Optimization system and method for improving effectiveness of external data source

Abstract

The invention relates to the technical field of big data processing and financial risk control, in particular to an optimization system and method for improving effectiveness of an external data source. According to the optimization system for improving the effectiveness of the external data source, the external data source is divided into a feature class, a verification class, an evidence class and model sub-data through a classification identification module, and marking is carried out; the multi-dimensional evaluation module calculates a comprehensive score for different types of configuration evaluation index systems; the dynamic optimization and switching module screens an optimal data source in combination with comprehensive scores and cost parameters, establishes an alternative library and realizes dynamic switching through drainage testing and periodic comparison; and the automatic scheduling module automatically triggers evaluation, test and comparison processes. According to the method, full-process closed-loop management is formed, the problems of difficulty in data source quality fluctuation monitoring, evaluation lagging and low switching efficiency are solved, and the reliability of data support in credit decision and the system adaptability are improved.

Description

Technical Field

[0001] This invention relates to the fields of big data processing and financial risk control technology, specifically to an optimized system and method for improving the effectiveness of external data sources. Background Technology

[0002] In the context of big data risk control for credit decision-making, the quality of external data sources directly affects the accuracy and reliability of these decisions. Currently, when using external data sources for big data risk control and credit decision-making in the financial sector, existing technologies have many problems in controlling the quality of data sources.

[0003] From a data acquisition perspective, data sources are diverse, covering multiple fields such as credit reporting, consumption, telecommunications, and government affairs. However, data acquisition channels are often unstable, and the update mechanisms of data sources are imperfect. Some external data sources are not updated in a timely manner, with long update cycles, failing to reflect changes in business operations promptly. When unexpected situations arise in a business's operations, credit decisions made based on lagging data can be biased. Regarding data quality, data sources such as e-commerce and social media suffer from data distortion. Fraudulent transactions inflate e-commerce transaction data, failing to accurately reflect users' spending power and repayment ability. Social media opinions are highly subjective, difficult to verify, and have a high error rate. Data coverage has blind spots; for certain specific groups, such as users in rural areas or those with no credit history, existing data sources provide limited effective information, making it difficult to comprehensively assess their credit risk. Furthermore, technical stability and cost are also prominent issues. Data source vendor systems are unstable; for example, network failures at financial terminals can affect the real-time risk control decisions of multiple financial institutions. Moreover, when multiple data sources are connected, system performance is prone to bottlenecks, interface response delays, and impacting credit approval efficiency.

[0004] In conclusion, new technologies and optimized systems are urgently needed to monitor and manage the quality of external data sources in order to ensure the scientific nature and accuracy of big data risk control credit decisions. Summary of the Invention

[0005] The present invention aims to provide an optimized system and method for improving the effectiveness of external data sources, in order to solve the problems of lagging evaluation, inefficient switching, and difficulty in ensuring long-term effectiveness of traditional external data sources.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: An optimization system for improving the effectiveness of external data sources includes: The data source classification and labeling module is used to classify and label external data sources, dividing them into feature data, verification data, evidence data, and model data, and adding corresponding category labels to each type of data. The multi-dimensional evaluation module is used to configure corresponding multi-dimensional evaluation indicator systems for external data sources with different category tags, and calculate the comprehensive score of the corresponding external data source according to the preset weights and scoring standards of each evaluation indicator. The dynamic optimization and switching module is used to select the current optimal data source based on the comprehensive score and cost parameters of each external data source; establish a candidate data source library, conduct traffic testing on the candidate data sources, and regularly compare the comprehensive performance of the candidate data sources with the current optimal data source. When the performance of the candidate data source is better than that of the current optimal data source, the module will automatically switch to the candidate data source. The automated scheduling module is used to configure threshold parameters for evaluation cycle, traffic generation test ratio, and performance comparison cycle, and automatically trigger the corresponding evaluation, testing, and comparison processes based on the threshold parameters.

[0007] The principle and advantages of this solution are as follows: In practical applications, based on the essential attributes and business uses of the data source, external data sources are accurately divided into four categories—feature-based, verification-based, evidence-based, and model-based—and labeled using a combination of rule matching and intelligent classification models, laying the foundation for subsequent differentiated evaluation. A multi-dimensional evaluation index system is configured for the characteristics of different data source categories, and a comprehensive score is quantitatively calculated using preset weights and scoring standards to achieve an objective evaluation of data source quality. A collaborative screening mechanism based on cost parameters and comprehensive scores is introduced to determine the current optimal data source; simultaneously, a candidate database is established, and the performance of candidate data sources is continuously verified through tiered traffic redirection testing. Combined with periodic automatic comparisons to trigger dynamic switching, this ensures that the system always uses the optimal data source. By configuring threshold parameters such as evaluation cycle and testing ratio through the automated scheduling module, the entire process is automatically triggered, forming a virtuous cycle of "precise classification - multidimensional evaluation - dynamic optimization - automated scheduling". This fundamentally solves problems such as difficulty in monitoring data source quality fluctuations, evaluation lag, and inefficient switching. The automation mechanism ensures dynamic matching between data source quality and business needs, improves the reliability of data support in credit decisions, reduces decision risks caused by data source quality fluctuations, and enhances the system's adaptability to diverse business scenarios.

[0008] Preferably, as an improvement, the data source classification and identification module includes: The data source metadata acquisition submodule is used to obtain descriptive information from external data sources; The rule engine submodule has a built-in initial classification rule library, which contains feature keywords and matching conditions; it matches the description information with the initial classification rule library and outputs preliminary classification results. The machine learning classification submodule trains a classification model based on a preset training sample set, which contains metadata about data sources with labeled categories; it receives the description information and preliminary classification results; when the matching degree of the preliminary classification results is lower than a preset threshold, it calls the classification model to perform secondary classification on the data source and outputs optimized classification results. The category label generation submodule generates standardized labels for the corresponding categories based on the preliminary classification results or optimized classification results. The human interaction submodule provides a visual interface for users to view classification results, receives correction instructions for classification results, and feeds back the corrected classification results to the rule engine submodule and the machine learning classification submodule to update the initial classification rule library and training sample set.

[0009] Technical Results: Through a dual classification mechanism combining a rule engine and machine learning classification, coupled with manual intervention for correction, this approach ensures both rapid and accurate classification of conventional data sources and high accuracy in classifying complex or novel data sources. The rule engine guarantees basic classification efficiency, the machine learning model solves the classification challenges of low-matching data sources, and manual intervention forms a closed loop of "classification-feedback-optimization," continuously improving the adaptability of the classification rule base and model, reducing subsequent evaluation biases caused by classification errors, and providing a precise classification foundation for multi-dimensional evaluation.

[0010] Preferably, as an improvement, the classification model is a text convolutional neural network model. The input of the model is a text vector describing the information, and the output is the probability distribution of four types of data. When the probability value of the category corresponding to the highest probability is higher than the threshold, it is determined as the optimized classification result.

[0011] Technical Results: Employing a text convolutional neural network as the classification model effectively captures semantic features and deep relationships within the data source descriptions. Compared to traditional rule matching or simple machine learning models, it demonstrates stronger classification capabilities for data sources with fuzzy descriptions and multiple feature intersections. When the initial classification matching degree is low, the classification result is determined by outputting a probability distribution and using a high-probability threshold, significantly reducing the misclassification rate of low-matching data sources and improving the robustness of the classification results. This lays the foundation for accurate adaptation of subsequent evaluation metrics.

[0012] Preferably, as an improvement, the data source classification and identification module further includes a tag association submodule, which is used to associate and store category tags with the access interface information and historical call records of the data source to form a data source classification archive. The classification archive supports retrieval and batch calling by category tags.

[0013] Technical Benefits: By associating category tags with data source access interface information and historical call records to form categorized profiles, integrated management of data source category attributes, technical parameters, and usage records is achieved. On one hand, it supports quick retrieval of target data sources by category tags, reducing the time cost of manual searching; on the other hand, the batch call function avoids operational errors from configuring data sources one by one, and the historical call records allow for tracing the usage effect of data sources, providing a basis for evaluating the long-term stability of data sources, thus improving the overall efficiency of data source management and retrieval.

[0014] Preferably, as an improvement, the dynamic optimization and switching module includes: The optimal data source filtering submodule is used to receive the comprehensive scores and cost parameters of each external data source, calculate the comprehensive evaluation index of each data source based on the preset weight allocation rules, and filter out the data source with the highest comprehensive evaluation index as the current optimal data source. The candidate data source management submodule is used to include external data sources that have not been selected as the current optimal data source into the candidate data source database, and to dynamically update the candidate data source database according to a preset elimination threshold. The traffic acquisition test submodule has a built-in traffic allocation engine, which is used to divert test samples from new application data according to the preset traffic acquisition test ratio, call the alternative data source in the alternative data source library to synchronously acquire data from the test samples, and record the overall test score and response time of the alternative data source. The periodic comparison submodule is configured with a time-triggered unit, which automatically triggers the comparison process according to a preset performance comparison cycle; The automatic switching execution submodule is used to receive switching trigger signals, perform data source switching operations, including updating the data source call interface address, adjusting the data acquisition logic, synchronously notifying related systems of data source change information, and generating switching log archives.

[0015] Technical Benefits: The dynamic optimization and switching module constructs a full lifecycle management mechanism for data sources through collaborative screening of comprehensive scores and cost parameters, dynamic management of the alternative database, lead generation testing and verification, and periodic automatic comparison. This resolves the quality fluctuation risk caused by reliance on a single traditional data source, ensuring the system always uses the data source with the best overall performance through automatic switching. Simultaneously, the combination of the alternative database and lead generation testing avoids business interruptions in the event of sudden data source failures, balancing data source quality and cost-effectiveness, and improving the stability of credit decisions.

[0016] Preferably, as an improvement, the traffic diversion test submodule further includes a sample stratification unit, which is used to stratify the new application data according to the user risk level and geographical distribution, so that the proportion of the diverted test samples in each stratification is consistent with the stratification proportion of the new application data.

[0017] Technical Effects: The sample stratification unit stratifies new application data according to user risk level and geographical distribution, while maintaining consistent stratification ratios between the test samples and the original data, ensuring the representativeness of the lead generation test samples. This avoids sample bias that may be caused by traditional random sampling (such as distortion of the proportion of high-risk user samples), enabling the test results of alternative data sources to truly reflect their performance in actual business scenarios, improving the reliability of alternative data source evaluation, and providing accurate basis for subsequent switching decisions.

[0018] Preferably, as an improvement, the automatic switching execution submodule further includes a grayscale switching unit, which is used to gradually increase the number of alternative data sources called according to a preset grayscale ratio when performing a data source switching operation, while monitoring the data acquisition success rate. When the success rate is stable above a preset threshold, the full switching is completed.

[0019] Technical Results: By gradually increasing the number of alternative data source calls and monitoring the success rate, a smooth transition for data source switching was achieved. This avoided system load fluctuations or data acquisition interruptions that could result from a direct full switch, ensuring business continuity. Simultaneously, success rate monitoring allowed for the timely detection of potential problems with alternative data sources and the termination of the switch, reducing switching risks and improving system stability.

[0020] Preferably, as an improvement, the optimal data source filtering submodule further includes a cost parameter dynamic correction unit, which is used to calculate the batch purchase discount coefficient based on the historical call volume of the data source, substitute the batch purchase discount coefficient into the calculation of the unit call cost, and obtain the corrected cost parameter.

[0021] Technical Benefits: By calculating the bulk purchase discount coefficient based on historical call volume, the unit call cost is corrected, making the cost parameters more closely match actual procurement scenarios. This solves the cost assessment bias caused by the neglect of bulk discounts in traditional fixed cost calculations, ensuring that the comprehensive evaluation indicators can truly reflect the balance between data source performance and cost, avoiding the misselection of the optimal data source due to cost overestimation or underestimation, and improving the economy and rationality of data source selection.

[0022] It also includes an optimization method for improving the effectiveness of external data sources, which is applied to an optimization system for improving the effectiveness of external data sources. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the structure of an optimization system to improve the effectiveness of external data sources. Detailed Implementation

[0024] The following detailed description illustrates the specific implementation method: The basic implementation examples are as follows: Figure 1 As shown: An optimization system for improving the effectiveness of external data sources includes: The data source classification and labeling module is used to classify and label external data sources, dividing them into feature data, verification data, evidence data, and model data, and adding corresponding category labels to each type of data.

[0025] Among them, feature-based data includes multi-head data used to extract business features; verification-based data includes educational background data, ID card number data, or mobile phone number data used to verify the authenticity of information; evidence-based data includes blacklist data used to identify high-risk users; and model score data includes credit score data used to assess the credit level of users.

[0026] The data source classification and identification module includes a data source metadata collection submodule, a rule engine submodule, a machine learning classification submodule, a category label generation submodule, and a human interaction submodule.

[0027] The data source metadata acquisition submodule is used to obtain descriptive information from external data sources. This descriptive information includes the data source name, data type description, data format, and purpose description. The rule engine submodule has a built-in initial classification rule library. This library contains feature keywords and matching conditions corresponding to feature-type data, verification-type data, evidence-type data, and model score data. The matching conditions of the initial classification rule library include: if the descriptive information contains multiple borrowing, behavioral characteristics, or user attributes, and the purpose description is "extracting feature variables," then it matches as feature-type data; if the descriptive information contains ID card verification, education verification, or mobile phone number authenticity, and the purpose description is "information verification," then it matches as verification-type data; if the descriptive information contains blacklists, graylists, or risk lists, and the purpose description is "high-risk identification," then it matches as evidence-type data; if the descriptive information contains credit scores, risk scores, or model output values, and the purpose description is "credit assessment," then it matches as model score data. The rule engine submodule matches the descriptive information with the initial classification rule library and outputs preliminary classification results. The machine learning classification submodule trains a classification model based on a preset training sample set, which contains data source metadata with labeled categories. The submodule receives the descriptive information and preliminary classification results. When the matching degree of the preliminary classification result is lower than a preset threshold, it calls the classification model to perform a secondary classification of the data source and outputs an optimized classification result. The classification model is a text convolutional neural network model. The input of the model is a text vector of the descriptive information, and the output is the probability distribution of four categories of data. When the probability value of the category corresponding to the highest probability is higher than the threshold, it is determined as the optimized classification result. In this embodiment, the threshold is preferably 0.8.

[0028] The category label generation submodule generates standardized labels for the corresponding categories based on the preliminary or optimized classification results. These standardized labels include feature categories, validation categories, evidence categories, and model scores. The human interaction submodule provides a visual interface for users to view the classification results, receives correction instructions for the classification results, and feeds the corrected classification results back to the rule engine submodule and the machine learning classification submodule to update the initial classification rule base and training sample set.

[0029] The data source classification and identification module also includes a tag association submodule, which is used to associate and store category tags with the access interface information and historical call records of the data source to form a data source classification archive. The classification archive supports retrieval and batch calling by category tags.

[0030] The multi-dimensional evaluation module is used to configure corresponding multi-dimensional evaluation indicator systems for external data sources with different category labels, and calculate the comprehensive score of the corresponding external data source according to the preset weights and scoring standards of each evaluation indicator. In this embodiment, the evaluation indicators for feature-type data include detection rate, coverage rate, and accuracy rate; the detection rate is used to measure the proportion of feature-type data that is correctly identified as positive among all samples that are actually positive, detection rate = number of true positive samples / (number of true positive samples + number of false negative samples); the coverage rate is used to evaluate the sample range covered by the features, coverage rate = number of features actually extracted / total number of features; the accuracy rate is used to measure the overall proportion of correct classification, accuracy rate = (number of true positive samples + number of true negative samples) / (number of true positive samples + number of true negative samples + number of false positive samples + number of false negative samples). The evaluation indicators for verification-type data include accuracy rate, completeness rate, and timeliness; the accuracy rate is used to measure the consistency between verification data and official or other reliable data sources, accuracy rate = number of correctly verified data entries / total number of data entries; the completeness rate is used to evaluate the completeness of verification data, completeness rate = number of entries without missing or ambiguous information / total number of data entries. Evaluation metrics for evidential data include completeness and accuracy. Evaluation metrics for model score data include consistency, stability, and discrimination. Consistency measures the consistency of model scores with other credit assessment standards, measured by correlation coefficient or Kendall's rank correlation coefficient. Stability refers to the stability of PSI-assessed model scores across different time periods or scenarios, measured by PSI, standard deviation, or coefficient of variation. Discrimination measures the ability of model scores to differentiate borrowers at different risk levels, measured by ROC curve or KS value. The pre-defined weights and scoring standards for each evaluation metric are as follows: for feature-based data, the search rate has a weight of 30%, coverage a weight of 20%, and accuracy a weight of 50%; for verification-based data, accuracy a weight of 40%, completeness a weight of 30%, and timeliness a weight of 30%; for evidential data, completeness and accuracy each have a weight of 50%; for model score data, consistency a weight of 30%, stability a weight of 30%, and discrimination a weight of 40%. The scoring standard for each evaluation metric is a tiered scoring system, assigning a corresponding score based on the range in which the actual value of the metric falls.

[0031] The dynamic optimization and switching module is used to select the current optimal data source based on the comprehensive score and cost parameters of various external data sources; establish a candidate data source library, conduct traffic testing on candidate data sources, and regularly compare the comprehensive performance of candidate data sources with the current optimal data source. When the performance of a candidate data source is better than that of the current optimal data source, the module will automatically switch to the candidate data source.

[0032] The dynamic optimization and switching module includes an optimal data source filtering submodule, a candidate database management submodule, a traffic generation testing submodule, a periodic comparison submodule, and an automatic switching execution submodule.

[0033] The optimal data source filtering submodule receives the comprehensive scores and cost parameters from various external data sources, calculates the comprehensive evaluation index of each data source based on a preset weighting rule, and selects the data source with the highest comprehensive evaluation index as the current optimal data source. The weighting rule assigns 50% weight to the comprehensive score and 50% to the cost parameter, which includes the unit call cost of the data source. The optimal data source filtering submodule also includes a cost parameter dynamic correction unit, which calculates a batch purchase discount coefficient based on the historical call volume of the data source, substitutes the batch purchase discount coefficient into the calculation of the unit call cost, and obtains the corrected cost parameter.

[0034] The candidate data source management submodule is used to include external data sources that have not been selected as the current optimal data source into the candidate data source database, sort them from high to low according to their comprehensive scores, and dynamically update the candidate data source database according to a preset elimination threshold. When the comprehensive score of a candidate data source is lower than the elimination threshold, it is removed from the candidate data source database.

[0035] The traffic referral testing submodule has a built-in traffic allocation engine. It allocates test samples from new application data according to a preset traffic referral testing ratio. It calls the top N candidate data sources in the candidate data source library to synchronously acquire data from the test samples, recording the overall test score and response time of each candidate data source. The traffic referral testing ratio is adjusted through a threshold configuration interface, where N is a configurable positive integer. The traffic referral testing submodule also includes a sample stratification unit, which stratifies new application data according to user risk level and geographical distribution, ensuring that the proportion of allocated test samples in each stratum matches the proportion of new application data in each stratum.

[0036] The periodic comparison submodule is configured with a time-sensitive triggering unit. This unit automatically triggers the comparison process according to a preset performance comparison cycle. During the comparison process, the periodic comparison unit extracts the historical comprehensive score of the current optimal data source and the test comprehensive score of the alternative data source, calculates the difference between the two, and generates a switching trigger signal when the difference meets a preset switching condition. The performance comparison cycle is adjusted through a threshold configuration interface. The switching condition in the periodic comparison submodule includes: the test comprehensive score of the alternative data source is higher than the historical comprehensive score of the current optimal data source for M consecutive performance comparison cycles, and the average difference between the two is greater than a preset threshold; where M is a configurable positive integer.

[0037] The automatic switching execution submodule receives switching trigger signals and executes data source switching operations, including updating the data source call interface address, adjusting data acquisition logic, synchronously notifying related systems of data source change information, and generating and archiving switching logs. The automatic switching execution submodule also includes a gray-scale switching unit, which, during data source switching operations, gradually increases the number of calls to alternative data sources according to a preset gray-scale ratio while monitoring the data acquisition success rate. When the success rate stabilizes above a preset threshold, a full switch is completed; the gray-scale ratio is adjusted through a threshold configuration interface.

[0038] The automated scheduling module is used to configure threshold parameters for evaluation cycle, traffic generation test ratio, and performance comparison cycle, and automatically trigger the corresponding evaluation, testing, and comparison processes based on the threshold parameters.

[0039] This application also includes an optimization method for improving the effectiveness of external data sources, applied to an optimization system for improving the effectiveness of external data sources.

[0040] The above descriptions are merely embodiments of the present invention, and common knowledge such as specific technical solutions and / or characteristics are not described in detail here. It should be noted that those skilled in the art can make various modifications and improvements without departing from the technical solutions of the present invention, and these should also be considered within the scope of protection of the present invention. These modifications and improvements will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. An optimization system for improving the effectiveness of external data sources, characterized in that, The data source classification and labeling module is used to classify and label external data sources, dividing them into feature data, verification data, evidence data, and model data, and adding corresponding category labels to each type of data. The multi-dimensional evaluation module is used to configure corresponding multi-dimensional evaluation indicator systems for external data sources with different category tags, and calculate the comprehensive score of the corresponding external data source according to the preset weights and scoring standards of each evaluation indicator. The dynamic optimization and switching module is used to select the current optimal data source based on the comprehensive scores and cost parameters of various external data sources. Establish a backup data source library, conduct traffic testing on the backup data sources, and regularly compare the comprehensive performance of the backup data sources with the current best data source. When the performance of the backup data source is better than that of the current best data source, automatically switch to the backup data source. The automated scheduling module is used to configure threshold parameters for evaluation cycle, traffic generation test ratio, and performance comparison cycle, and automatically trigger the corresponding evaluation, testing, and comparison processes based on the threshold parameters.

2. The optimization system for improving the effectiveness of external data sources according to claim 1, characterized in that, The data source classification and identification module includes: The data source metadata acquisition submodule is used to obtain descriptive information from external data sources; The rule engine submodule has a built-in initial classification rule library, which contains feature keywords and matching conditions; it matches the description information with the initial classification rule library and outputs preliminary classification results. The machine learning classification submodule trains a classification model based on a preset training sample set, which contains metadata about data sources with labeled categories; it receives the description information and preliminary classification results; when the matching degree of the preliminary classification results is lower than a preset threshold, it calls the classification model to perform secondary classification on the data source and outputs optimized classification results. The category label generation submodule generates standardized labels for the corresponding categories based on the preliminary classification results or optimized classification results. The human interaction submodule provides a visual interface for users to view classification results, receives correction instructions for classification results, and feeds back the corrected classification results to the rule engine submodule and the machine learning classification submodule to update the initial classification rule library and training sample set.

3. The optimization system for improving the effectiveness of external data sources according to claim 2, characterized in that: The classification model is a text convolutional neural network model. The input of the model is a text vector describing the information, and the output is the probability distribution of four types of data. When the probability value of the category corresponding to the highest probability is higher than the threshold, it is determined to be the optimized classification result.

4. The optimization system for improving the effectiveness of external data sources according to claim 2, characterized in that: The data source classification and identification module also includes a tag association submodule, which is used to associate and store category tags with the access interface information and historical call records of the data source to form a data source classification archive. The classification archive supports retrieval and batch calling by category tags.

5. The optimization system for improving the effectiveness of external data sources according to claim 1, characterized in that, The dynamic optimization and switching module includes: The optimal data source filtering submodule is used to receive the comprehensive scores and cost parameters of each external data source, calculate the comprehensive evaluation index of each data source based on the preset weight allocation rules, and filter out the data source with the highest comprehensive evaluation index as the current optimal data source. The candidate data source management submodule is used to include external data sources that have not been selected as the current optimal data source into the candidate data source database, and to dynamically update the candidate data source database according to a preset elimination threshold. The traffic acquisition test submodule has a built-in traffic allocation engine, which is used to divert test samples from new application data according to the preset traffic acquisition test ratio, call the alternative data source in the alternative data source library to synchronously acquire data from the test samples, and record the overall test score and response time of the alternative data source. The periodic comparison submodule is configured with a time-triggered unit, which automatically triggers the comparison process according to a preset performance comparison cycle; The automatic switching execution submodule is used to receive switching trigger signals, perform data source switching operations, including updating the data source call interface address, adjusting the data acquisition logic, synchronously notifying related systems of data source change information, and generating switching log archives.

6. The optimization system for improving the effectiveness of external data sources according to claim 5, characterized in that: The optimal data source filtering submodule also includes a cost parameter dynamic correction unit, which is used to calculate the batch purchase discount coefficient based on the historical call volume of the data source, substitute the batch purchase discount coefficient into the calculation of the unit call cost, and obtain the corrected cost parameter.

7. The optimization system for improving the effectiveness of external data sources according to claim 5, characterized in that: The traffic diversion test submodule also includes a sample stratification unit, which is used to stratify the new application data according to the user risk level and geographical distribution, so that the proportion of the diverted test samples in each stratification is consistent with the proportion of the new application data in each stratification.

8. An optimization system for improving the effectiveness of external data sources according to claim 5, characterized in that: The automatic switching execution submodule also includes a grayscale switching unit, which is used to gradually increase the number of alternative data sources called according to a preset grayscale ratio when performing data source switching operations, while monitoring the data acquisition success rate. When the success rate is stable above a preset threshold, the full switching is completed.

9. An optimization method for improving the effectiveness of external data sources, characterized in that: This is applied to an optimization system for improving the effectiveness of external data sources as described in any one of claims 1-8.

Images