A multi-service line index library construction method and system

Abstract

The application discloses a kind of multi-service line index library construction method and system, method includes: data acquisition and pre-processing, to the cleaning and standardization of multi-source heterogeneous data;Index screening, effective index is screened based on correlation analysis and the significance test algorithm of adaptive selection chi-square test or t test;Weight configuration, generate comprehensive weight using combination weighting algorithm, and dynamically adjust through business scene adaptation factor;Index iterative management, according to the rigid quantization rule based on continuous multiple period statistical verification result to index execution new start, disable archiving and version trace management;Business synergy linkage, based on unified associated field, the data link of index and multiple business modules is established, and the closed loop of "index trigger-business response-data backflow-index optimization" is formed, the beneficial effects of the application are that through multi-algorithm fusion driving and full life cycle regulation control, the high effectiveness, dynamic adaptability and deep synergy with business of index system are realized.

Description

Technical Field

[0001] This invention relates to the field of financial big data analysis and risk control technology, and in particular to a method and system for constructing a multi-business line indicator library. Background Technology

[0002] In existing technologies, accurate assessment of debtors' repayment ability is a core element in optimizing resource allocation and improving capital recovery efficiency in the disposal and mediation of non-performing assets by financial institutions. The accuracy of this assessment heavily relies on a scientific, comprehensive, and dynamically adaptable indicator system. However, current industry practices for managing and applying indicators to assess debtors' repayment ability generally suffer from the following core technical deficiencies: 1. Lack of uniformity in indicator system: Different business lines use independent evaluation indicators, with differences in indicator definitions, statistical methods, and weight allocation. The lack of a unified standard across business lines makes it impossible to compare evaluation results horizontally, making it difficult to achieve overall risk control and resource allocation.

[0003] 2. The selection and optimization of indicators are highly subjective: Existing indicators mostly rely on manual experience for screening, and the correlation between indicators and collection rate has not been verified through scientific statistical analysis. There are problems such as redundant and ineffective indicators and missing key indicators. At the same time, the indicator weights are fixed and cannot be dynamically adjusted according to business changes and market fluctuations. The effect will decay after long-term use.

[0004] 3. Non-standardized indicator management process: There is a lack of standardized process for adding, modifying and deactivating indicators. The relationship between indicators and evaluation models and business systems is chaotic. The iteration history cannot be traced, which is not conducive to model version management and compliance audit.

[0005] 4. Poor data adaptability and disconnect from business collaboration: The data sources of various business lines are heterogeneous and have different formats, which makes it difficult to collect indicator data and the quality is inconsistent. More importantly, the indicator library often exists only as an isolated basic data set and has failed to establish a deep and automated linkage mechanism with core business modules such as customer segmentation, adjustment strategy matching, and risk warning. As a result, the analytical value of the indicators cannot be efficiently transformed into actual business effectiveness.

[0006] While some existing related technologies involve indicator management, they are mostly limited to static storage and simple queries. Their technical shortcomings are: single and fixed algorithms, lacking the ability to filter and dynamically optimize driven by the fusion of multiple algorithms; lack of a standardized and quantifiable indicator control rule system; and high system architecture coupling, making it difficult to adapt to the differentiated needs of multiple business lines and the requirements of high-frequency iteration.

[0007] Therefore, there is an urgent need for a method and system for constructing an indicator library that can systematically solve the above problems and achieve scientific selection, dynamic optimization, standardized management throughout the entire life cycle of indicators, and deep collaboration with business operations. Summary of the Invention

[0008] The purpose of this invention is to provide a method and system for constructing a multi-business line indicator library.

[0009] To achieve the above objectives, the technical solution proposed by this invention is as follows: A method for constructing a multi-business line indicator library includes the following steps: S1. Data Acquisition and Preprocessing Steps: Collect raw indicator data from multiple heterogeneous data sources, and perform data cleaning and standardization on the raw indicator data to form standardized indicator data.

[0010] S2. Indicator screening steps: Calculate the standardized indicator data based on a preset statistical analysis algorithm, screen out effective indicators that are significantly related to the target variable, and eliminate redundant indicators; the statistical analysis algorithm includes at least correlation analysis, significance test and discrimination verification.

[0011] S3. Weight Configuration Steps: A combined weighting algorithm that integrates subjective business weights and objective data weights is used to assign comprehensive weights to each of the selected effective indicators. The combined weighting algorithm combines the Analytic Hierarchy Process (AHP) and the entropy weight method, and is dynamically adjusted according to the business scenario adaptation factor.

[0012] S4. Indicator Iteration Management Steps: Based on the preset rigid quantification rules, the effective indicators and their weights are managed throughout their entire lifecycle. The rigid quantification rules define the conditions for adding, discontinuing, and archiving indicators, and record the complete version iteration history.

[0013] S5. Business Collaboration and Linkage Steps: Establish a data association and closed-loop feedback mechanism between the indicators and multiple business modules, so that changes in the values ​​of the indicators trigger automatic adjustments to business strategies, and business execution effect data flows back to the indicator library to drive the dynamic optimization of the indicators and the comprehensive weights.

[0014] Step S1, the data acquisition and preprocessing steps, further includes: Data is collected through two modes: real-time API capture and batch offline file import.

[0015] The data cleaning process includes primary cleaning and secondary cleaning. Primary cleaning fills in missing values ​​using the median or default values ​​for the business scenario. Secondary cleaning identifies and marks outliers by combining the 3σ principle with box plots.

[0016] Step S2, the significance test in the indicator screening step, further includes: To determine the data type of the indicator, a chi-square test is used for categorical indicators and a t-test is used for continuous numerical indicators. The p-value obtained from the test is compared with the preset significance level threshold to determine whether the indicator is significantly effective.

[0017] Step S3, the combined weighting algorithm in the weight configuration step, further includes: The subjective weights of each indicator are calculated using the Analytic Hierarchy Process (AHP), and the objective weights of each indicator are calculated using the entropy weight method.

[0018] The comprehensive weight is calculated using a linear combination formula and a preset weight balance coefficient.

[0019] An adjustment factor adapted to the business scenario is introduced to dynamically fine-tune the comprehensive weight.

[0020] Step S4, the rigid quantification rules in the indicator iteration management step, further include: Newly added indicators must pass the validity verification within the first preset time period before they can be officially activated.

[0021] For existing indicators whose correlation coefficient is lower than the first preset threshold and whose significance test p-value is higher than the second preset threshold for consecutive periods of the second preset time period, they are automatically marked as pending deactivation and deactivation or archiving is performed after the buffer period.

[0022] Any change to any metric generates a corresponding version record to support full lifecycle audit traceability.

[0023] Step S5, the business collaboration and linkage steps, further include: Based on standardized association fields, data links are established between the indicators and the customer segmentation module, mediation strategy matching module, risk warning module, case management module, and communication record module.

[0024] Automatically trigger and match corresponding business strategies based on preset combination of indicator thresholds.

[0025] The system receives business execution performance data from the business modules and uses it as input data for a new round of indicator effectiveness evaluation and weight adjustment, forming a closed loop of "indicator triggering - business response - data feedback - indicator optimization".

[0026] A multi-business line indicator library construction system, comprising: The data acquisition and preprocessing module is used to collect raw indicator data from multiple heterogeneous data sources and perform data cleaning and standardization on the raw indicator data to form standardized indicator data.

[0027] The indicator screening module is used to calculate the standardized indicator data based on a preset statistical analysis algorithm, screen out effective indicators that are significantly related to the target variable, and eliminate redundant indicators; the statistical analysis algorithm includes at least correlation analysis, significance test and discrimination verification.

[0028] The weight configuration module is used to assign comprehensive weights to each of the selected effective indicators by adopting a combined weighting algorithm that integrates subjective business weights and objective data weights. The combined weighting algorithm combines the Analytic Hierarchy Process (AHP) and the entropy weight method, and is dynamically adjusted according to the business scenario adaptation factor.

[0029] The indicator iteration management module is used to manage the effective indicators and their weights throughout their entire lifecycle according to preset rigid quantification rules. The rigid quantification rules define the conditions for adding, disabling, and archiving indicators, and record the complete version iteration history.

[0030] The business collaboration module is used to establish a data association and closed-loop feedback mechanism between the indicator and multiple business modules, so that changes in the value of the indicator trigger automatic adjustments to the business strategy, and the business execution effect data flows back to the indicator library to drive the dynamic optimization of the indicator and the comprehensive weight.

[0031] The weight configuration module works in conjunction with the indicator iteration management module, specifically for: After each performance indicator iteration evaluation, based on the latest business performance data, the combined weighting algorithm is called again to update the comprehensive weight of each effective indicator, and the updated weight is synchronized to the business service layer through a hot update mechanism without restarting the system service.

[0032] The indicator iteration management module is further used for: Store and maintain a metadata database containing indicator definitions, statistical calibers, calculation rules, data sources, validity verification methods, and historical version information.

[0033] The system provides a visual management interface for configuring the metadata database and the rigid quantification rules, enabling differentiated adaptation and management of indicators for different business lines.

[0034] The system adopts a hierarchical loosely coupled architecture, which includes, from top to bottom: The front-end configuration layer provides a visual interaction and configuration interface.

[0035] The business service layer employs a microservice architecture to deploy the indicator filtering module, weight configuration module, and business collaboration module.

[0036] The data processing layer integrates a distributed computing framework to execute the statistical analysis algorithms.

[0037] The data storage layer adopts a hybrid storage architecture, in which structured indicator metadata and weight configuration data are stored in a relational database, while frequently accessed configuration data is cached in an in-memory database.

[0038] The beneficial effects of this invention are: 1. High effectiveness and dynamic adaptability: Through the screening and weighting mechanism of multi-algorithm fusion, the objectivity, effectiveness and scientific nature of the indicators are guaranteed. At the same time, the closed-loop feedback and hot update mechanism enable the indicator system to dynamically adapt to business changes and avoid the decay problem of traditional static models.

[0039] 2. Standardized and traceable management: The system achieves automated management of indicators throughout their entire lifecycle through rigid quantitative rules. From addition and discontinuation to archiving, everything is traceable. The complete version management function greatly improves the standardization of model risk management and the convenience of auditing.

[0040] 3. Deep Business Collaboration and Efficiency Transformation: By establishing a closed-loop linkage between indicators and multiple business modules, data analysis results are seamlessly embedded into business processes, realizing an automated flow from "data insight" to "business action" and then to "effect feedback," significantly enhancing the ultimate business value of indicators.

[0041] 4. High availability and scalability: Adopting a layered, loosely coupled microservice architecture, combined with distributed computing and hybrid storage technologies, the system has high concurrency processing capabilities, high availability, and good horizontal scalability, enabling it to smoothly adapt to business needs of different scales. Attached Figure Description

[0042] Figure 1 This is a flowchart illustrating the method for constructing a multi-business line indicator library according to Embodiment 1 of the present invention. Figure 2 This is a flowchart illustrating the multi-business line indicator library construction system according to Embodiment 1 of the present invention; Figure 3 This is a detailed algorithm flowchart of the indicator screening step in Embodiment 2 of the present invention. Detailed Implementation

[0043] The present invention will now be described in further detail with reference to the accompanying drawings. Example 1:

[0044] A method for constructing a multi-business line indicator library includes the following steps: S1. Data Acquisition and Preprocessing Steps: Collect raw indicator data from multiple heterogeneous data sources, and perform data cleaning and standardization on the raw indicator data to form standardized indicator data.

[0045] S2. Indicator screening steps: Based on the preset statistical analysis algorithm, the standardized indicator data is calculated to screen out effective indicators that are significantly related to the target variable and eliminate redundant indicators; the statistical analysis algorithm includes at least correlation analysis, significance test and discrimination verification.

[0046] S3. Weighting configuration steps: A combined weighting algorithm that integrates subjective business weights and objective data weights is used to assign comprehensive weights to each selected effective indicator. The combined weighting algorithm combines the Analytic Hierarchy Process (AHP) and the entropy weighting method, and is dynamically adjusted according to the business scenario adaptation factor.

[0047] S4. Indicator Iteration Management Steps: Based on the preset rigid quantification rules, the effective indicators and their weights are managed throughout their entire lifecycle. The rigid quantification rules define the conditions for adding, discontinuing, and archiving indicators, and record the complete version iteration history.

[0048] S5. Business Collaboration and Linkage Steps: Establish a data association and closed-loop feedback mechanism between indicators and multiple business modules, so that changes in indicator values ​​trigger automatic adjustments to business strategies, and business execution effect data flows back to the indicator library to drive the dynamic optimization of indicators and comprehensive weights.

[0049] The flowchart of the method for constructing a multi-business line indicator library in Embodiment 1 of the present invention is shown below. Figure 1 As shown.

[0050] Step S1, the data acquisition and preprocessing steps, further includes: Data is collected through two modes: real-time API capture and batch offline file import.

[0051] Data cleaning includes primary cleaning and secondary cleaning. Primary cleaning fills in missing values ​​using the median or default values ​​for the business scenario. Secondary cleaning identifies and marks outliers by combining the 3σ principle with box plots.

[0052] Step S2, the significance test in the indicator screening step, further includes: Calculate the Pearson correlation coefficient between each indicator and the target variable. For categorical indicators, construct a contingency table and perform the chi-square test to obtain a first p-value. For continuous numerical indicators, divide the sample into two groups according to the median and perform the t-test to obtain a second p-value. If the absolute value of the Pearson correlation coefficient is greater than or equal to a first preset threshold, and the first p-value or the second p-value is less than a second preset threshold, the corresponding indicator is determined to be a valid indicator.

[0053] The formula for calculating the Pearson correlation coefficient is as follows: , It is used to initially measure the linear correlation between the indicator and the target variable, namely the collection rate, where r is the Pearson correlation coefficient, ranging from [-1, 1], used to measure the degree of linear correlation between the indicator and the target variable, n is the sample size, i.e., the number of debtor cases involved in the calculation, and X... i Let i be the value of the i-th sample on a certain indicator. The mean of this indicator across all samples, Y i For the first i The values ​​of each sample in the target variable, such as the collection rate. The target variable is the mean of all samples.

[0054] The chi-square test formula is as follows: , Wherein, χ² is the chi-square statistic, used to measure the degree of association between the categorical indicator and the target variable. The observation frequency is the value in the i-th row and j-th column of the contingency table, which is the actual number of samples. Let be the expected frequency in the i-th row and j-th column of the contingency table, which is the theoretical sample size under the null hypothesis.

[0055] The t-test formula is the Welch t-test formula, and the formula for unequal variances is: , Where t is the t-statistic, used to measure whether there is a significant difference in the mean of the target variable between two groups of samples grouped by the median of the indicator. The first sample group represents the mean of the target variable, such as the sample group whose index value is less than or equal to the median. The second group of samples represents the mean of the target variable, such as the sample group whose index value is greater than the median. The variance of the first group of samples on the target variable. The variance of the second group of samples on the target variable. This represents the sample size of the first group. This represents the number of samples in the second group.

[0056] Cohen's effect size formula: , Where d is Cohen's d effect size, used to measure the actual discriminative power of a continuous indicator against the target variable. The mean of the first group of samples on the target variable, such as the high-value group. The mean of the second group of samples on the target variable, such as the low indicator value group. S pooledLet n be the pooled standard deviation of the two groups of samples, and n1 and n2 be the sample sizes of the first and second groups, respectively. , These are the variances of the first and second groups of samples on the target variable, respectively.

[0057] Step S3, the combined weighting algorithm in the weight configuration step, further includes: The subjective weights of each indicator are calculated using the Analytic Hierarchy Process (AHP), and the objective weights of each indicator are calculated using the entropy weight method.

[0058] The comprehensive weight is calculated using a linear combination formula and a preset weight balance coefficient.

[0059] An adjustment factor adapted to the business scenario is introduced to dynamically fine-tune the comprehensive weight.

[0060] Among them, the subjective weight calculation of AHP (Analytic Hierarchy Process): Judgment matrix and consistency test, constructing the judgment matrix ,in, Calculate the eigenvectors and the largest eigenvalue lambda. {max} Consistency check: , RI This is a random consistency index, obtained by looking up a table based on the matrix order n; the consistency ratio. Subjective weight vector .

[0061] in, A To determine the matrix, a {ij} Indicators i relative to indicators j Importance scale lambda {max} To determine the largest eigenvalue of a matrix, CI As a consistency indicator, CR For the consistency ratio, when CR < 0.1 At that time, it is considered that the judgment matrix passes the one-time test. The subjective weight vector is obtained by normalizing the eigenvector corresponding to the largest eigenvalue of the judgment matrix.

[0062] Among them, the objective weight calculation of the entropy weight method is as follows: Data standardization: a positive indicator , negative indicators , Calculate the first i The first item under the indicator j The proportion of each sample , Calculate information entropy ,in , Calculate objective weights , in, X {ij} For the first j The original value of each sample on the i-th indicator X {max} , X {min} Let these represent the maximum and minimum values ​​of the i-th indicator in all samples, respectively. N {ij} For the first j The standardized value of a sample on the i-th indicator, p {ij} For the first i The first item under the indicator j The proportion of each sample m The total number of samples, k It is a constant. E i Let i be the information entropy of the i-th indicator. For the first i The objective weight of each indicator n This represents the total number of indicators.

[0063] The formula for adjusting the business scenario adaptation factor is as follows: , in, This represents the final weight of the i-th metric after business scenario adaptation. W i Let β be the overall weight of the i-th indicator. k For the first k The business scenario adaptation factor corresponding to the business line is used to fine-tune the weights based on business characteristics, and its value range is... 0.9≤β≤1.1 .

[0064] The algorithm employs a combination of the Analytic Hierarchy Process (AHP) and entropy weighting to assign weights, balancing business experience with data objectivity to enhance the scientific nature of weight allocation. It also introduces dynamic adjustment factors to adapt to business changes. The core logic of the algorithm is as follows: First, using AHP, a three-tiered hierarchical structure is constructed: target layer (repayment ability assessment), criteria layer (four major indicator categories), and indicator layer. Five to eight business experts are invited to score and construct a judgment matrix. Subjective weights are calculated after a consistency test (CR < 0.1). Judgment matrices that fail the test are automatically prompted for expert correction. Second, using entropy weighting, the information entropy of each indicator is calculated based on historical data from the past six months. A smaller entropy value indicates higher indicator discrimination and a larger weight, yielding objective weights. Third, the algorithm uses linear... The combined formula yields a comprehensive weight, with subjective weight accounting for 40% and objective weight accounting for 60%, balancing business needs and data patterns. It also incorporates business scenario adjustment factors to adapt to the characteristics of different business lines. The indicator library rules impose constraints: for highly correlated indicators (comprehensive correlation coefficient ≥ 0.6), the weight is forcibly increased by 5%-10%; for low-correlation indicators (comprehensive correlation coefficient < 0.2), the weight is decreased by 20% or marked as pending observation. The upper limit for a single indicator weight is ≤ 0.3, and the lower limit is ≥ 0.05. The total weight is strictly 1, with real-time verification during configuration and automatic warnings for exceeding the range. It supports differentiated adjustments based on business lines, with an adjustment range of ±5%. Adjustments are automatically synchronized to the scoring model, and adjustment records are permanently archived for traceability.

[0065] Step S4, the rigid quantification rules in the indicator iteration management step, further include: New indicator activation conditions: Candidate indicators are officially activated only if, within a consecutive first preset number of evaluation periods, their correlation coefficient is greater than or equal to the first preset threshold and their significance test p-value is less than the second preset threshold.

[0066] Indicator deactivation conditions: If an activated effective indicator has a correlation coefficient less than a third preset threshold and a significance test p-value greater than or equal to the second preset threshold within a consecutive second preset number of evaluation periods, it will be automatically marked as pending deactivation and will be deactivated or archived after the preset buffer period expires.

[0067] The system establishes rigid iteration rules to ensure a streamlined and effective indicator library, with rules that are irreplaceable and compliant: new indicators must undergo a complete screening process, including preprocessing, correlation analysis, significance testing, and discrimination verification, and must pass verification for two consecutive months before being officially activated. An initial version is automatically generated upon activation. Indicators with a comprehensive correlation coefficient <0.2, p-value ≥0.05, and Cohen's d <0.5 for three consecutive months are automatically marked as "pending discontinuation." A one-month buffer period is required before discontinuation, and discontinuation or archiving is implemented after double verification. Archived indicators retain complete historical versions and data links, cannot be deleted, support full lifecycle traceability, and comply with financial compliance requirements. A full review of all indicators is conducted quarterly, and the indicator system is restructured annually. The results of the review and restructuring must be included in the compliance audit scope. During indicator iteration, if changes in statistical definitions are involved, a change description must be generated and linked to historical data to ensure data consistency.

[0068] Step S5, the business collaboration and linkage steps, further include: The unified association fields should include at least customer identifier, case identifier, and business line identifier; In response to the current value of a valid indicator satisfying a preset combination of policy triggering conditions, a policy execution instruction is sent to the corresponding business module. Receive business execution effect data returned by the business module after the strategy is executed. The business execution effect data includes at least the payment conversion rate, customer response status or risk handling results. The business execution performance data is associated with and stored with the corresponding historical snapshots of the metrics, and used as input data for re-executing the metric selection and weight configuration steps.

[0069] A multi-business line indicator library construction system, comprising: The data acquisition and preprocessing module is used to collect raw indicator data from multiple heterogeneous data sources and perform data cleaning and standardization on the raw indicator data to form standardized indicator data.

[0070] The indicator screening module is used to calculate standardized indicator data based on a preset statistical analysis algorithm, screen out effective indicators that are significantly related to the target variable, and eliminate redundant indicators; the statistical analysis algorithm includes at least correlation analysis, significance test and discrimination verification.

[0071] The weight configuration module is used to assign comprehensive weights to each selected effective indicator by employing a combined weighting algorithm that integrates subjective business weights and objective data weights. The combined weighting algorithm combines the Analytic Hierarchy Process (AHP) and the entropy weight method, and is dynamically adjusted according to the business scenario adaptation factor.

[0072] The dynamic adjustment formula is as follows: Significance test, If p < 0.05, the indicator is significant. If p ≥ 0.05, the indicator is not significant.

[0073] in, For the first i The weight values ​​of the indicators before adjustment For the first i The adjusted weight values ​​of the indicators p Obtained for significance test p The value is used to measure the statistical significance of the association between the indicator and the target variable. To ensure that the weight adjustment does not exceed the preset upper limit of 1.0, To ensure that the weight adjustment does not fall below the preset lower limit of 0.0.

[0074] The indicator iteration management module is used to manage the entire lifecycle of effective indicators and their weights according to preset rigid quantification rules. The rigid quantification rules define the conditions for adding, disabling, and archiving indicators, and record the complete version iteration history.

[0075] The business collaboration module is used to establish data association and closed-loop feedback mechanism between indicators and multiple business modules, so that changes in indicator values ​​trigger automatic adjustments to business strategies, and business execution effect data flows back to the indicator library to drive dynamic optimization of indicators and comprehensive weights.

[0076] The indicator screening module is further used to: calculate the Pearson correlation coefficient between each indicator and the target variable; for categorical indicators, construct a contingency table and perform a chi-square test to obtain a first p-value; for continuous numerical indicators, divide the sample into two groups according to the median and perform a t-test to obtain a second p-value; in response to the absolute value of the Pearson correlation coefficient being greater than or equal to a first preset threshold, and the first p-value or the second p-value being less than a second preset threshold, determine the corresponding indicator as a valid indicator.

[0077] The weight configuration module and the indicator iteration management module work together, specifically for: After each performance indicator iteration evaluation, based on the latest business performance data, the combined weighting algorithm is called again to update the comprehensive weight of each effective indicator, and the updated weight is synchronized to the business service layer through a hot update mechanism without restarting the system service.

[0078] The combined weighted linear fusion formula is as follows: , Among them, W i Let be the comprehensive weight of the i-th indicator, and be the result after integrating subjective and objective weights. Let be the subjective weight of the i-th indicator calculated using the Analytic Hierarchy Process (AHP). For the first i The objective weights of the indicators are calculated using the entropy weight method. α is a preset fusion coefficient used to adjust the relative importance of subjective weights and objective weights, and its value range is 0.4≤α≤0.6.

[0079] The indicator iteration management module is further used for: Store and maintain a metadata database containing indicator definitions, statistical calibers, calculation rules, data sources, validity verification methods, and historical version information.

[0080] The system provides a visual management interface for configuring the metadata database and rigid quantification rules, enabling differentiated adaptation and management of indicators for different business lines.

[0081] New indicators are activated only if, within a consecutive first preset number of evaluation periods, the candidate indicator meets the following conditions: the correlation coefficient is greater than or equal to the first preset threshold and the p-value of the significance test is less than the second preset threshold.

[0082] Indicator Deactivation: If an activated effective indicator has a correlation coefficient less than the third preset threshold and a significance test p-value greater than or equal to the second preset threshold for a consecutive second preset number of evaluation periods, it will be automatically marked as pending deactivation and will be deactivated or archived after the preset buffer period expires.

[0083] The system adopts a layered, loosely coupled architecture, which includes, from top to bottom: The front-end configuration layer provides an interface for configuring metrics, rules, and a version visualization interface.

[0084] The business service layer adopts a microservice architecture to deploy the indicator filtering module, weight configuration module, indicator iteration management module, and business collaboration module. The modules communicate with each other through a service registration and discovery mechanism.

[0085] The data processing layer integrates a distributed computing framework to perform batch computing tasks using various statistical analysis algorithms.

[0086] The data storage layer adopts a hybrid storage architecture, in which structured indicator metadata and weight configuration data are stored in a relational database, while frequently accessed configuration data is cached in an in-memory database.

[0087] The flowchart of the multi-business line indicator library construction system of Embodiment 1 of the present invention is shown below. Figure 2 As shown.

[0088] The front-end configuration layer is developed based on the Vue.js 3.0 framework and uses the ARCO DESIGN component library to encapsulate the interface into components, reducing development and maintenance costs. It integrates ECharts 5.0 visualization components and uses visualization forms such as heatmaps, histograms, and tree diagrams to display the correlation of indicators, weight distribution, application effects, and version iteration trajectory in real time. It supports responsive layout technology and is compatible with multiple terminals such as computers and tablets. The operation interface adopts drag-and-drop configuration logic to realize the visualization operation of indicator classification, weight adjustment, and other functions. The core interaction response time is ≤300ms to ensure smooth operation.

[0089] The business service layer is developed based on the Spring Boot 2.7 framework and adopts a microservice architecture to split into six core modules. Each module implements service registration and discovery through Spring Cloud Alibaba, supports elastic scaling and failover, uses the Feign component to realize cross-module and interface calls with external systems, adopts the Dubbo protocol to improve the efficiency of inter-service communication, and achieves an interface call success rate of ≥99.9%. Spring Security is introduced to implement interface permission control, and identity authentication is completed based on JWT tokens to ensure the security of core operations such as indicator configuration and version management. At the same time, operation logs are recorded through AOP aspect technology to support compliance auditing.

[0090] The data processing layer is the core technology hub of the system, responsible for the collection, cleaning, transformation, feature extraction and statistical analysis of indicator data. It integrates four core algorithm modules to build an irreplaceable algorithm engine barrier. It adopts the Spring JPA framework to implement ORM mapping, supports dynamic SQL generation and batch data operation, and improves data query efficiency by more than 30% compared with traditional JDBC. It integrates the Apache Spark computing framework to realize distributed processing of large-scale business data and support computationally intensive tasks such as indicator screening and correlation analysis.

[0091] The data storage layer employs a hybrid storage architecture. Structured data is stored in a MySQL 8.0 database, with master-slave replication technology enabling data backup. The slave database latency is ≤1 second. A table and database sharding strategy is also introduced, with tables partitioned by business line to improve the efficiency of large-scale data queries. High-frequency access data, such as commonly used indicator weights and business line configurations, is cached in Redis 6.0 using a String+Hash data structure, with a 30-minute expiration time and an active update mechanism, achieving a cache hit rate of ≥90% and a data access response time of ≤50ms. Unstructured data, such as indicator documents and analysis reports, is stored in the OSS distributed storage service, supporting breakpoint resume and version control. CDN accelerates static resource access, ensuring efficient document loading.

[0092] Each indicator contains complete metadata attributes. The indicator library rules run through the entire process of metadata definition, verification, association, and iteration, forming a closed-loop standardized system to ensure that indicators are standardized, traceable, and reusable. At the same time, it is compatible with the existing system data model and establishes a rigid association between indicators and entities such as cases and organizations. This rule system is one of the core technical barriers. It eliminates cross-business line indicator differences through clear and unified rules, while supporting differentiated expansion.

[0093] Among them, hierarchical rules are linked, and preset hierarchical rules are used, for example:

[0094] The tiered results are synchronized to the customer management system. High-capital customers are given priority in receiving premium service resources, such as recommendations for low-interest loan renewals. Medium-capital customers are matched with standard mediation strategies. Low-capital customers are subject to risk mitigation measures, such as asset verification.

[0095] Customer performance data after segmentation, such as the repayment rate of high-capability customers and the loss of contact rate of low-capability customers, are periodically fed back into the indicator library to adjust indicator weights. For example, if the "loss of contact label" has a high accuracy rate in identifying low-capability customers, its weight can be increased.

[0096] Layering rules can be modified through a visual configuration on the front end. The formula for comprehensively scoring a customer's repayment ability is as follows: , Where S is the customer's comprehensive repayment ability score, ranging from 0 to 100 points, with higher scores indicating stronger repayment ability, and n is the total number of valid indicators participating in the scoring. W i The comprehensive weight of the i-th indicator is... N i The normalized score of the i-th indicator is mapped to the 0-100 score range after normalization.

[0097] Wherein, the normalization formula is: Positive indicator normalization: , Negative indicator normalization: , in, N i Let be the normalized score of the i-th indicator, ranging from 0 to 100. X i Let i be the original value of the i-th indicator. X max The maximum value of the i-th indicator across all samples. X minLet be the minimum value of the i-th indicator among all samples.

[0098] A higher positive indicator value is more beneficial for assessing repayment ability, such as credit limit and the percentage of amount already repaid.

[0099] The higher the negative indicator value, the more unfavorable it is for the assessment of repayment ability, such as the number of overdue days and the total overdue amount.

[0100] The linkage logic of the mediation strategy matching module is to automatically match the optimal mediation strategy based on real-time indicator data, and the effect of strategy execution drives indicator iteration.

[0101] Indicator trigger conditions: Preset strategies trigger indicator combinations, for example: Strategy A (SMS reminder): Overdue days ≤ 30 days + most recent repayment record ≥ 1 + no write-off mark.

[0102] Strategy B (Telephone Mediation): 30 days < overdue days ≤ 90 days + outstanding principal < 50% + not out of contact.

[0103] Automatic strategy matching: The indicator library pushes indicator data to the mediation strategy engine in real time. The engine automatically matches the corresponding strategy based on the trigger conditions, synchronizes it to the mediation business system, and assigns it to the corresponding mediator.

[0104] Dynamic adjustment: Mediation execution data, such as SMS response rate, call connection rate, and repayment conversion rate, are fed back into the indicator database. The correlation between the indicators and the strategy effect is verified through a significance test algorithm. For example, if "recent repayment record" has a significant impact on the response rate of strategy B, the screening weight of this indicator can be strengthened; if an indicator, such as "age", has no significant impact on the strategy effect, it can be marked as a redundant indicator.

[0105] Strategy Iteration: Supports dynamic adjustment of strategy parameters based on changes in indicators. For example, when the "overdue days" indicator is updated to 120 days, the strategy will be automatically upgraded from B to C without manual intervention.

[0106] The linkage logic of the risk warning module is to trigger risk warnings through indicator thresholds, and the effectiveness of the indicators is verified by the result of the warning handling.

[0107] Warning indicator settings: Select highly sensitive indicators as warning triggers, for example: Level 1 Warning (High Risk): Sudden increase in overdue days ≥ 30 days / month + outstanding principal ≥ 80% + "Out of Contact" tag = Yes.

[0108] Level 2 Warning (Medium Risk): No repayment record in the past 3 months + total overdue amount increased by ≥50% month-on-month.

[0109] Level 3 Warning (Low Risk): Frequent adjustments to installment periods + sudden increase in credit limit utilization to ≥90%.

[0110] The linkage logic with the case management module is as follows: indicator data runs through the entire case lifecycle, supporting case handling decisions and progress tracking.

[0111] Case filing stage: through case id The system automatically determines the risk level of a case—high / medium / low—based on indicators such as loan principal, credit limit, and initial risk score from the associated indicator library, providing a basis for prioritizing case filing.

[0112] During the case handling stage: Real-time synchronization of indicators such as "overdue days", "recent communication records", and "repayment progress". For example, during the case handling process, if the "number of months since the last repayment" is updated to 0, that is, the customer has made a repayment, the case status is automatically marked as "partial repayment" and the subsequent handling strategy is adjusted.

[0113] During the case closure stage: Link indicators such as "percentage of repaid amount", "processing cycle" and "risk mitigation effect" to generate a case processing assessment report, which is synchronized to the compliance audit module. At the same time, the assessment results, such as "success rate of high-risk case processing", are fed back to the indicator library to optimize the statistical scope and weight of case-related indicators.

[0114] Among them, the linkage logic with the communication record module is to transform key information in the communication records into derived indicators, which in turn support the assessment of repayment ability.

[0115] Derivative metric generation: Structured data from communication records, such as number of communications, connection status, customer promised repayment amount, and promised repayment date, are converted into derived metrics, such as "communication connection rate in the past 3 months," "promised repayment fulfillment rate," and "repayment rate within 7 days after communication." These metrics are automatically entered into the metric library and associated with the corresponding customers. id / case id .

[0116] Indicator Application: Combining derivative indicators with basic indicators improves the accuracy of assessments. For example, customers with a "repayment fulfillment rate ≥ 80%" can have their repayment willingness score increased; customers with a "communication connection rate of 0 in the past 3 months" are automatically marked as "suspected to be out of contact," triggering corresponding warnings and mediation strategies.

[0117] Data closed loop: Updates to communication records are synchronized to the indicator library in real time. Changes in derived indicators drive dynamic adjustments to customer segmentation and mediation strategies, forming a closed loop of "communication-indicators-business".

[0118] Example 2:

[0119] This technical solution can be applied to typical "guardian" business lines, such as the disposal of non-performing credit card assets of banks, where this method is used to build an indicator library and achieve dynamic optimization.

[0120] 1. Data Acquisition and Preprocessing: The system accesses basic characteristic data such as debtor age, gender, and credit limit from the customer information system in real time via RESTful API, and imports data such as overdue days, total overdue amount, and number of months since the last repayment from the repayment record system in batches. The data preprocessing module is activated to fill in the few missing values ​​in the "number of months since the last repayment" field with the average value of all customers in this business line; extreme outliers in "overdue days" that exceed the upper limit of the box plot are marked and manually reviewed by risk control personnel.

[0121] 2. Indicator Screening: The indicator screening module analyzes over 50 candidate indicators. First, it calculates the Pearson correlation coefficient between each indicator and the target variable "30-day repayment rate," initially screening indicators with correlation coefficients higher than 0.3. Then, it proceeds to the significance test: For the "lost contact label," a categorical indicator, the system automatically uses a chi-square test to construct a contingency table with "whether repayment was received," calculating a p-value of 0.001, which is less than the preset significance level of 0.05, thus deemed significant. For the "overdue days," a continuous indicator, the system uses a t-test, dividing customers into two groups based on the median overdue days and comparing the average repayment rates of the two groups, obtaining a p-value of 0.02, also deemed significant. Finally, the Cohen's d's ratio is used to verify the indicator's discriminative power. Through this step, over 50 candidate indicators are reduced to 23 highly effective indicators.

[0122] 3. Weight Configuration: Upon activation of the weight configuration module, on one hand, business experts are invited to use the Analytic Hierarchy Process (AHP) to compare the importance of 23 indicators pairwise, constructing a judgment matrix. After passing a consistency check (CR < 0.1), a set of subjective weights is calculated. On the other hand, based on historical data from the past six months, the system applies the entropy weight method to calculate the objective weights of each indicator. Finally, the system uses a linear combination formula... α is set to 0.4, the comprehensive weight of each indicator is calculated, and an adjustment factor of 1.05 is applied to the weight of indicators such as "overdue days" based on the characteristic of the "Guardian" business line which focuses on short-term repayment.

[0123] Indicator Iteration Management: Three months after the indicator goes live, the system automatically triggers a periodic evaluation. The evaluation report shows that the comprehensive correlation coefficient of the "debtor's age" indicator is below 0.2 for three consecutive months and the p-value is above 0.05. According to the preset rigid quantification rules, the indicator iteration management module automatically marks the indicator as "to be discontinued" and changes its status to "discontinued" after a one-month buffer period and archives it to the historical version repository. The version number of the whole process is upgraded from V1.2.0 to V1.3.0, and the update reasons and operation records are fully saved.

[0124] Business collaboration: A debtor named "Zhang San", customer id The "overdue days" metric for case C1001 has been updated to 45 days. The business collaboration module captures this change in real time and, based on preset rules (30 days < overdue days ≤ 90 days + non-disconnection status), automatically triggers the mediation strategy matching engine. The case is assigned to the "telephone mediation" strategy, and the task is pushed to mediator "Li Si's" workbench. Li Si completes the telephone communication and records "customer promises to repay within 3 days" in the system. This communication record is converted into a derived metric of "promised repayment" and fed back into the metric library. One week later, if Zhang San repays on time, this positive result will be used as positive sample data for the next weight optimization calculation.

[0125] The detailed algorithm flowchart for the indicator selection step in Embodiment 2 of the present invention is shown below. Figure 3 As shown.

[0126] Example 3:

[0127] Within the same system framework, this technical solution enables differentiated adaptation and management of indicators for "Takeoff" business lines with different business characteristics, such as the back-end asset disposal of small consumer loans.

[0128] 1. Dedicated Indicator Configuration: Unlike the "Guardian" business line, the "Soaring" business line focuses more on customers' long-term behavioral patterns and collection scores. Therefore, the system administrator added two dedicated indicators for the "Soaring" business line through the visual interface of the front-end configuration layer: "M2 Score," which comes from an external credit scoring model and "monthly customer-level period count." When defining the indicator metadata, "indicator" will be used... id “business” line Attributes such as "Take Off" are linked to the "Take Off" business line.

[0129] 2. Differentiated Rule Adaptation: For the common indicator "number of months initiating follow-up calls", the two business lines have different evaluation focuses. Through the configuration interface of the business collaboration module, the system configures independent statistical rules and scoring ranges for the "Soaring" business line. For example, the "number of months initiating follow-up calls" is divided into four ranges: "0-3 months / 3-24 months / 24-36 months / ≥36 months", and different score ranges are set for each range. This customized rule configuration only affects the scoring calculation of the "Soaring" business line and will not interfere with other business lines.

[0130] 3. Data isolation and independent iteration: The system is based on the business line identifier. line To achieve data isolation, the effectiveness evaluation and weight calculation of indicators for the "Soaring" business line only use the exclusive data of its business line. For example, when conducting a significance test on the "M2 score" indicator, only case data from the "Soaring" business line is extracted for analysis. This allows the indicator system of this business line to be independently and flexibly iterated and optimized without being bound by globally unified rules.

[0131] Example 4:

[0132] This embodiment focuses on how the data loop formed by the synergistic interaction between indicators and business modules drives the adaptive optimization of the entire indicator system.

[0133] 1. Triggering and Response: The risk warning module has a preset Level 1 warning rule: "Monthly increase in overdue days ≥ 30 days AND outstanding principal ≥ 80% AND 'out of contact' tag = 'Yes'". This warning is triggered when the debtor "Wang Wu" case... id When the C9876 indicator data meets all three conditions, the system automatically generates a high-risk warning message and pushes it to the senior risk control manager via instant messaging.

[0134] 2. Performance Data Feedback: The risk control manager takes emergency action on the alerted case, including initiating an asset verification process. After the case is resolved, the results, such as "whether any funds were recovered within 30 days" or "whether the case was ultimately written off," are used as performance data and fed back through the case file. id It is associated with historical indicator snapshots in the indicator library and then sent back to the data storage layer.

[0135] 3. Driving indicator iteration: In the next monthly evaluation cycle, the indicator screening module re-analyzes the effectiveness of the warning-related indicators, and the significance test module SignificanceTester extracts all cases that have triggered the first-level warning and their final effect data.

[0136] The analysis results show that the "lost contact tag" has a very high significance for the final low recovery rate, i.e., p<0.01, while the significance of the "percentage of outstanding principal" is relatively low, i.e., p=0.08. Based on the test results, the weight configuration module automatically increases the weight of the "lost contact tag" in the early warning model, for example, from 0.9 to 1.0, while decreasing the weight of the "percentage of outstanding principal". This adjustment takes effect in real time through a hot update mechanism, making future risk warnings more accurate. This process is fully automated, realizing an adaptive closed loop of "business feedback → data analysis → model optimization".

[0137] Working principle: A layered, loosely coupled system architecture is constructed, adopting a four-layer structure: front-end configuration layer, business service layer, data processing layer, and data storage layer. Each layer interacts with data through standardized interfaces, implemented using microservices, Spring Boot, and Spring Cloud Alibaba technology stacks. The data processing layer integrates Apache Spark to support large-scale distributed computing. Four core algorithm engines are built: 1) an indicator selection algorithm based on correlation analysis, significance testing, and Cohen's d's discrimination validation; 2) an AHP-entropy weighting algorithm combining subjective business experience and objective data patterns; 3) a multi-dimensional verification algorithm; and 4) an indicator significance testing and dynamic weight adjustment algorithm. A full lifecycle indicator library rule system is established: a four-dimensional rule framework of "standardized definition + rigid verification + standardized iteration + differentiated adaptation," including an indicator classification system, metadata attribute definitions, indicator iteration rules (such as rigid conditions for adding / deactivating / archiving), weight constraint rules, and real-time data synchronization latency ≤1 second (e.g., single indicator weight upper limit ≤0.3, lower limit ≥0.05). An indicator-business collaborative linkage mechanism is implemented through customer... id case id business line ,debt no Four core related fields establish a unified cross-module association system, realizing a closed-loop linkage of "indicator triggering - business response - data feedback - indicator optimization" between the indicator library and five major business modules, including customer segmentation, mediation strategy matching, risk warning, and case management. The system integrates four dimensions: multi-algorithm fusion screening and optimization, full life cycle rule-based management and control, multi-business line differentiated adaptation, and business collaborative linkage, forming a complete closed loop from indicator definition to business application and iterative optimization.

[0138] The beneficial effects of this invention are: high effectiveness and dynamic adaptability: through a multi-algorithm fusion screening and combination weighting mechanism, the objectivity, effectiveness, and scientific nature of the indicators are guaranteed. Simultaneously, the closed-loop feedback and hot update mechanism enable the indicator system to dynamically adapt to business changes, avoiding the decay problem of traditional static models; standardized and traceable management: through rigid quantitative rules, the entire lifecycle management of indicators is automated, from addition and deactivation to archiving, all are documented. Complete version management functions greatly improve the standardization of model risk management and the convenience of auditing; deep business collaboration and efficiency transformation: by establishing closed-loop linkage between indicators and multiple business modules, data analysis results are seamlessly embedded into business processes, achieving automated flow from "data insight" to "business action" and then to "effect feedback," significantly enhancing the final business value of the indicators; high availability and scalability: adopting a layered, loosely coupled microservice architecture, combined with distributed computing and hybrid storage technology, the system possesses high concurrency processing capabilities, high availability, and good horizontal scalability, enabling smooth adaptation to business needs of different scales.

[0139] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for constructing a multi-business line indicator library, characterized in that, Includes the following steps: S1. Data Acquisition and Preprocessing Steps: Collect raw indicator data from multiple heterogeneous data sources, and perform data cleaning and standardization on the raw indicator data to form standardized indicator data. S2. Indicator screening step: Calculate the standardized indicator data based on a preset statistical analysis algorithm, screen out effective indicators that are significantly related to the target variable, and eliminate redundant indicators; the statistical analysis algorithm includes at least correlation analysis, significance test and discrimination verification. S3. Weighting configuration steps: A combined weighting algorithm that integrates subjective business weights and objective data weights is used to assign comprehensive weights to each of the selected effective indicators; the combined weighting algorithm combines the Analytic Hierarchy Process (AHP) and the entropy weight method, and is dynamically adjusted according to the business scenario adaptation factor. S4. Indicator Iteration Management Steps: Based on preset rigid quantification rules, the effective indicators and their weights are managed throughout their entire lifecycle. The rigid quantification rules define the conditions for adding, discontinuing, and archiving indicators, and record the complete version iteration history. S5. Business Collaboration and Linkage Steps: Establish a data association and closed-loop feedback mechanism between the indicators and multiple business modules, so that changes in the values ​​of the indicators trigger automatic adjustments to business strategies, and business execution effect data flows back to the indicator library to drive the dynamic optimization of the indicators and the comprehensive weights.

2. A multi-business line indicator library construction system, characterized in that, include: The data acquisition and preprocessing module is used to collect raw indicator data from multiple heterogeneous data sources and perform data cleaning and standardization on the raw indicator data to form standardized indicator data. The indicator screening module is used to calculate the standardized indicator data based on a preset statistical analysis algorithm, screen out effective indicators that are significantly related to the target variable, and eliminate redundant indicators; the statistical analysis algorithm includes at least correlation analysis, significance test and discrimination verification. The weight configuration module is used to assign comprehensive weights to each of the selected effective indicators by adopting a combined weighting algorithm that integrates subjective business weights and objective data weights; the combined weighting algorithm combines the Analytic Hierarchy Process (AHP) and the entropy weight method, and is dynamically adjusted according to the business scenario adaptation factor. The indicator iteration management module is used to manage the effective indicators and their weights throughout their entire lifecycle according to preset rigid quantification rules. The rigid quantification rules define the conditions for adding, disabling, and archiving indicators, and record the complete version iteration history. The business collaboration module is used to establish a data association and closed-loop feedback mechanism between the indicator and multiple business modules, so that changes in the value of the indicator trigger automatic adjustments to the business strategy, and the business execution effect data flows back to the indicator library to drive the dynamic optimization of the indicator and the comprehensive weight.

3. The method for constructing a multi-business line indicator library as described in claim 1, characterized in that, Step S1, the data acquisition and preprocessing steps, further includes: Data is collected through two modes: real-time API capture and batch offline file import. The data cleaning process includes primary cleaning and secondary cleaning. Primary cleaning fills in missing values ​​using the median or default values ​​for the business scenario. Secondary cleaning identifies and marks outliers by combining the 3σ principle with box plots.

4. The method for constructing a multi-business line indicator library as described in claim 1, characterized in that, Step S2, the significance test in the indicator screening step, further includes: To determine the data type of the indicator, a chi-square test is used for categorical indicators and a t-test is used for continuous numerical indicators. The p-value obtained from the test is compared with the preset significance level threshold to determine whether the indicator is significantly effective.

5. The method for constructing a multi-business line indicator library as described in claim 1, characterized in that, Step S3, the combined weighting algorithm in the weight configuration step, further includes: The subjective weights of each indicator are calculated using the Analytic Hierarchy Process (AHP), and the objective weights of each indicator are calculated using the entropy weight method. The comprehensive weight is calculated using a linear combination formula and a preset weight balance coefficient. An adjustment factor adapted to the business scenario is introduced to dynamically fine-tune the comprehensive weight.

6. The method for constructing a multi-business line indicator library as described in claim 1, characterized in that, Step S4, the rigid quantification rules in the indicator iteration management step, further include: Newly added indicators must pass the validity verification within the first preset time period before they can be officially activated. For existing indicators whose correlation coefficient is lower than the first preset threshold and whose significance test p-value is higher than the second preset threshold for consecutive periods of the second preset time period, they are automatically marked as pending discontinuation and discontinuation or archiving operations are performed after the buffer period. Any change to any metric generates a corresponding version record to support full lifecycle audit traceability.

7. The method for constructing a multi-business line indicator library as described in claim 1, characterized in that, Step S5, the business collaboration and linkage steps, further include: Based on standardized association fields, data links are established between the indicators and the customer segmentation module, mediation strategy matching module, risk warning module, case management module, and communication record module. Automatically trigger and match corresponding business strategies based on preset combination of indicator thresholds; The system receives business execution performance data from the business modules and uses it as input data for a new round of indicator effectiveness evaluation and weight adjustment, forming a closed loop of "indicator triggering - business response - data feedback - indicator optimization".

8. The multi-business line indicator library construction system as described in claim 2, characterized in that, The weight configuration module and the indicator iteration management module work together, specifically for: After each performance indicator iteration evaluation, based on the latest business performance data, the combined weighting algorithm is called again to update the comprehensive weight of each effective indicator, and the updated weight is synchronized to the business service layer through a hot update mechanism without restarting the system service.

9. A multi-business line indicator library construction system as described in claim 2, characterized in that, The indicator iteration management module is further used for: Store and maintain a metadata database containing indicator definitions, statistical calibers, calculation rules, data sources, validity verification methods, and historical version information; The system provides a visual management interface for configuring the metadata database and the rigid quantification rules, enabling differentiated adaptation and management of indicators for different business lines.

10. A multi-business line indicator library construction system as described in claim 2, characterized in that, The system adopts a hierarchical loosely coupled architecture, which includes, from top to bottom: The front-end configuration layer provides a visual interactive and configuration interface. The business service layer adopts a microservice architecture to deploy the indicator filtering module, weight configuration module, and business collaboration module. The data processing layer integrates a distributed computing framework to execute the statistical analysis algorithms. The data storage layer adopts a hybrid storage architecture, in which structured indicator metadata and weight configuration data are stored in a relational database, while frequently accessed configuration data is cached in an in-memory database.

Images