion methods based on graph databases, and data fusion methods based on machine learning algorithms.

[0067] The mapping rule between business type and data fusion method is a pre-stored decision logic within the computing device. Essentially, it's a knowledge model that associates business scenario classification with optimal analysis strategies. This rule can determine the data fusion method the computing device should invoke based on the business type.

[0068] Data fusion methods refer to algorithms used by computing devices to extract deep features, identify hidden patterns, and generate quantitative analysis results from multi-dimensional business-related data.

[0069] Specifically, the business rule-based data fusion method refers to the hard filtering and classification of business-related data by executing a series of predefined logical judgment conditions (i.e., "business rules") formulated by domain experts. This is used to quickly and reliably execute deterministic decisions and ensure that approvals meet basic compliance requirements and security bottom lines.

[0070] Data fusion based on statistical methods refers to the application of probability theory and mathematical statistical models (such as regression analysis, analysis of variance, and outlier detection) to perform quantitative statistics, trend fitting, and significance analysis on business-related data. This aims to identify key data characteristics and their correlations that affect business approval results, quantify the degree of abnormal fluctuations in business indicators, and predict the future development trend of the business request, thus providing objective, data-driven quantitative basis for decision-making.

[0071] Data fusion methods based on graph databases refer to using graph theory principles to construct a network graph of business entities (e.g., users, enterprises, accounts) and their relationships (e.g., transactions, social interactions), and using graph algorithms (e.g., community detection, centrality measurement, path analysis) to mine potential associations and structural patterns. This is used to deeply mine complex relationships, identify potential abnormal collaborative groups or abnormal association chains, and reveal the systemic structural characteristics and contradictions hidden in the relationship network.

[0072] Data fusion methods based on machine learning algorithms refer to the use of machine learning models (such as gradient boosting trees, support vector machines, or neural networks) to automatically learn the complex nonlinear mapping relationship between massive, high-dimensional features and the final approval results, thereby achieving end-to-end pattern recognition and prediction. This is used to capture and quantify complex and hidden risk patterns that are difficult to describe with explicit rules or simple statistics, and is particularly good at adaptively responding to new and changing abnormal business behaviors.

[0073] Different business types have inherent differences in security priorities, data patterns, and decision-making logic. If computing devices ignore these differences and employ a single or fixed combination of analysis methods, it will inevitably lead to insufficient mining of key features, thus severely impacting the accuracy of the final approval conclusion. Therefore, this step establishes an adaptation relationship between the inherent characteristics of the business and data processing capabilities, enabling computing devices to dynamically allocate appropriate data fusion methods for specific business scenarios. This ensures that key features are fully mined, thereby improving the accuracy and scenario adaptability of approval conclusions.

[0074] In some embodiments, the preset mapping rule between service types and data fusion methods is implemented as a mapping configuration table pre-installed in the storage unit of the computing device. This configuration table uses the service type as the primary key, and its associated values ​​are identifiers of one or more data fusion methods. After obtaining the service type from S101, the computing device can directly determine the target data fusion method by querying this configuration table.

[0075] One possible implementation is that the computing device uses the service type as the query key to perform an exact match search in the mapping configuration table. If a matching item is found, the target data fusion method for this approval process is determined based on one or more data fusion method identifiers in that matching item.

[0076] For example, in a mapping configuration table for a security verification scenario, if the business type of the business to be approved is innovative product insurance, then its possible target data fusion methods can be mapped as: data fusion method based on machine learning algorithms, and data fusion method based on statistical methods.

[0077] In other embodiments, the preset mapping rule between business types and data fusion methods can be a lightweight machine learning selection model (e.g., random forest or shallow neural network). The computing device takes the business type and several key features initially extracted from the business-related data as input feature vectors, processes them through the model, and the output layer nodes of the model correspond to the recommendation weights of different data fusion methods. The computing device finally selects one or more methods with the highest weights as the target data fusion method.

[0078] One possible implementation involves the computing device concatenating business types (typically vectorized using one-hot encoding, etc.) and / or key features extracted from business-related data (such as resource inventory, historical business approval rates, etc.) into a feature vector. This feature vector is then input into a machine learning selection model. Through forward computation, the model generates a recommendation weight or probability score for each candidate data fusion method at the output layer. The computing device ultimately determines the final target data fusion method combination based on a preset strategy (e.g., selecting the top K methods with the highest weights, or selecting all methods with weights exceeding a specific threshold).

[0079] For example, K can be 2, 3 or 4, and the specific value of K is not limited in the embodiments of this application.

[0080] Furthermore, after determining the target data fusion method, the computing device can further introduce a resource-aware execution optimization strategy. Specifically, the computing device will evaluate the computing resource consumption corresponding to each target data fusion method and the current load status of the computing device in real time. If the computing device load is too high, computationally intensive methods (such as database data fusion methods) can be dynamically and temporarily removed from the method combination, prioritizing the execution of time-sensitive methods (such as business rules) to ensure the response speed of the approval service, and triggering a supplementary analysis process after the computing device load decreases.

[0081] S103. Process business-related data through each target data fusion method to obtain the data fusion result corresponding to each target data fusion method.

[0082] Among them, the data fusion result is a specific and directly applicable conclusion derived from the processing of the original business data by each target data fusion method. For example, the data fusion result can be a specific quantitative indicator such as "the blacklist rule has been triggered", "the transaction frequency has increased abnormally in the past month", "there are financial transactions with 3 high-risk entities", or "the model predicts a default probability of 85%".

[0083] This step allows each selected data fusion method to play its unique role, examining the pending approval process from different perspectives. Specifically, business rule-based methods ensure that approvals meet basic compliance and security requirements; statistical analysis methods excel at identifying anomalous behaviors that deviate from normal patterns from historical data; graph database methods focus on revealing complex relationships and network structures between entities; and machine learning models can perform deep feature extraction and nonlinear relationship modeling on the aforementioned data fusion results and business-related data. Through these data fusion results, computing devices can construct a three-dimensional, comprehensive evaluation view, effectively overcoming the technical limitations of single-analysis method perspectives, thereby significantly improving the accuracy and reliability of approval decisions.

[0084] In some embodiments, the computing device can independently process business-related data using each target data fusion method to obtain different data fusion results. That is, multiple data fusion methods are executed in parallel.

[0085] One possible implementation is that the computing device creates multiple independent task threads or processes to simultaneously distribute the complete business-related data to various target data fusion methods for processing. After all methods return results, the data is then uniformly processed into subsequent decision-making steps.

[0086] In other embodiments, the computing device may also process the output of the previous target data fusion method as the input data of the next target data fusion method according to the data fusion order between the target data fusion methods, that is, multiple data fusion methods are executed in series.

[0087] S104. Based on the business type and the data fusion result of each business, conduct business approval for the business to be approved, and obtain the business approval result of the business to be approved.

[0088] As mentioned above, multiple data fusion results are intermediate conclusions from different analytical dimensions and may be inconsistent. For example, a rule engine might suggest "rejection" due to a violation of a certain clause, while a machine learning model might give a judgment of "low security but passable" due to a good overall pattern. Therefore, this step requires a comprehensive consideration of the various data fusion results and, based on the different priorities of security, compliance, and business development goals for different business types, to form the final business approval result.

[0089] In some embodiments, the computing device has a pre-built library of decision integration strategies bound to different business types. Each strategy clearly defines how to interpret and integrate various data fusion results to form a final conclusion. When performing this step, the computing device first calls the corresponding decision integration strategy according to the business type determined in S101, and then uses all the data fusion results obtained in S103 as input to execute the logic defined by the strategy.

[0090] One possible implementation, for business types with high resource exchange volumes or extremely high compliance requirements, adopts a security-first decision-making strategy. Under this strategy, the conclusions of data fusion methods based on business rules and data fusion methods based on graph databases have the highest priority. If either method's output contains a low-security conclusion, the computing device determines the approval result as "disapproved."

[0091] Another possible implementation approach, for standardized business types with small resource exchange volumes or mature business models, adopts an efficiency-first integration strategy. Under this strategy, the computing device first confirms that the data fusion method based on business rules has not triggered any veto items. Subsequently, it mainly relies on the high-confidence predictions output by the data fusion method based on machine learning algorithms to automatically complete the approval process. This strategy maximizes approval efficiency while maintaining basic security.

[0092] Another possible approach, for complex business types with novel business models or highly conflicting analytical conclusions, employs a comprehensive evaluation-based integration strategy. Under this strategy, the computing device uses a weighted fusion algorithm to assign dynamic weights to the results of each data fusion method. It performs multi-dimensional evaluations based on business scenario characteristics and the correlation between results, ultimately generating a balanced conclusion that comprehensively considers all evidence.

[0093] For details, please refer to the implementation method. Figure 5 The details of this and related information will not be elaborated here.

[0094] The business approval method provided in this application achieves intelligent processing by establishing a matching mechanism between business types and data fusion methods. Specifically, this solution first establishes a comprehensive decision-making information foundation by acquiring business types and multi-dimensional business-related data. Then, based on preset mapping rules, it determines differentiated data fusion method combinations for different business types, ensuring the adaptability of data processing to business characteristics. Furthermore, by executing multiple fusion algorithms, it fully leverages the advantages of rule engines, statistical analysis, graph computing, and machine learning. Finally, it comprehensively evaluates the multi-source fusion results based on business types to form the business approval result.

[0095] The following section introduces the process of data processing using data fusion methods.

[0096] As can be seen from the foregoing, the modes in which computing devices invoke the target data fusion method can include: serial invocation.

[0097] In a chain of calls, the data fusion order of the target data fusion method can be determined by the following rules: Determined based on data dependencies: The computing device has a pre-built dependency graph between methods. If the input of one target data fusion method depends on the output of another target data fusion method, the computing device will force the dependent method to be executed first. For example, the statistical features generated by a data fusion method based on statistical methods are often used as one of the input features of a data fusion method based on machine learning algorithms. In this case, the statistical method must be executed before the machine learning method.

[0098] Based on computational complexity and timeliness requirements, the computing device dynamically plans the data fusion order according to predefined complexity indicators for each method and the timeliness requirements of the current business. This typically follows a "fast first, slow later" or "simple first, complex later" principle, prioritizing fast, lightweight methods (such as data fusion methods based on business rules) followed by computationally time-consuming, complex methods (such as data fusion methods based on graph databases). This approach helps to quickly generate preliminary conclusions and terminate the process early if necessary.

[0099] Fixed priority determination based on configuration: The computing device maintains a static priority list of all data fusion methods. After determining the target combination of methods for sequential execution, the computing device calls and executes them in descending order of execution priority according to the list.

[0100] In some embodiments, the serial execution process of step S103 can be implemented as follows: according to the data fusion order between the target data fusion methods, the target data fusion methods are executed sequentially to process business-related data, and the data fusion result corresponding to each target data fusion method is obtained.

[0101] Among them, the input data of the target data fusion method that ranks later in the data fusion order includes at least the data fusion result output by the target data fusion method preceding it.

[0102] Specifically, this implementation organizes multiple target data fusion methods into an ordered data fusion chain. In this chain, the output of a preceding method (i.e., its data fusion result) serves as one of the inputs for subsequent methods. This design allows information to flow and add value between methods, enabling subsequent methods to conduct deeper analysis based on the higher-level "features" or "conclusions" extracted by preceding methods. This execution method is particularly suitable for complex scenarios where the analytical logic has a clear progressive relationship, gradually transforming coarse raw data into refined decision-making criteria.

[0103] For example, suppose the sequential data fusion order of the target data fusion method is: data fusion method based on statistical methods, data fusion method based on business rules, data fusion method based on graph databases, and data fusion method based on machine learning algorithms. The specific data processing procedure is as follows: First, the computing device executes a data fusion method based on statistical methods to perform quantitative analysis on business-related data and generate statistical results including "the transaction frequency of the applicant has increased abnormally by 200% in the past month".

[0104] Subsequently, the computing device inputs the aforementioned statistical results and business-related data into a data fusion method based on business rules, triggering the "abnormal transaction frequency" rule and generating the corresponding rule triggering result.

[0105] Next, the computing device inputs the rule triggering result and all the aforementioned data into a graph database-based data fusion method. Through graph analysis, it is found that the applicant is closely related to three high-risk entities, and a graph analysis result containing this relationship is generated.

[0106] Finally, the computing device inputs all the aforementioned data fusion results into a data fusion method based on machine learning algorithms. This method comprehensively evaluates and nonlinearly models multi-dimensional information such as abnormal transaction behavior, rule triggering, and high-risk correlation networks, ultimately outputting a quantitative business security score, which directly represents the overall security status of the business pending approval.

[0107] It can be seen that the execution order of the target data fusion method is not fixed. The computing device can dynamically combine and adjust the sequential data fusion order based on the type of service, the real-time load of the computing device, or the presence or absence of specific features in the service-related data. This application embodiment does not limit the specific arrangement of the data fusion order.

[0108] In some embodiments, when the target data fusion method includes a graph database-based data fusion method, such as Figure 3 As shown, the data fusion method based on graph databases may include the following steps: S201. Based on business-related data, identify the interaction objects that have resource interaction relationships with the applicant.

[0109] The interaction object can be an entity that can establish a resource exchange relationship with the applicant, such as an individual, an enterprise, or other business approval equipment.

[0110] As mentioned above, relevant approval mechanisms typically focus on the applicant's own information; however, such mechanisms struggle to effectively identify systemic security issues arising from the actions of related parties. Therefore, in this embodiment, by proactively identifying and defining interaction objects that have direct resource interaction relationships with the applicant, a foundation is provided for placing them within a relationship network for in-depth analysis, thereby enabling the assessment of potential business security from the perspective of network relationships.

[0111] In some embodiments, to improve the efficiency and relevance of analysis, the computing device may introduce a filtering mechanism when identifying interaction objects. For example, the computing device may have preset resource interaction types, and only include counterparties involved in these resource interaction types in the set of interaction objects, thereby focusing on the core relationship network that is most likely to influence approval decisions.

[0112] One possible implementation involves the computing device extracting resource interaction event records and business request records of the requesting object from business-related data. By parsing the "opponent" field in these records, or through correlation queries, all unique entities that have had resource interactions with the requesting object within a specific time window (such as the past 180 days) are aggregated and identified as the interaction objects.

[0113] S202. Construct a resource interaction relationship graph using the application object and the interaction object as nodes.

[0114] The resource interaction relationship graph is a data model that uses a graph structure to represent the resource interaction network between entities. In this graph, nodes represent various entities participating in resource interaction (including requesting objects and interaction objects), and edges represent the specific resource interaction relationships that occur between entities.

[0115] For example, based on the direction of resource flow, the resource interaction relationship can include: resource input relationship, resource output relationship, resource circulation relationship, and resource transfer relationship. Alternatively, based on the stability of resource exchange frequency, the resource interaction relationship can be classified into high-frequency stable relationship, low-frequency occasional relationship, and periodic pulse relationship. This application does not limit the specific implementation of the resource interaction relationship.

[0116] Simply listing interactive objects only provides discrete relationship information, but cannot demonstrate the strength, density, and topological structure of these relationships. Therefore, in this embodiment, a resource interaction relationship graph is constructed to integrate discrete interaction relationships into an organic network model, providing a structured data foundation for subsequent graph-based deep feature extraction.

[0117] One possible implementation involves the computing device importing the requesting object and all interacting objects as nodes into a graph database. Based on resource interaction event records, directed or undirected edges are established between nodes with interactive relationships. Each edge can be appended with attributes such as resource interaction amount, frequency, and most recent interaction time, forming a weighted resource interaction relationship graph.

[0118] In some embodiments, to improve the efficiency of graph construction, the computing device may employ an incremental update mechanism. For recurring interaction objects, nodes are not repeatedly created; instead, the weights and time-sensitivity attributes of their associated edges are updated, thereby constructing a dynamically evolving resource interaction relationship graph.

[0119] S203. Analyze the topological structure of the applicant in the resource interaction relationship graph, and obtain the resource exchange relationship characteristic index of the applicant as the data fusion result of the data fusion method based on graph database.

[0120] Among them, the resource exchange relationship characteristic index characterizes the stability and scope of influence of the resource interaction network of the applicant.

[0121] Specifically, the resource exchange relationship characteristic indicators may include at least one of the following: degree centrality, betweenness centrality, proximity centrality, eigenvector centrality, and network constraint coefficient.

[0122] Degree centrality: refers to the number of interactive objects directly connected to the applicant, reflecting the scale of its direct influence in the resource interaction graph.

[0123] Betweenness centrality: measures the degree to which an applicant acts as a mediator in the resource interaction graph, reflecting its control over resource flow.

[0124] Proximity centrality: Characterizes the average distance from the requesting object to all other nodes in the resource interaction graph, reflecting its efficiency in information acquisition and resource access.

[0125] Eigenvector centrality: It considers not only the number of direct connections of the applicant, but also the importance of its connected objects, and comprehensively evaluates its position in the resource interaction graph.

[0126] Network constraint coefficient: measures the degree to which the applicant is structurally constrained in the resource interaction relationship graph. The lower the coefficient, the more non-redundant connections it has, and the stronger the graph structure stability.

[0127] For example, suppose the resource interaction relationship graph of a certain application object A contains 125 nodes, and it has active resource interactions with 98 of these nodes in the past 30 days. The corresponding resource exchange relationship feature indicators can be: degree centrality 0.784 (standardized value), betweenness centrality 0.215, graph density change rate in the past 30 days +0.33, and monthly variance of eigenvector centrality 0.087.

[0128] Specifically, the degree centrality of 0.784 indicates that it has direct business connections with 78.4% of the nodes in the network, while the monthly variance of the eigenvector centrality of 0.087 reflects the stability of this core position; meanwhile, the betweenness centrality of 0.215 indicates that it plays a moderate mediating role in resource flow. Furthermore, the graph density change rate of +0.33 shows that its business relationship network is continuously strengthening and developing healthily. These indicators can characterize the robust business network of this applicant during a period of healthy expansion.

[0129] In some embodiments, to enhance the time sensitivity of feature indicators, the computing device can assign weights to edges in the resource interaction relationship graph based on a time decay function. Specifically, weights can be calculated using exponential or linear decay functions based on the timestamps of resource interaction events, so that recently occurring resource interactions contribute more significantly when calculating topology indicators. In this way, the calculated resource exchange relationship feature indicators can more accurately reflect the current (rather than historical) influence and stability of the applicant in the resource interaction relationship graph.

[0130] In some embodiments, to adapt to the assessment needs of development trends in specific business scenarios, the computing device can also calculate dynamic graph evolution indicators. For example, by constructing resource interaction relationship graph subgraphs for different time windows (e.g., monthly), macro-indicators such as the graph density change rate and average path length change of these subgraphs can be calculated and compared; or the fluctuation variance and evolution trend of the core topological indicators (e.g., degree centrality, eigenvector centrality) of the application object nodes at different periods can be calculated. These dynamic graph indicators can capture the structural evolution patterns of the application object's resource interaction relationship graph, thereby enhancing the ability of feature indicators to predict business security trends.

[0131] One possible implementation involves the computing device employing a graph computing engine to perform a series of graph algorithm calculations on the requesting node, including but not limited to: calculating degree centrality to measure its direct connectivity, calculating betweenness centrality to assess its network control capability, executing community detection algorithms to determine its community characteristics, and calculating clustering coefficients to analyze the density of its neighborhood network. The outputs of these algorithms are then combined to form a multi-dimensional resource exchange relationship characteristic index.

[0132] In some embodiments, when the target data fusion method includes a data fusion method based on machine learning algorithms, such as Figure 4 As shown, a data fusion method based on machine learning algorithms may include the following steps: S301. Generate a feature vector based on at least one of attribute information, resource status information, and behavioral information, and / or the data fusion result obtained through other data fusion methods.

[0133] In this context, a feature vector is a data structure that transforms multi-source heterogeneous data into a standardized numerical vector that can be processed by a machine learning model. Each dimension represents a specific feature after feature engineering. Specifically, the original business data and the results of various data fusion methods differ significantly in data type, scale, and distribution, making them unsuitable for direct and effective processing by machine learning models. Therefore, this embodiment of the application achieves data standardization and structuring through a feature vector generation step, providing standardized input for subsequent machine learning models.

[0134] This step employs a multi-source data fusion strategy based on the following technical considerations: First, attribute information, resource status information, and behavioral information characterize the applicant's status from three orthogonal dimensions: static qualifications, dynamic capabilities, and historical performance. Combining these three aspects avoids the bias of a single-dimensional assessment. Second, the introduction of other data fusion methods is based on the idea of ​​cascaded fusion, aiming to use the deep features extracted by previous methods (such as rule engines, statistical analysis, and graph calculations) as new evidence input, enabling machine learning models to perform higher-level pattern recognition. This design effectively solves the technical problems of insufficient information from a single data source and the isolation of conclusions from various analytical methods, achieving a layer-by-layer refinement from raw data to intelligent features.

[0135] One possible implementation involves the following process for the computing device to generate feature vectors: First, numerical features are standardized to eliminate the influence of dimensions; categorical features are one-hot encoded or target encoded to convert them into numerical form. Then, the computing device concatenates and aligns the raw features from attribute information, resource status information, and behavioral information with the resulting features output from other data fusion methods.

[0136] In addition, during this process, the computing device will also perform feature selection. By evaluating the correlation between features and approval results, it will retain information-rich features, eliminate redundant features, and finally generate an optimized, fixed-dimensional feature vector.

[0137] For example, suppose we are processing a business approval case containing 235 original features. After feature selection, the computing device retains the 87 features with the highest information content. These features include: 15 core features selected from attribute information (e.g., credibility score 0.82), 22 key features selected from resource status information (e.g., available resource ratio 0.67), 28 behavioral features selected from behavioral information (e.g., standardized value of resource interaction frequency in the past 30 days 1.24), and 22 deep features selected from the results of other data fusion methods (e.g., spectral centrality 0.784, rule triggering identifier 0). The final generated feature vector has a dimension of 87, where the value of each feature has been standardized and lies in the range [0,1] or [-1,1], forming a standardized input that can be directly processed by the machine learning model.

[0138] S302. By processing business types and feature vectors through a preset business approval model, the business security score of the business to be approved is obtained as the data fusion result of the data fusion method based on machine learning algorithm.

[0139] Among them, the business type is used to indicate the weight allocation of the security assessment in the preset business approval model.

[0140] By using business type as key contextual information, the model can understand the specific business scenario being evaluated, thereby dynamically adjusting its internal decision boundaries or feature weights to achieve accurate and scenario-adaptive security assessments.

[0141] In some embodiments, the preset business approval model is a machine learning model trained based on historical approval data. Its specific type can be selected according to the application scenario, such as a gradient boosting decision tree model or a deep neural network model.

[0142] Among them, the gradient boosting decision tree model is an ensemble learning model that effectively captures complex nonlinear relationships and interaction effects in features by iteratively training multiple weak decision trees and combining their prediction results. The gradient boosting decision tree model is particularly suitable for approval scenarios with relatively clear business rules, moderate feature dimensions, and a need for strong model interpretability.

[0143] For example, in the approval process for equipment leasing, the features typically include explicit numerical variables (such as equipment value, average daily usage time, and historical maintenance records) and categorical variables (such as equipment type and leasing scenario). Gradient boosting decision tree models can effectively learn the complex nonlinear relationships between these features (such as the interactive impact of equipment value and usage time on equipment depreciation rate), and at the same time, provide an intuitive basis for approval decisions by ranking the importance of features.

[0144] Deep neural network models are neural network models containing multiple hidden layers. They possess powerful representation learning capabilities, excelling particularly at handling high-dimensional, sparse feature vectors, and can automatically learn hierarchical feature representations. Deep neural network models are therefore more suitable for modern internet business approval scenarios characterized by complex business models, high feature dimensionality, and hidden interaction relationships.

[0145] For example, when evaluating an application for an innovative financial product that integrates e-commerce transactions, social networks, and behavioral logs, its feature vectors are often thousands of dimensions high and contain a large number of sparse features (such as click sequences and text description embeddings). Deep neural network models, with their multi-layered nonlinear transformation capabilities, can automatically extract deep, abstract risk patterns from the original high-dimensional features, identifying collaborative abnormal behavioral characteristics that are difficult for traditional models to detect, thereby addressing the risks of rapidly evolving online businesses.

[0146] For example, for tree ensemble models such as gradient boosting decision trees, the output business security score is typically represented as a probability value between 0 and 1, such as 0.85. For deep network models such as deep neural network models, the output can be a probability value in the same form, or it can be a percentage score after linear transformation, such as 85. In addition, kernel method models such as support vector machines can output distance-based decision values, and logistic regression models can output log odds values. This application does not limit the specific numerical form of the business security score.

[0147] In some embodiments, when the target data fusion method includes a business rule-based data fusion method, the business rule-based data fusion method may include the following steps: First, the computing device filters a set of applicable business rules from a preset rule base according to the business type. Second, the business-related data is matched one by one with the selected business rule triggering conditions to identify all triggered target business rules. Then, the rule conclusions that may conflict are coordinated according to a preset conflict resolution strategy. Finally, the decision conclusions of all triggered rules are combined to generate a rule decision result as the data fusion result of the method.

[0148] For example, suppose the business type of the application to be approved is K-03, and the associated data includes: a credibility score of 65 (out of 100), a requested resource quantity of 50,000 units, a available resource quantity of 45,000 units, and a resource interaction frequency of 85 times in the past 30 days. The computing device matches the data according to a preset rule base: Rule B-01 requires that the credibility score is <70 and the requested resource quantity is > the available resource quantity, and Rule B-05 requires that the interaction frequency in the past 30 days is >80. Both rules are triggered, and the final rule decision result is generated: {Number of rule triggers: 2, Highest security level: Medium, Security label: ["Insufficient resources", "High-frequency resource exchange"]}.

[0149] In some embodiments, when the target data fusion method includes a statistical method-based data fusion method, the statistical method-based data fusion method may include the following steps: First, the computing device extracts a sequence of key indicators within a preset time period from business-related data. Second, statistical characteristics are calculated on the key indicator sequences to obtain multidimensional statistics, including central tendency, dispersion, and distribution pattern. Then, the calculated statistics are compared and analyzed with a benchmark statistical model established based on historical data to identify anomalous features that significantly deviate from the normal pattern. Finally, based on the severity and combination of anomalous features, statistical analysis results are generated as the data fusion result of the method.

[0150] For example, assuming a user's resource interaction data over the past 90 days (e.g., average daily interaction count of 21.5, standard deviation of 7.8; average single interaction amount of 2350, standard deviation of 1250), the computing device compares it with historical benchmarks: the average interaction count deviation is +28.3%, the interaction amount coefficient of variation deviation is +86.5%, and the outlier rate is 12.2%. Through comprehensive analysis, a comprehensive anomaly score of 76.5 (out of 100) is generated, identifying the main anomalies as abnormal fluctuations in interaction amount and abnormal interaction frequency, with the security level assessed as medium security.

[0151] It should be understood that by establishing a sequential execution mechanism for data fusion methods, a hierarchical feature extraction and decision support process is constructed. This sequential processing approach allows subsequent methods to conduct further analysis based on the in-depth results of preceding methods, resulting in progressive information refinement. In particular, by mining network topology features using graph database methods and then combining them with machine learning methods for comprehensive scoring and prediction, the depth of feature mining is ensured, the intelligence level of the final decision is improved, and the processing capability of computing devices for complex business data is effectively enhanced.

[0152] In some embodiments, when significant differences or contradictions exist among multiple data fusion results, the computing device will initiate a conflict resolution mechanism to ensure the reliability of the final approval conclusion. In this case, such as... Figure 5 As shown, S104 specifically includes the following steps: S401. Based on the preset result conflict detection rules, perform conflict detection on each data fusion result to obtain the conflict detection result.

[0153] Among them, the result conflict detection rules include contradictory combinations between different data fusion results.

[0154] Result conflict detection rules are predefined logical judgment conditions in computing devices used to identify logical contradictions between different data fusion results. A contradictory combination refers to a combination of two or more data fusion results that is predefined as logically conflicting. These combinations reflect that different data fusion methods, when analyzing the same pending approval business, have reached mutually exclusive conclusions in terms of decision-making logic.

[0155] Because different data fusion methods are based on different analytical dimensions and algorithmic principles, they may produce contradictory conclusions in complex business scenarios. Therefore, this step, through systematic conflict detection, can promptly identify and label these contradictions, providing a basis for subsequent conflict resolution and ensuring the reliability of the final approval conclusion.

[0156] In some embodiments, the contradictory combination includes at least one of the following: Security level conflict combination: The security level output by the data fusion method based on business rules differs from the security level output by the data fusion method based on machine learning algorithms by more than the preset level.

[0157] Conflicting conclusions: The final conclusions of the data fusion method based on business rules are directly opposed to the final conclusions of the data fusion method based on machine learning algorithms.

[0158] Numerical conflict combination: The anomaly score output by the data fusion method based on statistical methods and the security score output by the data fusion method based on machine learning algorithms are respectively located in the preset conflict interval.

[0159] For example, suppose the contradictory combination configuration table defines that when the method based on business rules outputs a negative conclusion, while the method based on machine learning algorithms outputs a positive conclusion, a contradictory combination of data fusion results is formed.

[0160] In some embodiments, the computing device may determine preset conflict detection rules in at least one of the following ways: Based on expert experience configuration: Domain experts directly define typical conflict combinations and their conflict levels based on business knowledge and risk preferences, forming an initial conflict detection rule base. Based on historical conflict data mining: The computing device analyzes conflict cases in historical approval processes, automatically discovers frequently occurring conflict combination patterns through association rule mining algorithms, and adds them to the conflict detection rule base. Based on machine learning dynamic optimization: The computing device trains a conflict prediction model using labeled conflict cases. This model can automatically identify new conflict combination patterns and dynamically update the conflict detection rules.

[0161] S402. If at least one data fusion result combination in the conflict detection results is a contradictory combination, a conflict report shall be generated.

[0162] A conflict report is a structured document generated by a computing device that details the identified conflict combinations, the data fusion methods involved, the output results of each method, and their confidence levels.

[0163] Specifically, simple conflict identification is insufficient to support subsequent decision-making; complete conflict context information is required. This step generates a detailed conflict report, providing sufficient decision-making basis for human intervention and ensuring the quality and efficiency of conflict resolution.

[0164] One possible implementation involves the computing device formatting the collision detection results according to a preset template, including key information such as the collision type, the data fusion methods involved, the timestamps of the output results of each method, and the confidence score. The report is presented using a combination of machine-readable JSON format and human-readable natural language descriptions.

[0165] In some embodiments, the computing device can automatically convert structured conflict data into a detailed analysis report containing conflict descriptions, causal analysis, and handling recommendations by invoking a natural language generation engine. Alternatively, it can generate a visual analysis report using a data visualization component, presenting the conflict combinations and the confidence levels of each method's results in the form of radar charts or comparative bar charts.

[0166] S403. In response to the conflict confirmation results based on the conflict report input, determine the valid data fusion results.

[0167] Among them, the valid data fusion results include data fusion results that have not triggered conflict rules and data fusion results that have been confirmed as valid and have conflicts in the conflict confirmation results.

[0168] "Human" refers to an expert in the field of the business to be approved who has the authority to approve using computing devices.

[0169] Specifically, in cases where computing devices produce conflicting results, human experience is introduced for final confirmation. This human-machine collaboration ensures both processing efficiency and decision-making quality in complex situations.

[0170] One possible implementation involves the computing device pushing conflict reports to designated approval experts and providing an intuitive human-computer interaction interface for experts to view conflict details. Experts then confirm the conflict outcome based on their business knowledge and experience, and the computing device updates the valid data fusion results according to the experts' confirmation.

[0171] In some embodiments, the computing device can construct a conflict decision-making knowledge base based on expert confirmation results. Specifically, the computing device records key elements in each conflict confirmation process, including the conflict type, the data fusion methods involved and their confidence levels, the final adoption results by experts, and the rationale for the decisions. Through continuous learning of this knowledge base, the computing device can gradually build a conflict resolution model to handle repetitive or patterned conflict scenarios, thereby reducing reliance on human intervention and improving the long-term operational efficiency of the computing device.

[0172] S404. Based on the business type and the effective data fusion results, the business approval results of the business to be approved are obtained.

[0173] The final decision-making process, based on business type and the results of effective data fusion, is grounded in the following technical considerations: First, different business types have varying risk tolerances and approval strategies, requiring different decision-making logics. Second, the results of effective data fusion represent validated multi-dimensional analytical conclusions, and combining these conclusions can form a more comprehensive and reliable basis for approval. This design ensures that the final approval result not only meets the requirements of the specific business scenario but also fully considers the advantages of multi-method collaborative analysis.

[0174] For details on the specific implementation of S404, please refer to the following text. Figure 6 This will not be elaborated upon here.

[0175] It should be understood that a verification and coordination system for the results of multi-method fusion has been established through a result conflict detection mechanism and a manual confirmation process. This solution effectively solves the problem of inconsistencies that may occur in the fusion of multi-source data. Through human-machine collaborative decision-making, it ensures reliability in complex business scenarios while maintaining the efficiency and standardization of the approval process.

[0176] In some embodiments, the computing device can implement the specific process of obtaining the business approval result of the business to be approved based on the business type and the effective data fusion result in S404 based on a preset weighted fusion decision strategy.

[0177] In this case, the process is as follows Figure 6 As shown, S404 may specifically include the following steps: S501. Based on the business type and the preset mapping rule set between business type and fusion result weight, determine the weight value corresponding to each valid data fusion result.

[0178] The preset mapping rule set between business types and fusion result weights defines the proportion of importance of each type of data fusion result in the final decision under different business types.

[0179] In some embodiments, the computing device can autonomously construct and optimize a mapping rule set between the business type and the weight of the fusion result by analyzing the consistency between the data fusion results in historical approval data and the final manual approval result.

[0180] Specifically, the computing device collects historical business approval records. Each record includes the business type, the output results of each data fusion method, and the final manual approval conclusion. For each business type, the computing device calculates the proportion of results from each data fusion method that match the manual approval conclusion. This proportion, after normalization, becomes the initial weight of each method for that business type. The computing device can periodically execute this process to achieve self-learning and continuous optimization of the weight mapping rule set.

[0181] For example, for business category "K-03", after analyzing 1000 past records, the computing device found that the consistency rate between the machine learning algorithm-based method and manual approval was 85%, the business rule-based method was 70%, and the statistical method was 65%. After normalization, the initial weight assignments were 0.39, 0.32, and 0.29, respectively. This weight set was stored in the mapping rule set for subsequent approval.

[0182] In other embodiments, the computing device can generate the mapping rule set by integrating a domain expert knowledge base and combining it with knowledge graph reasoning technology. The computing device parses the descriptions of risk characteristics of different business types in the expert knowledge base, automatically infers the relative importance of each data fusion method in different business scenarios, and thus constructs an initial weight mapping rule set. This application embodiment does not limit the method for obtaining the preset mapping rule set of business types and fusion result weights.

[0183] One possible implementation method, the specific implementation process of S501 can be represented as follows: the computing device uses the current business type as the primary key, queries the preset weight mapping table, and obtains the weight configuration vector corresponding to the business type.

[0184] For example, for business type "K-03", the query yields the following weight configuration: {Results based on business rules: 0.2, Results based on statistical methods: 0.3, Results based on graph databases: 0.1, Results based on machine learning algorithms: 0.4}. Subsequently, the computing device extracts the corresponding weight values ​​from this weight configuration vector and assigns them to each result based on the type of currently existing valid data fusion results.

[0185] S502. The weight values ​​of each valid data fusion result are weighted and summed to obtain the business evaluation score of the business to be approved.

[0186] Specifically, since the results of data fusion reflect business characteristics from different dimensions and have varying degrees of importance, directly using these results makes it difficult to reach a unified decision. By using weighted summation, the multi-dimensional analytical conclusions can be effectively integrated into a comparable comprehensive indicator, providing a quantitative basis for the final approval decision.

[0187] In some embodiments, the computing device adopts a corresponding standardization processing strategy according to the type of each data fusion result: for data fusion results based on business rules, the results are converted into standardized scores through level mapping rules based on the security level and triggering rule characteristics they contain; for data fusion results based on statistical methods, the results are obtained through anomaly score conversion rules based on their anomaly indicators and statistical characteristics; for data fusion results based on graph databases, the results are calculated through graph feature aggregation functions based on their network topology feature indicators; and for data fusion results based on machine learning algorithms, the output business security score is directly used as the standardized score.

[0188] After obtaining the standardized scores of the fusion results of each valid data, the business evaluation score of the business to be approved is calculated according to the weighted summation formula.

[0189] For example, suppose the business type of a pending application is "K-03" (innovative product insurance). The valid data fusion result after conflict confirmation includes the following four: The data fusion result based on business rules is: {Number of rule triggers: 2, Highest security level: "Medium", Security label: ["Insufficient resources", "High-frequency resource exchange"]}. The corresponding standardization process is as follows: The computing device maps the "Medium" security level to a baseline score of 0.6 according to the preset mapping rules. At the same time, based on the "Insufficient resources" and "High-frequency resource exchange" labels, the adjustment coefficients of -0.1 and -0.05 are obtained from the influence coefficient table, respectively. The standardized score is calculated using the formula: s1=0.6+(-0.1)+(-0.05)=0.45.

[0190] The data fusion result based on statistical methods is: {Overall Anomaly Score: 76.5, Main Anomaly Dimensions: ["Amount Volatility", "Interaction Frequency"], Statistical Confidence Level: 0.92}. The corresponding standardization process is as follows: The computing device first converts the percentage-based anomaly score into a baseline score: 1 - 76.5 / 100 = 0.235. Considering the existence of two main anomaly dimensions, 0.1 points are deducted from each dimension, and a confidence level weighting is introduced: s² = 0.235 × 0.92 - 0.1 - 0.1 = 0.016.

[0191] The data fusion result based on the graph database is: {degree centrality: 0.784, betweenness centrality: 0.215, network constraint coefficient: 0.35}. The corresponding standardization process is as follows: The computing device uses a predefined network health assessment function: s3 = 0.5 × 0.784 + 0.3 × (1 - 0.215) + 0.2 × 0.35 = 0.392 + 0.236 + 0.07 = 0.698.

[0192] The data fusion result based on the machine learning algorithm is: {Business security score: 0.9, confidence level: 0.95}. The corresponding standardization process is as follows: the business security score is directly used as the standardization score: s4=0.9.

[0193] Then, the computing device queries the weight mapping rule set based on the business type "K-03" to obtain the weight configuration of each method: the result weight based on business rules is 0.15, the result weight based on statistical methods is 0.20, the result weight based on graph database is 0.25, and the result weight based on machine learning algorithm is 0.40.

[0194] Finally, a weighted sum is calculated based on the scores obtained after the above standardization process: S=(w1×s1)+(w2×s2)+(w3×s3)+(w4×s4)=(0.15×0.45)+(0.20×0.016)+(0.25×0.698)+(0.40×0.9)=0.6052.

[0195] S503. If the business evaluation score is greater than the preset evaluation threshold, the business approval result is determined to be passed.

[0196] The assessment threshold refers to the critical value set for business assessment scores to determine the approval outcome. This threshold is associated with the business type, and different thresholds can be configured for different business types.

[0197] For example, the evaluation threshold may be 0.5, 0.6 or 0.75, etc., and the specific value of the evaluation threshold is not limited in the embodiments of this application.

[0198] Specifically, the business evaluation score is a continuous quantitative value, while the approval decision requires a clear binary conclusion (pass / fail). By comparing it with a preset threshold, the conversion from continuous scoring to discrete decision-making can be achieved, ensuring the clarity and enforceability of the approval conclusion.

[0199] One possible implementation is that the computing device first queries the corresponding evaluation threshold based on the current business type. Then, it compares the business evaluation score calculated by S502 with this threshold: if the business evaluation score is greater than the threshold, an "approved" approval result is generated; otherwise, a "disapproved" approval result is generated. The approval result may include information such as the business evaluation score and a summary of the decision basis.

[0200] For example, suppose the evaluation threshold for business type "K-03" is set to 0.6. The computing device compares the business evaluation score of 0.605 obtained in the previous example with this threshold: since 0.605 > 0.6, the business approval result is determined to be "passed".

[0201] In some embodiments, the computing device may also implement S503 as follows: if the business evaluation score equals a preset evaluation threshold, determine that the business approval result is passed. This application embodiment does not limit the implementation of the threshold boundary judgment.

[0202] In some embodiments, to improve the adaptability of the computing device to business scenarios, the computing device may adopt a dynamic threshold adjustment mechanism. The computing device periodically analyzes the approval effectiveness indicators (such as false approval rate and false rejection rate) of each business type, and automatically optimizes and adjusts the evaluation thresholds based on these indicators, so that the computing device can achieve the optimal approval rate while ensuring the quality of approval.

[0203] It should be understood that a precise decision-making fusion is achieved through a business-type adaptive weight allocation mechanism. This solution can dynamically adjust the contribution of each data fusion result to the final decision based on the characteristics of different business types, ensuring a precise match between approval standards and business needs. This differentiated weight allocation strategy reflects the adaptability of computing devices to different business scenarios and guarantees the objectivity and consistency of the decision-making process through quantitative weighted summation.

[0204] In some embodiments, S503 determines that the business approval result is unsuccessful if the business evaluation score is less than or equal to a preset evaluation threshold. In this case, the computing device can also analyze the approval result to obtain an analysis report indicating that the approval result is unsuccessful. In this case, the process is as follows: Figure 7 As shown, S503 specifically includes the following steps: S601. If the business approval result is not approved, determine the weighted evaluation value of each valid fusion result by using the weight value corresponding to each valid fusion result.

[0205] The weighted evaluation value of each effective fusion result refers to the value obtained by multiplying the data fusion result score by its corresponding weight, reflecting the specific contribution of the result to the final business approval result.

[0206] Specifically, in order to analyze the specific reasons why pending business applications are rejected, it is necessary to quantify the contribution of each valid data fusion result to the final business evaluation score, thereby identifying the influencing factors of the business approval result being rejected.

[0207] This process can be referred to in the example content of S502 above, and will not be described in detail here.

[0208] S602. Compare the weighted evaluation value of each valid fusion result with its corresponding preset single evaluation threshold to determine the substandard data fusion results whose weighted evaluation value is less than the corresponding preset evaluation threshold.

[0209] Substandard data fusion results refer to data fusion results whose weighted evaluation values ​​fail to meet the minimum requirements preset for this type of result.

[0210] The weighted evaluation value of each valid fusion result and its corresponding preset individual evaluation threshold refer to the minimum standard that the weighted evaluation value must reach, which is independently set for each type of data fusion method. In addition, the sum of the weighted evaluation value of each valid fusion result and its corresponding preset individual evaluation threshold is equal to the evaluation threshold preset in S503.

[0211] In some embodiments, the computing device can obtain the preset single evaluation threshold corresponding to each fusion result by querying a configuration table of preset single evaluation thresholds for different service types.

[0212] In other embodiments, the computing device can dynamically calculate the preset individual evaluation threshold corresponding to each fusion result. Specifically, based on the score distribution of various data fusion results in historical approval data and combined with the risk preference of the current business type, the computing device dynamically calculates the individual evaluation threshold that each type of result should reach. For example, for risk-averse business types, the computing device can set a strict individual threshold based on the lower quantile (e.g., the 25th percentile) of the score for this type of result in historical data.

[0213] One possible implementation is that the computing device iterates through each valid data fusion result and performs the following comparison operation: compares the weighted evaluation value of the current result with the single evaluation threshold preset for this type of result; if the weighted evaluation value is less than or equal to the corresponding threshold, then the result is marked as a substandard data fusion result.

[0214] For example, suppose the preset individual evaluation thresholds for each data fusion result under a certain business category are: 0.08 for results based on business rules, 0.12 for results based on statistical methods, 0.15 for results based on graph databases, and 0.25 for results based on machine learning algorithms. If, in a certain approval process, the weighted evaluation value of the result based on statistical methods is 0.09 (less than 0.12), then this result is determined to be a substandard data fusion result.

[0215] S603. Based on the fusion results of non-compliant data and business-related data, generate a business approval result report.

[0216] The purpose of generating business approval result reports is to clearly reveal the specific reasons for approval failure, provide applicants with clear directions for improvement, and provide data support for the continuous optimization of computing equipment.

[0217] In some implementations, computing devices can perform root cause analysis based on the input data (such as business-related data or data fusion results from other data fusion methods) corresponding to the substandard data fusion results. This allows them to identify key characteristic data affecting the substandard data fusion results from the business-related data. Optimization direction analysis is then conducted based on this key characteristic data. Finally, a business approval result report is generated based on the key characteristic data and the optimization direction.

[0218] During this process, the computing device can perform root cause analysis by invoking a preset analysis strategy that matches the data fusion method type corresponding to the non-compliant result: If the failure to meet the target stems from a data fusion method based on business rules, then a rule condition matching backtracking process is performed: the triggered rule is parsed, the specific business data field on which the rule condition depends is located, and this field and its abnormal values ​​are identified as key feature data. For example, if the rule is triggered by "login count in the last 30 days < 5", then the key feature is "login count in the last 30 days = 4".

[0219] If the failure to meet the target score stems from a data fusion method based on a machine learning model, then feature contribution analysis is performed: using the model interpreter, the contribution of each input feature to the low score prediction is calculated, and the negative feature with the highest contribution is identified as the key feature data. For example, the analysis found that "historical overdue number" contributed more than 40% to the low score.

[0220] If the non-compliance result stems from a data fusion method based on statistical approaches, then the source of the anomaly is traced: the comprehensive anomaly score output is analyzed, and its internal statistical indicators (such as variance and month-on-month growth rate) are broken down and analyzed. The indicator that deviates most significantly from the historical benchmark is identified as the key feature data. For example, the comprehensive anomaly mainly stems from the "week-on-week growth rate of transaction amount" reaching a peak of 200%.

[0221] If the failure to meet the target stems from a data fusion method based on graph databases, then graph pattern recognition is performed: analyzing the resource interaction graph, and using graph query and community detection algorithms, identifying abnormal subgraph structures or path patterns that lead to a decrease in risk score. For example, it was identified that the applicant and three known high-risk nodes formed a closed-loop guarantee circle.

[0222] Furthermore, the process by which the computing device performs optimization direction analysis based on this key feature data is as follows: State assessment. The specific values ​​of key feature data are compared with preset normal ranges or ideal states to quantify the degree of deviation.

[0223] Strategy matching. Based on the identity of key feature data (i.e., what business meaning it represents) and its deviations, feasible optimization directions are matched from a pre-built optimization strategy knowledge base. For example, if the key feature is "login count in the last 30 days" and the value is too low, the matched optimization direction is "increase account activity"; if the key feature is "volatility of cash flow" and the value is too high, the matched optimization direction is "maintain stability of cash flow".

[0224] Suggestion generation. The matched optimization directions are combined with specific business scenarios to generate concrete and actionable natural language descriptions, forming improvement suggestions.

[0225] Finally, the computing device integrates the key feature data obtained from the root cause analysis, the improvement suggestions from the optimization direction analysis, and the non-compliance results themselves to generate a structured business approval result report.

[0226] In some embodiments, the business approval result report may include: business approval result, non-compliant data fusion result, key feature data, and optimization direction suggestions.

[0227] It should be understood that by establishing a mechanism for identifying and generating reports on non-compliance results, a transparent approval service is provided. This solution not only provides approval conclusions but also accurately identifies the key factors leading to approval failures, offering applicants clear directions for improvement. This refined results analysis enhances the transparency of the approval process and increases the added value of the service through targeted improvement suggestions, achieving a functional upgrade from simple approval to intelligent service.

[0228] In some embodiments, to improve the long-term performance and user experience of computing devices, the devices can build personalized approval strategies based on the applicant's historical approval performance. This aims to identify and adapt to the behavioral characteristics of different applicants, and improve approval efficiency while ensuring security by dynamically optimizing the allocation of approval resources.

[0229] In this case, the process is as follows Figure 8 As shown, the method in this embodiment also includes the following steps: S701. Obtain the approval efficiency data of each of the multiple evaluated business processes of the applicant.

[0230] Among them, approval efficiency data refers to a set of quantitative indicators generated in the historical approval process that can reflect the efficiency and quality of approval. The approval efficiency data for each evaluated business includes: approval time, number of manual conflict confirmations, and deviation of business evaluation results.

[0231] Specifically, approval time refers to the total time from business submission to the determination of the approval result. The number of times manual conflict confirmation is required refers to the number of times in the approval process, due to conflicts in data fusion results, that the final confirmation is performed manually. The business evaluation result deviation measures the degree of difference between the business evaluation result of the evaluated business and the actual business behavior of the applicant.

[0232] In some embodiments, the process by which the computing device obtains the business evaluation result deviation for each evaluated service can be represented as follows: After a preset observation period ends, the computing device, for an evaluated service, uses its actual business behavior data during the observation period as new business-related data, and re-executes S101 to S104 of the embodiments of this application to obtain a post-evaluation business result that reflects its true business performance. Subsequently, by comparing the degree of difference between the initial business evaluation result and the post-evaluation business result, the business evaluation result deviation is calculated.

[0233] The degree of difference is calculated using an absolute difference normalization method, i.e.: Business evaluation result deviation = |Initial business evaluation result - Post-evaluation business evaluation result| / Scoring range. Specifically, the scoring range is a predefined value. For example, when the computing device uses a percentage system, the scoring range is 100; when the computing device uses a probability score from 0 to 1, the scoring range is 1.

[0234] In addition, the computing device can also determine the approval time and the number of manual conflict confirmations for the evaluated business by parsing the approval logs. The manual conflict confirmation time will have a specific identifier in the approval log, and the computing device can obtain the number of manual conflict confirmations by counting the number of such events recorded during the approval process of the business.

[0235] Through the above embodiments, the computing device can obtain approval efficiency data for each evaluated business.

[0236] S702. Summarize the approval efficiency data of multiple evaluated businesses to determine the approval strategy pattern for applicants.

[0237] Among them, the approval strategy model is a representative pattern of approval behavior identified based on the applicant's historical approval efficiency characteristics. The approval strategy model reflects the typical characteristics of the applicant in terms of approval efficiency, decision complexity, and result accuracy.

[0238] Specifically, pattern induction refers to the process of extracting representative combinations of features from discrete approval efficiency data. Approval strategy patterns can include types such as efficient and stable, routine processing, and complex and controversial. The efficient and stable pattern is characterized by short approval time, few manual conflict confirmations, and low deviation of business evaluation results; the routine processing pattern is characterized by all indicators being at a moderate level; and the complex and controversial pattern is characterized by long approval time, many manual conflict confirmations, and high deviation of business evaluation results.

[0239] In some embodiments, the process by which the computing device performs pattern induction on the approval efficiency data of multiple evaluated services can be represented as follows: the computing device converts the approval efficiency data of each evaluated service into a feature vector, where each dimension corresponds to an approval efficiency index; a clustering algorithm is used to analyze the feature vector to identify the main approval feature clusters of the application object; and based on the feature combination of the cluster centers, the approval strategy pattern type of the application object is determined.

[0240] In addition, computing devices can also use time-series analysis methods to analyze the changing trends of approval efficiency data over time and dynamically adjust approval strategies. This time-series analysis includes trend detection and identification of change points for indicators such as approval time, number of manual conflict confirmations, and deviation of business evaluation results.

[0241] S703. Based on the approval strategy model, generate personalized data fusion strategies for applicants.

[0242] The personalized data fusion strategy defines the execution priority and combination of various data fusion methods in the subsequent evaluation of the applicant.

[0243] Specifically, the generation of personalized data fusion strategies is based on the mapping relationship between approval strategy models and the characteristics of data fusion methods.

[0244] For example, for efficient and stable models, an efficiency-first strategy is adopted to increase the execution priority of lightweight methods such as business rules and statistical methods; for complex and controversial models, an accuracy-first strategy is adopted to ensure that all data fusion methods are executed and to strengthen the result conflict detection mechanism.

[0245] In some embodiments, the process of a computing device generating a personalized data fusion strategy can be represented as follows: the computing device queries a preset strategy template library according to the approval strategy mode to obtain a basic strategy framework; fine-tunes the parameters of the basic strategy based on the specific approval efficiency indicators of the applicant to determine the execution weight and triggering conditions of various data fusion methods; and finally generates a complete strategy configuration including method execution sequence, timeout settings and conflict handling mechanisms.

[0246] In addition, computing devices can continuously optimize personalized data fusion strategies based on the effectiveness of strategy execution through reinforcement learning mechanisms. The effectiveness of strategy execution is quantitatively evaluated using indicators such as the approval time of subsequent business processes, the number of manual conflict confirmations, and the deviation of business evaluation results.

[0247] It should be understood that by summarizing patterns based on historical approval efficiency data, personalized data fusion strategies are generated for specific applicants. This solution realizes a shift in approval strategies from a "one-size-fits-all" approach to precise customization. By dynamically optimizing the combination of methods and execution priorities, it significantly improves the processing efficiency of subsequent business while ensuring approval quality, and enhances the self-learning and continuous optimization capabilities of the approval mechanism for this application.

[0248] In some embodiments, since the approval of a business transaction may occur at different stages of the transaction, traditional solutions lack the ability to dynamically monitor the execution process. To address this issue, computing devices also possess continuous monitoring capabilities for approved transactions. In this case, the specific implementation of the computing device's monitoring function is as follows: The computing device periodically acquires the post-approval business-related data generated by the applicant during the business execution process, using a preset time period as the monitoring interval. After acquiring the data, the computing device processes the post-approval business-related data by executing S101-S104 and related implementation methods, and outputs the dynamic business security score at the current time point.

[0249] Furthermore, the dynamic business security score is compared and analyzed with the initial business approval result to calculate the business evaluation result deviation (which can be understood as the business evaluation result deviation in S701). When the deviation exceeds the preset threshold for the business type, a corresponding security warning signal is generated.

[0250] It should be understood that this solution aims to establish a closed-loop business management system covering the entire lifecycle from business approval to execution, thereby improving the overall security level of business management.

[0251] In some embodiments, based on the dynamic business security score in the above embodiments, the model parameters of the machine learning model algorithm in this embodiment can also be fed back. Specifically, the computing device uses the dynamic business security score obtained in each monitoring period and the corresponding post-approval business-related data as new training samples and incorporates them into the training dataset of the machine learning model. By retraining the preset business approval model (such as the machine learning model described in S302) using the dataset containing the new samples, continuous optimization of its model parameters can be achieved.

[0252] like Figure 9 This is a schematic diagram of a business approval device provided in an embodiment of this application. Figure 9 As shown, the business approval device includes: an acquisition module 901 and a processing module 902.

[0253] The acquisition module 901 is used to acquire the business type of the pending approval business and the business association data of the application objects associated with the pending approval business. The business association data of the application objects includes: the attribute information of the application objects, the resource status information of the application objects, and the behavior information of the application objects.

[0254] Processing module 902 is used to determine at least one target data fusion method from a preset set of data fusion methods based on the business type and preset business type-data fusion method mapping rules. The preset data fusion methods include: data fusion methods based on business rules, data fusion methods based on statistical methods, data fusion methods based on graph databases, and data fusion methods based on machine learning algorithms. Business-related data is processed using each target data fusion method to obtain the corresponding data fusion result for each method. Based on the business type and each data fusion result, the pending business is approved to obtain the approval result for the pending business.

[0255] In other embodiments, the processing module 902 is further configured to sequentially execute the target data fusion methods to process business-related data according to the data fusion order among the target data fusion methods, thereby obtaining the data fusion result corresponding to each target data fusion method. The input data of the target data fusion method that is later in the data fusion order includes at least the data fusion result output by its preceding target data fusion method.

[0256] Target data fusion methods include graph database-based data fusion methods and / or machine learning algorithm-based data fusion methods.

[0257] Processing module 902 is specifically used to identify interaction objects that have resource interaction relationships with the applicant based on business-related data. It constructs a resource interaction relationship graph using the applicant and interaction objects as nodes. The topological structure of the applicant in the resource interaction relationship graph is analyzed to obtain resource exchange relationship characteristic indicators of the applicant, which are then used as the data fusion result of the graph database-based data fusion method. These resource exchange relationship characteristic indicators characterize the stability and influence range of the applicant's resource interaction network.

[0258] The processing module 902 is specifically used to generate feature vectors based on at least one of attribute information, resource status information, and behavioral information, and / or data fusion results obtained through other data fusion methods. The business type and feature vectors are processed by a preset business approval model to obtain a business security score for the business to be approved, which serves as the data fusion result of the data fusion method based on machine learning algorithms. The business type is used to instruct the preset business approval model to adjust the weight allocation for security assessment.

[0259] In other embodiments, the processing module 902 is specifically used to perform conflict detection on each data fusion result based on preset result conflict detection rules, and obtain conflict detection results. The result conflict detection rules include contradictory combinations between different data fusion results. If at least one data fusion result combination is a contradictory combination in the conflict detection results, a conflict report is generated. In response to the conflict confirmation result input by the user based on the conflict report, a valid data fusion result is determined. The valid data fusion result includes data fusion results that did not trigger conflict rules and data fusion results that were confirmed as valid and conflicting in the conflict confirmation results. Based on the business type and the valid data fusion results, the business approval result for the business to be approved is obtained.

[0260] In other embodiments, the processing module 902 is specifically used to determine the weight value corresponding to each valid data fusion result based on the business type and a preset mapping rule set between business type and fusion result weights. The weight values ​​of each valid data fusion result are weighted and summed to obtain the business evaluation score of the business to be approved. If the business evaluation score is greater than a preset evaluation threshold, the business approval result is determined to be passed.

[0261] In other embodiments, the processing module 902 is further configured to determine that the business approval result is unsuccessful if the business evaluation score is less than or equal to a preset evaluation threshold. If the business approval result is unsuccessful, a weighted evaluation value is determined for each valid fusion result based on its corresponding weight value. The weighted evaluation value of each valid fusion result is compared with its corresponding preset single-item evaluation threshold to determine the substandard data fusion results whose weighted evaluation values ​​are less than the corresponding preset evaluation threshold. Based on the substandard data fusion results and related business data, a business approval result report is generated.

[0262] In other embodiments, the attribute information of the applicant includes: basic information of the applicant and credibility information of the applicant. The resource status information of the applicant includes: resource inventory information, available resource information, resource input information, and resource output information of the applicant. The behavioral information of the applicant includes: resource interaction event records, business request records, and business browsing records of the applicant.

[0263] In other embodiments, the processing module 902 is further configured to acquire approval efficiency data for each of the applicant's multiple evaluated business processes. The approval efficiency data for each evaluated business process includes: approval time, number of manual conflict confirmations, and deviation of business evaluation results. The deviation of business evaluation results measures the degree of difference between the business evaluation results of the evaluated business and the applicant's actual business behavior. Pattern summarization is performed on the approval efficiency data for each of the multiple evaluated business processes to determine the applicant's approval strategy pattern. Based on the approval strategy pattern, a personalized data fusion strategy is generated for the applicant. The personalized data fusion strategy defines the execution priority and combination method of various data fusion methods in the subsequent evaluation processes for the applicant.

[0264] The business approval device provided in this application embodiment can execute the method shown in the above method embodiment. Its implementation principle and beneficial effects can be referred to the relevant description in the method embodiment, and will not be repeated here.

[0265] Figure 10 This is a schematic diagram of a business approval device provided in an embodiment of this application. Figure 10 As shown, the business approval device includes: a memory 1001, a transceiver 1002, and at least one processor 1003.

[0266] Transceiver 1002 is used to interact with other devices to send and receive data.

[0267] For example, in this embodiment of the application, the transceiver 1002 can be used to obtain the business type of the business to be approved and the business association data of the application object associated with the business to be approved, or to send the business approval result to the terminal device of the application object.

[0268] The memory 1001 stores computer program code, which includes computer instructions. These computer instructions run in the aforementioned business approval device to implement the method described in the above-described method embodiments. For example, the memory may include high-speed random access memory (RAM), and may also include non-volatile memory (NVM), such as at least one disk storage device, or a USB flash drive, portable hard drive, read-only memory, disk, or optical disc, etc.

[0269] Processor 1003 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. Processor 1003 can also be other general-purpose processors. The general-purpose processor can be a microprocessor or any conventional processor.

[0270] The memory 1001, transceiver 1002, and processor 1003 are communicatively connected. For example, the memory 1001 and transceiver 1002 can be connected to the processor 1003 via a system bus to complete mutual communication. The system bus can be a peripheral component interconnect (PCI) bus, an extended industry standard architecture (EISA) bus, an industry standard architecture (ISA) bus, etc. The system bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the figure, but this does not mean that there is only one bus or one type of bus.

[0271] Optionally, the memory 1001 can be either independent or integrated with the processor 1003. When the memory 1001 is configured independently, it is connected to the processor 1003 via a system bus.

[0272] This application also provides a chip for executing instructions, which is used to perform the business approval method described in the above embodiments.

[0273] This application also provides a computer-readable storage medium storing computer instructions. When these computer instructions are executed by a processor, they are used to implement the technical solution of the business approval method described in the above embodiments. Specifically, when the computer instructions are executed by a processor, the business approval device can execute the technical solution of the business approval method described in the above embodiments.

[0274] This application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium, and when the at least one processor executes the computer program, it can implement the technical solution of the business approval method in the above embodiments.

[0275] The aforementioned computer-readable storage media can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as static random-access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The computer-readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0276] An exemplary computer-readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Of course, the computer-readable storage medium can also be a component of the processor. The processor and the computer-readable storage medium can reside in application-specific integrated circuits (ASICs). Alternatively, the processor and the computer-readable storage medium can exist as discrete components in an electronic control unit or main control device; this application does not limit this.

[0277] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.

[0278] The modules described as separate components may or may not be physically separate. The components shown as modules 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 modules can be selected to implement the solution of this embodiment according to actual needs.

[0279] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The unit composed of the above modules can be implemented in hardware or in the form of hardware plus software functional units.

[0280] The integrated modules described above, implemented as software functional modules, can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods of the various embodiments of this application.

[0281] It should be understood that the steps of the method disclosed in the embodiments of this application can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules in the processor.

[0282] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0283] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A business approval method, characterized in that, include: Obtain the business types of the pending approval business and the business association data of the application objects associated with the pending approval business; The business-related data of the applicant includes: the applicant's attribute information, the applicant's resource status information, and the applicant's behavior information; Based on the business types and the preset business type-data fusion method mapping rules, at least one target data fusion method is determined from the preset data fusion methods; the preset data fusion methods include: data fusion methods based on business rules, data fusion methods based on statistical methods, data fusion methods based on graph databases, and data fusion methods based on machine learning algorithms. The business-related data is processed by each of the target data fusion methods to obtain the data fusion result corresponding to each target data fusion method. Based on the business type and each of the data fusion results, the pending business is approved to obtain the business approval result of the pending business.

2. The method according to claim 1, characterized in that, The process of processing the business-related data through each of the target data fusion methods to obtain the data fusion result corresponding to each target data fusion method includes: According to the data fusion order among the target data fusion methods, the target data fusion methods are executed sequentially to process the business-related data, thereby obtaining the data fusion result corresponding to each target data fusion method; wherein, the input data of the target data fusion method ranked later in the data fusion order includes at least the data fusion result output by the previous target data fusion method; The target data fusion method includes the graph database-based data fusion method and / or the machine learning algorithm-based data fusion method; The data fusion method based on graph database includes the following steps: Based on the business-related data, identify the interaction objects that have resource interaction relationships with the applicant. Using the application object and the interaction object as nodes, construct a resource interaction relationship graph; The topological structure of the applicant in the resource interaction relationship graph is analyzed, and the resource exchange relationship characteristic index of the applicant is obtained as the data fusion result of the graph database-based data fusion method; the resource exchange relationship characteristic index characterizes the stability and influence range of the applicant's resource interaction network. The data fusion method based on machine learning algorithms includes the following steps: A feature vector is generated based on at least one of the attribute information, the resource status information, and the behavioral information, and / or the data fusion result obtained through other data fusion methods. The business type and the feature vector are processed by a preset business approval model to obtain the business security score of the business to be approved as the data fusion result of the data fusion method based on machine learning algorithm; the business type is used to instruct the preset business approval model to adjust the weight allocation of the security assessment.

3. The method according to claim 1, characterized in that, The process of approving the pending business based on the business type and each data fusion result to obtain the business approval result for the pending business includes: Based on a preset result conflict detection rule, conflict detection is performed on each of the data fusion results to obtain a conflict detection result; the result conflict detection rule includes contradictory combinations between different data fusion results. If at least one of the data fusion results in the conflict detection results is a contradictory combination, a conflict report is generated. In response to the conflict confirmation result input by the human, a valid data fusion result is determined; wherein, the valid data fusion result includes the data fusion result that has not triggered the conflict rule and the data fusion result that has been confirmed as valid and has a conflict in the conflict confirmation result; Based on the business type and the effective data fusion results, the business approval result of the business to be approved is obtained.

4. The method according to claim 3, characterized in that, The process of obtaining the business approval result for the pending business based on the business type and the effective data fusion result includes: Based on the business types and the preset mapping rule set between business types and fusion result weights, determine the weight value corresponding to each of the effective data fusion results; The weighted sum of the weight values ​​of each valid data fusion result is used to obtain the business evaluation score of the business to be approved; If the business evaluation score is greater than the preset evaluation threshold, the business approval result is determined to be passed.

5. The method according to claim 4, characterized in that, The method further includes: If the business evaluation score is less than or equal to a preset evaluation threshold, the business approval result is determined to be unsuccessful. If the business approval result is not approved, the weighted evaluation value of each valid fusion result is determined by the weight value corresponding to each valid fusion result. The weighted evaluation value of each effective fusion result is compared with its corresponding preset single evaluation threshold to determine the substandard data fusion results whose weighted evaluation value is less than the corresponding preset evaluation threshold; Based on the fusion results of the non-compliant data and the business-related data, a business approval result report is generated.

6. The method according to claim 1, characterized in that, The attribute information of the application object includes: the basic information of the application object and the credibility information of the application object; The resource status information of the applicant includes: the applicant's resource inventory information, the applicant's available resources information, the applicant's resource input information, and the applicant's resource output information; The behavioral information of the applicant includes: the applicant's resource interaction event records, the applicant's business request records, and the applicant's business browsing records.

7. The method according to any one of claims 1-6, characterized in that, The method further includes: Obtain approval efficiency data for each of the applicant's multiple evaluated business processes; the approval efficiency data for each evaluated business process includes: approval time, number of manual conflict confirmations, and business evaluation result deviation; the business evaluation result deviation is used to measure the degree of difference between the business evaluation result of the evaluated business process and the applicant's actual business behavior; By summarizing the approval efficiency data of the various evaluated business processes, the approval strategy pattern of the applicant is determined. Based on the approval strategy model, a personalized data fusion strategy is generated for the applicant; the personalized data fusion strategy defines the execution priority and combination of various data fusion methods in the subsequent evaluation business for the applicant.

8. A business approval device, characterized in that, include: The acquisition module is used to acquire the business type of the business to be approved and the business association data of the application object associated with the business to be approved; The business-related data of the applicant includes: the applicant's attribute information, the applicant's resource status information, and the applicant's behavior information; The processing module is used to determine at least one target data fusion method from a preset set of data fusion methods based on the business type and preset business type and data fusion method mapping rules; the preset data fusion methods include: a data fusion method based on business rules, a data fusion method based on statistical methods, a data fusion method based on graph databases, and a data fusion method based on machine learning algorithms; process the business-related data through each target data fusion method to obtain the data fusion result corresponding to each target data fusion method; and perform business approval on the business to be approved based on the business type and each data fusion result to obtain the business approval result of the business to be approved.

9. A business approval device, characterized in that, include: A memory and at least one processor; the memory is communicatively connected to the processor; the memory is used to store computer program code, the computer program code including computer instructions; when the processor executes the computer instructions, it causes the business approval device to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, are used to implement the method as described in any one of claims 1-7.

11. A computer program product, characterized in that, When the computer program product is run on a computer / executed by the computer's processor, it implements the method as described in any one of claims 1-7.

Images