Rule parameter adjusting method and device, electronic equipment and program product

Abstract

The invention discloses a rule parameter adjusting method and device, electronic equipment and a program product. The rule parameter adjusting method comprises the following steps: acquiring current rule configuration information from a rule engine, wherein the rule configuration information comprises a current rule parameter; obtaining user historical data associated with the current rule parameter; based on a preset tuning target of a to-be-optimized index, performing dynamic optimization processing by using the user historical data to obtain optimized rule parameters; and the optimized rule parameters are applied to the rule engine, and new rule configuration information is obtained. According to the rule parameter adjustment method and device, the electronic equipment and the program product, dynamic adjustment of the rule parameters can be achieved in an automatic mode, and the limitation of a manual adjustment mode in the prior art is overcome.

Description

Technical Field

[0001] The present disclosure relates to the field of computer application technology, and in particular to a rule parameter adjustment method, device, electronic device and program product. Background Art

[0002] In the field of financial technology (such as risk control), rule parameters (such as thresholds) will change according to factors such as time, credit conditions, and user group characteristics. At present, rule parameter adjustments are mainly completed manually: business personnel need to analyze historical data offline and manually set rule parameters such as thresholds based on the analysis results. This manual method is not only complex in statistical analysis, but also limited by human work and rest time, and cannot modify rule parameters such as thresholds in a timely manner in response to emergencies around the clock.

[0003] Therefore, it is desirable to provide an improved rule parameter adjustment scheme.

[0004] The content of this background technology description is only for facilitating understanding of the relevant technology in this field and is not regarded as an admission of the prior art. Summary of the invention

[0005] The embodiments of the present disclosure aim to provide a rule parameter adjustment method, device, electronic device and program product, which realize dynamic adjustment of rule parameters in an automated manner, thus overcoming the limitations of the manual adjustment method in the prior art.

[0006] In a first aspect, an embodiment of the present disclosure provides a rule parameter adjustment method, which may include:

[0007] Acquire current rule configuration information from the rule engine, wherein the rule configuration information includes the rule configuration to be processed and the corresponding current rule parameters;

[0008] Acquire user history data associated with the current rule parameters;

[0009] Based on the preset tuning target of the indicator to be optimized, the user historical data is used to perform dynamic optimization processing to obtain optimized rule parameters; and

[0010] The optimized rule parameters are applied to the rule engine to obtain new rule configuration information.

[0011] In some embodiments, the rule parameters include, for example, rule thresholds.

[0012] In some embodiments, the preset tuning target based on the indicator to be optimized uses the user historical data to perform dynamic optimization processing to obtain optimized rule parameters, including:

[0013] Based on the user historical data, using a preset algorithm to construct at least one association model between a feature and at least one indicator, wherein the user historical data includes user historical feature data and historical indicators, and the at least one indicator includes the indicator to be optimized; and

[0014] Determine the correlation model corresponding to the indicator to be optimized, and use the corresponding correlation model to determine the feature to be optimized and its corresponding rule parameters;

[0015] Based on the preset tuning target of the indicator to be optimized, the corresponding association model is used to dynamically optimize the rule parameters corresponding to the feature to be optimized to obtain the optimized rule parameters.

[0016] In some embodiments, the preset algorithm is an XGBoost algorithm. In these embodiments, the method of constructing at least one association model between a feature and at least one indicator based on the user historical data using the preset algorithm includes:

[0017] Obtaining training sample data for training the at least one association model according to the user history feature data and the history execution result of the user history data, wherein the training sample data includes a training set and a validation set; and

[0018] Based on the training set, using the XGBoost algorithm to iteratively construct a decision tree sequence to obtain at least one trained association model, each association model including a decision tree set;

[0019] For each iteration, a new decision tree is constructed based on the prediction error of the previous decision tree, and the performance index of the association model is evaluated using the validation set until the performance index meets the preset iteration termination condition.

[0020] In some embodiments, determining the association model corresponding to the indicator to be optimized, and using the corresponding association model to determine the feature to be optimized and its corresponding rule parameter, includes:

[0021] Determine the correlation model corresponding to the indicator to be optimized;

[0022] Determine the influence of each feature in the decision tree set of the corresponding association model on the indicator to be optimized; and

[0023] The feature to be optimized is determined according to the influence, and a rule parameter corresponding to the feature to be optimized is obtained.

[0024] In some embodiments, based on the preset tuning target of the indicator to be optimized, the rule parameters corresponding to the feature to be optimized are dynamically optimized using the corresponding association model to obtain the optimized rule parameters, including:

[0025] Based on the association model corresponding to the indicator to be optimized, determining the correlation between each feature to be optimized and the indicator to be optimized;

[0026] Determining, based on the correlation, an adjustment direction and / or adjustment range of the rule parameters corresponding to each feature to be optimized; and

[0027] The rule parameters corresponding to each feature to be optimized are dynamically adjusted along the adjustment direction and / or within the adjustment range to obtain the optimized rule parameters that meet the preset tuning objectives.

[0028] In some embodiments, the dynamically adjusting the rule parameters corresponding to each feature to be optimized along the adjustment direction and / or within the adjustment range includes:

[0029] Determine the updated rule parameters corresponding to each feature to be optimized;

[0030] Based on the updated rule parameters, using the association model to perform prediction analysis on the historical sample data to obtain a prediction result;

[0031] If the prediction result does not meet the preset tuning target, the rule parameter is further adjusted along the adjustment direction and / or within the adjustment range according to a preset step size to obtain a further updated rule parameter.

[0032] Optionally, if the prediction result meets the preset tuning target, the updated rule parameter is determined as the optimized rule parameter.

[0033] In some embodiments, the rule parameter adjustment method further includes:

[0034] Acquire user performance data corresponding to the rule engine to which the optimized rule parameters are applied;

[0035] evaluating whether the actual indicator value corresponding to the indicator to be optimized in the user performance data meets the preset tuning target; and

[0036] When the actual indicator value does not meet the preset optimization target, the dynamic optimization process is re-performed using the corresponding association model based on the preset optimization target of the indicator to be optimized.

[0037] In some embodiments, obtaining current rule configuration information from the rule engine includes:

[0038] The acquired rule configuration information is filtered according to the preset filtering settings to obtain the rule configuration to be processed and the corresponding current rule parameters.

[0039] In some embodiments, the applying the optimized rule parameters to the rule engine to obtain new rule configuration information includes:

[0040] Generate a version number corresponding to the new rule configuration information; and / or

[0041] Determine whether the optimized rule parameters require manual review:

[0042] If necessary, applying the rule parameters that have passed manual review to the rule engine;

[0043] If not necessary, the optimized rule parameters are directly applied to the rule engine.

[0044] In a second aspect, an embodiment of the present disclosure provides a rule parameter adjustment device, comprising:

[0045] A configuration acquisition unit, used to acquire current rule configuration information from a rule engine, wherein the rule configuration information includes current rule parameters;

[0046] A data acquisition unit, used to acquire user history data associated with the current rule parameters;

[0047] An optimization unit, configured to perform dynamic optimization processing using the user historical data based on a preset optimization target of the indicator to be optimized, to obtain optimized rule parameters; and

[0048] The application unit is used to apply the optimized rule parameters to the rule engine to obtain new rule configuration information.

[0049] In a third aspect, an embodiment of the present disclosure provides an electronic device, which may include: a processor and a memory storing a computer program, wherein the processor is configured to implement the method described in the first aspect when executing the computer program.

[0050] In a fourth aspect, an embodiment of the present disclosure provides a program product, comprising a computer program, wherein when the computer program is executed by a processor, the method described in the first aspect is implemented.

[0051] In a fifth aspect, an embodiment of the present disclosure provides a storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the method described in the first aspect.

[0052] The rule parameter (such as threshold) adjustment method provided by the embodiment of the present disclosure obtains the current rule configuration information from the rule engine; obtains the user history data associated with the current rule parameter (such as threshold); based on the preset tuning target of the indicator to be optimized, uses the user history data to perform dynamic optimization processing to obtain the optimized rule parameter (such as threshold); and applies the optimized rule parameter (such as threshold) to the rule engine to obtain new rule configuration information, thereby realizing the automated rule parameter (such as threshold) adjustment from configuration acquisition, parameter optimization to update application, and improving the efficiency and real-time performance of parameter (such as threshold) adjustment.

[0053] Other optional features and technical effects of the embodiments of the present disclosure are partially described below, and partially can be understood by reading this document. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. The elements shown are not limited by the proportions shown in the accompanying drawings. The same or similar reference numerals in the accompanying drawings represent the same or similar elements, wherein:

[0055] Figure 1 A first flow chart of a rule parameter adjustment method according to an embodiment of the present disclosure is shown;

[0056] Figure 2 A second flow chart of the rule parameter adjustment method according to an embodiment of the present disclosure is shown;

[0057] Figure 3 A third flow chart of a rule parameter adjustment method according to an embodiment of the present disclosure is shown;

[0058] Figure 4 A fourth flow chart of the rule parameter adjustment method according to an embodiment of the present disclosure is shown;

[0059] Figure 5 A fifth flow chart of a rule parameter adjustment method according to an embodiment of the present disclosure is shown;

[0060] Figure 6 A sixth flow chart of the rule parameter adjustment method according to an embodiment of the present disclosure is shown;

[0061] Figure 7 A seventh flow chart of a rule parameter adjustment method according to an embodiment of the present disclosure is shown;

[0062] Figure 8 An architectural diagram of a rule parameter adjustment system capable of implementing the rule parameter adjustment method according to an embodiment of the present disclosure is shown;

[0063] Fig. 9 An exemplary structural diagram of a rule parameter adjustment device according to an embodiment of the present disclosure is shown; and

[0064] Fig.10 An exemplary structural diagram of an electronic device capable of implementing the method according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0065] In order to make the purpose, technical solution and advantages of the present disclosure more clear, the present disclosure is further described in detail below in conjunction with specific implementation methods and drawings. Here, the illustrative implementation methods and descriptions of the present disclosure are used to explain the present disclosure, but are not intended to limit the present disclosure.

[0066] As used herein, the term "including" and its variations mean open inclusion, i.e., "including but not limited to". Unless otherwise stated, the term "or" means "and / or". The term "based on" means "based at least in part on". The terms "an example embodiment" and "an embodiment" mean "at least one example embodiment". The term "another embodiment" means "at least one some embodiments". The terms "first", "second", etc. may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0067] In the field of financial technology (such as risk control), the setting of rule parameters (such as rule thresholds) directly affects the operating effect of the business system. At present, the adjustment of rule thresholds mainly relies on manual methods: business personnel need to analyze historical data offline and manually set rule parameters such as thresholds based on the analysis results. This manual method is not only complex in statistical analysis, but also limited by the working hours of people, and it is impossible to modify rule parameters such as thresholds in time for emergencies around the clock.

[0068] In view of the above problems, the embodiments of the present disclosure provide a method for adjusting rule parameters (such as threshold values). The method for adjusting rule parameters (such as threshold values) realizes the automation of rule parameter (such as threshold value) adjustment by automatically acquiring rule configuration, analyzing historical data, performing dynamic optimization and automatically updating parameters (such as threshold values) and other technical means, and can complete the whole process from configuration acquisition to rule parameter (such as threshold value) update without continuous human intervention. Compared with the prior art, the solution of the embodiments of the present disclosure improves the processing efficiency of rule configuration through an automated configuration acquisition and screening mechanism.

[0069] The technical solution of the embodiment of the present disclosure is described in detail below with reference to the accompanying drawings.

[0070] like Figure 1 As shown, in an embodiment of the present disclosure, a rule parameter adjustment method is provided. In an exemplary embodiment of the present disclosure, the rule parameter includes or is a rule threshold. And in the multiple exemplary embodiments described below, the rule threshold will be used as an example for description, but the present disclosure is not limited thereto.

[0071] Combined with reference Figure 1 and Figure 8 The embodiment of the present disclosure also provides a rule parameter (such as threshold) adjustment system that can implement the rule parameter (such as threshold) adjustment method. Figure 8 In the illustrated embodiment, the rule parameter (such as threshold) adjustment system is applied to the risk control field, and may be included or incorporated into a risk control (rule) engine. However, those skilled in the art will appreciate that the rule parameter (such as threshold) adjustment method, system, device, electronic device, program product, and storage medium of the disclosed embodiment may be applied to other occasions, including but not limited to other application scenarios or services of Fintech, or other scenarios or services in the non-Fintech field, such as recruitment, intelligent decision-making (BI), business access, credit scoring, and other scenarios or services involving rule engines. In some specific embodiments, the rule parameter (such as threshold) adjustment scheme of the disclosed embodiment will be described using the risk control scenario as an example, and the disclosed embodiment is not limited to the specific applications listed.

[0072] Continue to refer Figure 1 The rule parameter (such as threshold value) adjustment method may include 110 to 140:

[0073] 110: Obtain current rule configuration information from the rule engine, where the rule configuration information includes the rule configuration to be processed and the corresponding current rule parameters.

[0074] In some embodiments, Figure 8 As shown, a rule engine, such as a risk control (rule) engine, can generate rule configuration information including rule configuration and corresponding rule parameters (such as thresholds). In some embodiments, the rule engine, such as a risk control (rule) engine, can generate the current rule configuration information when the parameters (such as thresholds) are automatically adjusted last time, as further described below. In other embodiments, the current rule configuration information can be generated by other means, such as manually set or partially manually set and partially automatically generated, and the present disclosure does not limit this.

[0075] Accordingly, combined with reference Figure 8 In this step, for example, the rule configuration information (e.g., synchronized by the rule) can be obtained through the configuration acquisition unit, wherein the rule configuration information can be generated by the risk control (rule) engine. In some embodiments, the rule configuration information can also include specific version number and version effective time and other information. In some embodiments, the version number and version effective time and other information can be implemented through other steps of adjusting the rule parameters (e.g., thresholds), as further described below.

[0076] In some embodiments, the step of obtaining the current rule configuration information from the rule engine also includes: 110' (not shown): filtering the obtained rule configuration information according to the preset filtering settings to obtain the rule configuration to be processed and the corresponding current rule parameters (such as thresholds). Specifically, the rule set that needs to be automatically adjusted can be manually set, and the obtained rule configuration information can be filtered to obtain the rule configuration and corresponding parameters (such as thresholds) that need to be monitored, so that the rule configuration that needs to be optimized can be determined in a targeted manner, thereby improving the efficiency of configuration processing. For example, according to the pre-set filtering conditions, the rule configuration that needs to be automatically adjusted and the features involved in these rule configurations and their corresponding thresholds can be filtered out from all rule configurations, for example, the user identity information involved in the risk control scenario, The appropriate feature items contained in the credit information. Here, it should be understood that the information, data and signals involved in this disclosure are all authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data comply with the relevant laws, regulations and standards of relevant countries and regions.

[0077] In some embodiments, the relevant log information of the configuration file obtained (such as synchronization) can also be stored in the log (data) library, so that the data in the log library can be used to generate relevant reports within the parameter (such as threshold) adjustment period. Specifically, the report may include but is not limited to: adjustment records of rule parameters (such as thresholds), comparative analysis of effects before and after adjustment, statistics on the triggering of each rule, time distribution of policy adjustment and other dimensions. These reports can provide data support and decision-making basis for subsequent policy optimization and rule adjustment, and are also convenient for relevant personnel to track and evaluate the actual effect of rule adjustment. In some embodiments, these reports can also be automatically generated and pushed according to different time dimensions (such as daily, weekly, monthly, etc.).

[0078] 120: Get user history data associated with the current rule parameters.

[0079] In some embodiments, Figure 8 As shown and described above, the obtained rule configuration information is screened to obtain the rule configuration and corresponding parameters (such as thresholds) that need to be monitored, thereby determining the rule list that needs to be updated this time and the rule parameters (such as thresholds) involved. Figure 8 It is further shown that the system can determine the rule list that needs to be updated this time and its corresponding rule thresholds by reading the configuration file.

[0080] In some embodiments, the system will screen the configuration file according to the manually set dimensions described above, such as the preset update cycle, the given specific business type, the rule priority, etc., to determine the list of rules that need to be updated this time. In some embodiments, the update cycle may include but is not limited to daily updates, weekly updates, monthly updates, and other time dimensions. In some embodiments, the business type may include but is not limited to different business links such as pre-loan, mid-loan, and post-loan.

[0081] In a further embodiment, the rule list to be updated this time, i.e., the user history data associated with, for example, the filtered current rule parameters (such as thresholds), can be determined accordingly, and the user history data can be pulled from the data source accordingly. In some embodiments, the data source can be stored in any suitable form and stored in any suitable database. In a specific example, the database storing the data source includes, for example, HIVE, but other databases are conceivable.

[0082] Accordingly, the historical data obtained here will be used in subsequent dynamic optimization processing to determine more optimal rule parameters (such as thresholds).

[0083] 130: Based on the preset tuning target of the indicator to be optimized, dynamic optimization processing is performed using user historical data to obtain optimized rule parameters.

[0084] In the embodiments of the present disclosure, for example, Figure 8 As shown, the rule parameters (such as thresholds) can be dynamically optimized based on the preset tuning target of the indicator to be optimized. Figure 8 As shown, the system can determine the indicator to be optimized and the tuning target of the indicator to be optimized (which can be referred to as the tuning target) according to given conditions (such as business needs). In the embodiments of the present disclosure, the indicator to be optimized can involve different consideration indicators, such as but not limited to various risk performance indicators. In a specific example, the indicator to be optimized can include a downgraded overdue rate. In this article, the downgraded overdue rate has the usual meaning in the financial or credit field, more specifically in the risk control field, and specifically refers to the ratio of loans downgraded from a normal repayment status to an overdue status.

[0085] In the embodiments of the present disclosure, the preset tuning target of the indicator to be optimized may be any suitable value, including but not limited to, an (absolute) numerical value, a comparative value, or a combination of the two. As an example, the preset tuning target of the indicator to be optimized (such as the downgraded overdue rate) may be an (absolute) numerical value, such as 5% (or less). In another example, the preset tuning target of the indicator to be optimized (such as the downgraded overdue rate) may be an (absolute) numerical value or a comparative value, such as being optimized (such as decreased) by more than 20% compared to the previously set indicator to be optimized (such as the downgraded overdue rate) or the indicator to be optimized (such as the downgraded overdue rate) obtained based on historical data statistics. In another example, the preset tuning target of the indicator to be optimized (such as the downgraded overdue rate) may be a combination of an (absolute) numerical value and a comparative value. As mentioned above, the indicators to be optimized and their tuning targets can be flexibly configured according to given conditions, such as actual business needs. These given conditions include but are not limited to adjustment cycle settings (such as whether to include time factors such as weekends), business volume indicators (such as maximum reduction ratio, minimum submission volume, submission volume target, etc.), business efficiency indicators (such as approval rate target, minimum approval rate, etc.), and other aspects.

[0086] In the embodiments of the present disclosure, the acquired historical data of the user, including various characteristic data of the user and various indicators of the user, such as the performance during and after the loan, can be utilized, and a preset algorithm (such as the XGBoost algorithm) can be used to analyze and determine the correlation between the user characteristic data and various performance indicators, and the associated rule parameters (such as thresholds) can be dynamically optimized based on the correlation relationship related to the indicators to be optimized, so as to obtain the optimized rule parameters (such as thresholds).

[0087] In some embodiments, Figure 2 As shown, the preset tuning target based on the indicator to be optimized uses the user historical data to perform dynamic optimization processing to obtain the optimized rule parameters (such as thresholds), which may include 210 to 230:

[0088] 210: Based on the user historical data, use a preset algorithm to construct at least one association model between the feature and at least one indicator.

[0089] In these embodiments, the user history data includes user history feature data and history indicators. In these embodiments, the at least one indicator is preferably a plurality of indicators and includes an indicator to be optimized.

[0090] In some embodiments, the user's historical feature data may include, but is not limited to, the user's basic attribute information (such as age, occupation, income level, etc.), behavioral characteristics (such as historical transaction frequency, transaction amount distribution, etc.), credit characteristics (such as historical overdue situation, credit score, etc.), and other feature data in multiple dimensions. In some embodiments, the indicators, such as historical indicators, mainly include various business indicators. For example, when used in the field of risk control, the indicators (such as historical indicators) may include various risk performance indicators, including but not limited to business indicators during and after the loan, such as default rate, fraud rate, overdue rate and other key business indicators, more specifically, such as downgrade overdue rate.

[0091] In the embodiment of the present disclosure, the association relationship between the feature and the indicator can be obtained by constructing an association model between the feature and the indicator. In the embodiment of the present disclosure, the association relationship between the feature and the indicator not only covers the direct association between a single feature and the indicator, but also covers the influence of the interaction between the features on the indicator, which is not limited by the present disclosure.

[0092] More specifically, a preset algorithm, such as a preset machine learning algorithm, such as an XGBoost algorithm, may be used to analyze the correlation between features and indicators.

[0093] like Figure 3 As shown, the step of constructing at least one association model between a feature and at least one indicator based on the user historical data using a preset algorithm may include steps 310 to 330:

[0094] 310: Obtain training sample data for training at least one association model according to the user historical feature data and the historical execution results of the user historical data.

[0095] In these embodiments, the training sample data includes a training set and a validation set. In an optional embodiment, the training sample data may also include a test set, which may be used, for example, in the optional 620 below, for example, in the process of dynamically optimizing the rule parameters (such as thresholds), using the association model to perform predictive analysis to obtain a prediction result, as further described below.

[0096] In some embodiments, the training set, validation set, and optional test set may be directly extracted from the user historical feature data and historical execution results of the user historical data. In other embodiments, the user historical data, more specifically the user historical feature data and / or historical execution results, may be preprocessed, and the training set, validation set, and optional test set may be obtained from the preprocessed data.

[0097] In some embodiments, the preprocessing may include cleaning, standardization, outlier processing, merging, and other preprocessing of feature data.

[0098] As mentioned above, the user's historical feature data may include feature data of multiple dimensions, such as basic information dimension (including user age, occupation type, income level, education level, etc.), behavioral feature dimension (including historical transaction frequency, transaction amount distribution, transaction time characteristics, etc.), credit performance dimension (including historical overdue situation, repayment performance, credit score changes, etc.). Therefore, according to needs, feature data can be directly extracted or part or all of the feature data can be preprocessed to obtain feature data in training sample data.

[0099] In some embodiments, historical performance data of users may be collected, including but not limited to key risk indicators such as whether a default has occurred, the number of overdue days, and fraud marks. Similarly, these historical performance data may be directly extracted as Y labels of training sample data (corresponding to at least one, for example, multiple indicators mentioned above) as needed, for example, when these historical performance data directly correspond to at least one, for example, multiple indicators mentioned above. In other embodiments, the historical performance data may also be preprocessed to obtain the Y labels of the training sample data.

[0100] In some embodiments, when constructing training samples, any data segmentation strategy can be used to obtain the above-mentioned training set, validation set, and optional test set. For example, the processed data can be divided into a training set, a validation set, and an optional test set according to a preset ratio (such as 7:3 or 8:2 for the training set and the validation set, or 7:2:1 for the training set, the validation set, and the test set).

[0101] The present disclosure does not limit the method of obtaining the above-mentioned training set and verification set and the optional test set and the method of preprocessing.

[0102] 320: Based on the training set, use the XGBoost algorithm to iteratively construct a decision tree sequence to obtain at least one trained association model, each association model including a decision tree set.

[0103] In this step, the XGBoost (eXtreme Gradient Boosting) algorithm can be used to iteratively train to obtain at least one trained association model. The XGBoost (eXtreme Gradient Boosting) algorithm is an ensemble learning algorithm based on gradient boosting decision trees. In some embodiments, the hyperparameters of the model can be determined, including but not limited to the maximum tree depth, the minimum number of leaf node samples, the learning rate, the feature sampling ratio, etc.

[0104] In this step, at least one, such as multiple association models corresponding to at least one predetermined business indicator, such as multiple business indicators, are constructed to obtain association relationships between different indicators and related features. The at least one, such as multiple association models, include an association model corresponding to the indicator to be optimized.

[0105] During the training process, the XGBoost algorithm builds a sequence of decision trees by step-by-step iteration. For each iteration, a new decision tree is built based on the prediction error of the previous decision tree. In a specific embodiment, the first multi-layer decision tree can be built based on the training data of the training set, and each node divides the data into two parts by selecting the optimal split point. Then, a new decision tree can be iteratively built by calculating the error between the prediction results and the true value of all the decision trees that have been built.

[0106] In the disclosed embodiment, any suitable iteration termination condition can be set as needed. In some embodiments, the performance index of the currently constructed decision tree set can be verified in each iteration, and it can be determined whether the preset iteration termination condition is met according to the performance index. In the disclosed embodiment, the performance index may include prediction effect or other evaluation indicators, including but not limited to AUC (area under the curve), KS value (Kolmogorov-Smirnov statistic), precision, recall rate, etc. Thus, the iteration can be terminated when the performance index of the association model is improved by less than a certain threshold or the performance index decreases, for example, when the error fluctuation between the prediction result of the decision tree and the true value is less than a preset threshold. As a supplementary or alternative embodiment, other iteration termination conditions can be set, including but not limited to reaching the maximum number of iterations.

[0107] By constructing the at least one association model, the correlation between the business indicator and the relevant features can be obtained. For example, there may be a significant correlation between the user's historical overdue record and future default risk, or there may be a significant correlation between certain transaction behavior characteristics and fraud risk.

[0108] 220: Determine the association model corresponding to the indicator to be optimized, and determine the feature to be optimized and its corresponding rule parameters according to the corresponding association model.

[0109] In these embodiments, the association model corresponding to the indicator to be optimized may be determined in this step. For example, if the indicator to be optimized is the downgraded overdue rate, an association model specifically used to predict the downgraded overdue rate risk may be selected.

[0110] In these embodiments, since each association model includes a set of decision trees, in this step, the features related to the association model can be determined according to the association model that can determine the association relationship between the features and the indicators. In some exemplary embodiments, for example, a model mapping table can be maintained to record the correspondence between each association model and its corresponding indicator.

[0111] In the embodiments of the present disclosure, each rule may include at least one corresponding feature variable and a judgment condition of the feature variable. In this embodiment, the judgment condition of the feature variable may include a comparison operator and a threshold value, wherein the comparison operator may include any one of the following comparison operators: greater than (>), less than (<), equal to (=), greater than or equal to (≥), less than or equal to (≤) and not equal to (!=), and the threshold value is a specific value used for comparison. Accordingly, in these embodiments, the feature-associated rules and the corresponding rule thresholds can be determined based on the determined features and the optional rule configuration information.

[0112] In some embodiments, the features associated with the correlation model corresponding to the indicator to be optimized may be determined as the features to be optimized, and then the corresponding rule thresholds or other rule parameters may be determined.

[0113] In other embodiments, the associated features may be screened to select features that have a certain influence on the index to be optimized. Figure 4 As shown, the step of determining the association model corresponding to the indicator to be optimized, and determining the feature to be optimized and its corresponding rule parameter (such as a threshold) according to the corresponding association model may include 410 to 430:

[0114] 410: Determine the association model corresponding to the indicator to be optimized.

[0115] In these embodiments, the association model corresponding to the indicator to be optimized may be determined, similar to that described above.

[0116] 420: Determine the influence of each feature in the decision tree set of the corresponding association model on the optimization index.

[0117] In these embodiments, in this step, the influence of each feature in the decision tree set of the corresponding association model can be determined. In these embodiments, since each association model includes a decision tree set, these decision trees jointly reflect the influence of the feature on the target indicator. Therefore, the influence of each feature on the indicator to be optimized can be quantitatively evaluated based on the decision tree set in the selected association model.

[0118] In some specific embodiments, the situation in which each feature in the decision tree set is used as a split node can be analyzed, and the influence of the relevant features on the index to be optimized can be determined based on the split degree information of the decision tree set. In some embodiments, the influence can be determined based on the split position of the feature, that is, the node position where it appears in the decision tree. For example, the closer to the root node of the tree, the stronger its influence. For example, in a model for predicting the risk of downgraded overdue rate, if the feature "historical overdue times" often appears in the top node of the decision tree, and its split can significantly improve the prediction effect, then the feature has a higher influence. In other embodiments, the influence can be determined based on the number of times the feature is split in the decision tree. For example, the more times a feature appears, the more important the feature is. In some embodiments, the reduction in prediction error brought about by each split can be considered. The larger the reduction, the more significant the influence of the feature corresponding to the split.

[0119] As an optional embodiment, when determining the influence, the positive or negative correlation between the feature and the indicator to be optimized may also be determined, such as positive correlation, negative correlation or zero correlation. For example, features with zero correlation may be screened out.

[0120] 430: Determine the feature to be optimized according to the influence, and obtain the rule parameters corresponding to the feature to be optimized.

[0121] In these embodiments, in this step, based on the previously determined feature influence, features that have a significant impact on the index to be optimized can be screened out as features to be optimized. In some embodiments, the screening process can be performed using an influence threshold or a sorting method.

[0122] In these embodiments, for the determined features to be optimized, the corresponding rule parameters (such as thresholds) can be obtained from the current rule configuration information. In some embodiments, the current parameter (such as threshold) settings of the rules related to these features to be optimized can be recorded as the starting point for subsequent dynamic optimization.

[0123] 230: Based on the preset tuning target of the indicator to be optimized, the rule parameters corresponding to the feature to be optimized are dynamically optimized using the corresponding association model to obtain the optimized rule parameters.

[0124] In these embodiments, in this step, the rule parameters (such as thresholds) of the determined features to be optimized can be dynamically optimized according to the preset optimization targets. Here, the rule parameters (such as thresholds) can be iteratively adjusted based on the optimization targets of the indicators to be optimized using the association model.

[0125] like Figure 5As shown, a specific embodiment of dynamically optimizing rule parameters (such as thresholds) is shown. Specifically, based on the preset optimization target of the indicator to be optimized, the rule parameters (such as thresholds) corresponding to the feature to be optimized are dynamically optimized using the corresponding association model to obtain the optimized rule parameters (such as thresholds), which may include 510 to 540:

[0126] 510: Determine the correlation between each feature to be optimized and the indicator to be optimized based on the association model corresponding to the indicator to be optimized.

[0127] In these embodiments, in this step, the correlation model constructed by the XGBoost algorithm can be used to obtain the correlation between the feature to be optimized and the target indicator. This correlation may include the direction of correlation (positive correlation or negative correlation) and may also include a quantitative assessment of the degree of correlation. In some embodiments, the correlation between the feature and the indicator determined by the association model may include linear and nonlinear relationships. As an example, the number of historical overdue payments may be linearly positively correlated with the risk of downgraded overdue rate.

[0128] 520: Determine, based on the correlation, the adjustment direction and / or adjustment range of the rule parameters corresponding to each feature to be optimized.

[0129] In these embodiments, in this step, based on the correlation between the previously determined feature and the indicator to be optimized, the appropriate adjustment direction and adjustment range can be determined for the rule parameter (such as the threshold) of each feature to be optimized. In some embodiments, for example, the adjustment direction of the rule parameter (such as the threshold) can be determined based on the qualitative positive and negative correlation. For example, for a feature (such as the number of historical overdue payments) that is significantly positively correlated with the indicator to be optimized, such as the risk of downgraded overdue rate, it can be determined that the adjustment direction of the rule parameter (such as the threshold) is to reduce the rule parameter (such as the threshold). In some embodiments, for example, the adjustment range of the rule parameter (such as the threshold) can be determined based on the quantitative correlation degree. Accordingly, the initially updated rule parameter (such as the threshold) determined based on this can fall within the adjustment range. In other embodiments, the adjustment direction and adjustment range of the rule parameter (such as the threshold) can be determined simultaneously. For example, the initially updated rule parameter (such as the threshold) can be determined to correspond to the end point of the adjustment range, and the adjustment direction of the subsequently updated rule parameter, such as the threshold (if any), is to adjust to the other end point of the adjustment range.

[0130] 530: Dynamically adjust the rule parameters corresponding to each feature to be optimized along the adjustment direction and / or within the adjustment range to obtain optimized rule parameters that meet the preset tuning target.

[0131] In these embodiments, in this step, the rule parameters (such as thresholds) may be dynamically adjusted and optimized based on the previously determined adjustment direction and / or adjustment range.

[0132] In a further embodiment, Figure 6 As shown, the dynamically adjusting the rule parameters (such as thresholds) corresponding to each feature to be optimized along the adjustment direction and / or within the adjustment range to obtain the optimized rule parameters (such as thresholds) that meet the preset tuning target may include 610 to 650:

[0133] 610: Determine the updated rule parameters corresponding to each feature to be optimized.

[0134] In these embodiments, in this step, an initial update parameter (such as a threshold) may be determined for each feature to be optimized.

[0135] In some embodiments, the setting of the initial update parameter (eg, threshold) may be determined according to the adjustment direction and / or adjustment range.

[0136] In some specific embodiments, for features with determined adjustment directions, for example, the original parameter (such as a threshold) or the parameter obtained by adding the original parameter (such as a threshold) to a given step size (such as a threshold) can be selected as the initial update parameter (such as a threshold), that is, as the initial update point.

[0137] In some specific embodiments, for a feature with a determined adjustment range, any suitable point within the adjustment range, such as an endpoint, may be used as an initial update parameter (such as a threshold), that is, as an initial update point.

[0138] In some specific embodiments, for features with a determined adjustment range and adjustment direction, any suitable point within the adjustment range, such as an endpoint, such as an endpoint determined according to the adjustment direction, can be used as an initial update parameter (such as a threshold) (i.e., the endpoint is used as the initial update point). For example, if the adjustment direction is an increasing direction (such as when the tuning goal is to reduce the value of the index to be optimized and the feature is negatively correlated with the index, or when the tuning goal is to increase the value of the index to be optimized and the feature is positively correlated with the index), the lower end point of the adjustment range is used as the initial update point, and vice versa.

[0139] In some embodiments, the initial update position may be determined based on the strength of the correlation. For example, for a feature that is strongly correlated with the indicator to be optimized, an adjustment range endpoint may be selected as the initial value to obtain the maximum optimization effect.

[0140] In some other embodiments, the interaction relationship between multiple features to be optimized may be considered simultaneously, and their initial update parameters (such as thresholds) may be set in a coordinated manner. As an example, assuming that there are multiple features A, B, and C to be optimized, on the basis of setting the threshold A' corresponding to feature A, the correlation between feature A and feature B can be taken into account, and based on the initial update threshold A' corresponding to feature A, the initial update threshold B' corresponding to feature B is set, while the initial update threshold C' corresponding to feature C is set independently; for example, more specifically, for example, based on the positive correlation between feature A and feature B, when the initial update threshold A' is set to select one end point of the threshold adjustment range of feature A (such as the endpoint representing the lower limit of the threshold), according to the positive correlation between the two, the initial update threshold B' can be set to select the same side endpoint of the threshold adjustment range of feature B, that is, select the endpoint representing the lower limit of the threshold adjustment range of B; in addition, specifically, for example, based on the negative correlation between feature A and feature B, when the initial update threshold A' is set to select one end point of the threshold adjustment range of feature A (such as the endpoint representing the lower limit of the threshold), according to the negative correlation between the two, the initial update threshold B' can be set to select the opposite side endpoint of the threshold adjustment range of feature B, that is, select the endpoint representing the upper limit of the threshold of the threshold adjustment range of B.

[0141] 620: Based on the updated rule parameters, the historical sample data is predicted and analyzed using the association model to obtain a prediction result.

[0142] In these embodiments, this step uses the above-mentioned preset algorithm, such as the association model constructed by the XGBoost algorithm, to predict and evaluate the effect after adopting the new parameters (such as thresholds). Specifically, the updated rule parameters (such as thresholds) can be applied to the association model, and the characteristic values ​​of the corresponding characteristic items of the historical sample data can be processed using the association model to obtain the prediction results. In some embodiments, the historical sample data is, for example, the aforementioned test set, but the present disclosure is not limited to this.

[0143] In some embodiments, the prediction result may be a prediction result that has undergone a variety of post-processing, including but not limited to averaging, median, and other processing.

[0144] 630: Determine whether the prediction result meets the preset tuning target.

[0145] In these embodiments, this step judges the optimization effect of the current parameter (such as the threshold) by comparing the prediction result with the preset tuning target.

[0146] 640: If the prediction result does not meet the preset tuning target, continue to adjust the rule parameters along the adjustment direction and / or within the adjustment range according to the preset step size to obtain further updated rule parameters (such as thresholds).

[0147] In these embodiments, this step is to find the optimal parameter (such as threshold) by iterative adjustment. In some embodiments, a fixed step size can be used. In some embodiments, an adaptive step size strategy can be used. Initially use a larger step size to quickly explore, and automatically reduce the step size when approaching the target to achieve fine adjustment. For example, the step size can be set to a certain proportion of the current gap with the target.

[0148] In some other embodiments, a strategy of collaborative updating of multiple features may be adopted. Here, the interactive relationship between multiple features to be optimized may be considered, and the interrelated parameters (such as thresholds) corresponding to the associated feature combinations may be updated, such as but not limited to, alternate updating, parallel updating, partial parallel updating, nested updating, and the like strategies for updating the rule parameters (such as thresholds) corresponding to multiple features.

[0149] In some embodiments of the present disclosure, alternating update means that when multiple parameters are updated, a strategy of updating in turns is adopted. For example, if there are three parameters A, B, and C, parameter A is updated in round 1, parameter B is updated in round 2, and parameter C is updated in round 3, and this process is repeated for subsequent updates.

[0150] In some embodiments of the present disclosure, parallel updating means that when multiple parameters are updated, a strategy of updating them in each update round is adopted. For example, if there are three parameters A, B, and C, then parameters A, B, and C are updated in round 1, round 2, and round 3.

[0151] In some embodiments of the present disclosure, nested update means that when multiple parameters are updated in each update round, they are updated in the update round in a preset order until the preset result meets the preset tuning target. For example, there are three parameters, A, B, and C. When updating in this round, they can be updated in the order of parameter A first, parameter B second, and parameter C last. After each parameter is updated, it can be determined whether the preset result meets the preset tuning target. If not, continue to update the next sequence of parameters. For example, after parameter A is updated, if the preset result does not meet the preset tuning target, parameter B is updated; if after parameter B is updated, the preset result meets the preset tuning target, the update can be terminated, otherwise, parameter C is updated.

[0152] As an example, for example, the threshold A' corresponding to feature A can be updated first, and the thresholds B' and C' corresponding to features B and C remain unchanged. If the prediction result of the association model using the updated threshold A' and the unupdated thresholds B' and C' still does not meet the preset tuning target in step 630, the threshold B' corresponding to feature B can be updated, and the thresholds A' and C' of features A and C remain unchanged. If the prediction result still does not meet the preset tuning target, the thresholds are updated alternately in sequence.

[0153] In other embodiments, when multiple features are updated collaboratively, the interaction between multiple features to be optimized can also be considered. As an example, for example, considering that features A, B, and C have a synchronous association relationship, the threshold A' corresponding to feature A, the threshold B' corresponding to feature B, and the threshold C' corresponding to feature C can be updated synchronously and in parallel. As another example, for example, considering that features A and B have a synchronous association relationship, the threshold A' corresponding to feature A and the threshold B' corresponding to feature B can be partially updated synchronously, and the thresholds (A', B') and threshold C' can be updated alternately in sequence. As another example, the threshold A' corresponding to feature A, the threshold B' corresponding to feature B, and the threshold C' corresponding to feature C can be updated in a nested manner, so that, for example, the threshold C' of feature C can be updated first within its corresponding threshold range; if the prediction result still does not meet the preset tuning target, then the threshold B' of feature B is updated within its corresponding threshold range; if the prediction result still does not meet the preset tuning target, then the threshold A' of feature A is updated within its corresponding threshold range, and vice versa.

[0154] 650: If the prediction result meets the preset tuning target, the updated rule parameters are determined as optimized rule parameters.

[0155] In a preferred embodiment, a log may be generated when updated rule parameters (such as thresholds) that meet the requirements are obtained, and recorded in a log such as Figure 8 In the log library shown.

[0156] As an alternative embodiment, if suitable parameters (such as thresholds) cannot be found through automatic optimization, a manual intervention mechanism may be introduced to perform analysis and verification by manually setting parameters (such as thresholds).

[0157] 140: Apply the optimized rule parameters to the rule engine to obtain new rule configuration information.

[0158] In these embodiments, this step applies the new rule parameters (such as thresholds) obtained by system optimization to the rule engine.

[0159] Optionally, specific version information may be generated for the new rule configuration, including the version number and the effective time of the version. Optionally, the version information may also include parameter (such as threshold) optimization details and / or configuration update details, such as but not limited to optimization time, optimized index, optimized feature, optimized parameter (such as threshold), original parameter (such as threshold), and other information.

[0160] In some embodiments, manual review of optimized rule parameters (such as thresholds) may also be introduced. Here, applying the optimized rule parameters (such as thresholds) to the rule engine to obtain new rule configuration information includes: determining whether the optimized rule parameters (such as thresholds) require manual review. Figure 8 As shown, if manual review is required, the rule parameters (such as thresholds) that have passed manual review will be applied to the rule engine, such as the risk control (rule) engine; if manual review is not required, the optimized rule parameters (such as thresholds) will be directly applied to the rule engine, such as the risk control (rule) engine.

[0161] As an optional embodiment, the rule parameter (such as threshold) adjustment method may further include 710 to 730:

[0162] 710: Obtain user performance data corresponding to the rule engine to which the optimized rule parameters are applied.

[0163] In these embodiments, after the new rule parameters (such as thresholds) are put into use, user performance data can be continuously collected, that is, the "user historical data" from the time the new rule parameters (such as thresholds) are put into use to the current time, similarly including feature data and indicator data, which will not be elaborated here.

[0164] 720: Evaluate whether the actual indicator value corresponding to the indicator to be optimized in the user performance data meets the preset tuning target.

[0165] In these embodiments, the collected user performance data can be analyzed and evaluated to observe whether the actual indicator value corresponding to the indicator to be optimized in the user performance data meets the preset tuning target. Similarly, the indicator data can also be processed in a related manner, such as averaging, median, etc. For example, when the optimization goal is to reduce the degraded overdue rate, it can be analyzed whether the user's related overdue performance under the new rule parameters (such as thresholds) has achieved the expected reduction target. Optionally, the evaluation process can last for a certain period of time to ensure that stable and reliable evaluation results are obtained. Alternatively, the evaluation can at least partially use user performance data after the new rule parameters (such as thresholds) are put into use, and optionally can include user historical data before the new rule parameters (such as thresholds) are put into use.

[0166] 730: When the actual indicator value does not meet the preset tuning target, the dynamic optimization process is re-performed using the corresponding association model based on the preset tuning target of the indicator to be optimized.

[0167] In these embodiments, if the evaluation results show that the actual effect does not reach the expected goal, a new round of optimization process can be started. The implementation of this step can refer to Figures 2 to 6 The embodiments shown are not described in detail here.

[0168] The above describes the rule parameter (such as threshold) adjustment method of multiple embodiments of the present disclosure.

[0169] In a further optional embodiment, the rule parameter (such as threshold) adjustment method can periodically perform all the above optimization processes according to a preset automatic adjustment time interval to achieve all-weather and closed-loop (such as Figure 8 As an example and not limitation, the cycle can be set according to business needs, such as being set to be executed every minute, every fifteen minutes, every thirty minutes, every hour, 2 hours, 4 hours, 6 hours, 8 hours, every day, every week, or every month.

[0170] In a further optional embodiment, an analysis report can be automatically generated during the optimization process, and the report includes but is not limited to analysis data in multiple dimensions, such as the changing trend of rule parameters (such as thresholds), changes in key business indicators, achievement of optimization goals, and records of abnormal situations.

[0171] The rule parameter (such as threshold) adjustment method of the disclosed embodiment effectively solves the problem of low efficiency in traditional manual acquisition and screening of rule configuration information through an automated rule configuration acquisition and screening mechanism, and realizes accurate screening and efficient acquisition of rule configuration.

[0172] The rule parameter (such as threshold) adjustment method of the further embodiment of the present disclosure adopts a feature-indicator association analysis method based on machine learning, which overcomes the limitation of the traditional method of over-reliance on manual experience. The association model constructed by the XGBoost algorithm can accurately quantify the influence of the feature on the business indicator, which is conducive to the automatic optimization and adjustment of parameters (such as thresholds).

[0173] In addition, the rule parameter (such as threshold) adjustment method of a further embodiment of the present disclosure realizes a closed-loop feedback mechanism for parameter (such as threshold) adjustment, and realizes closed-loop optimization of parameter (such as threshold) adjustment by continuously monitoring the actual effect after optimization and timely adjusting the optimization strategy according to the feedback results.

[0174] In some embodiments of the present disclosure, Fig. 9 As shown, a rule parameter (such as threshold) adjustment device 900 is also provided, which may at least include: a configuration acquisition unit 910, a data acquisition unit 920, an optimization unit 930 and an application unit 940.

[0175] The configuration acquisition unit 910 is configured to acquire current rule configuration information from a rule engine, where the rule configuration information includes current rule parameters (such as thresholds).

[0176] The data acquisition unit 920 is configured to acquire user history data associated with the current rule parameter (such as a threshold).

[0177] The optimization unit 930 is configured to perform dynamic optimization processing based on a preset optimization target of the indicator to be optimized using the user historical data to obtain optimized rule parameters (such as thresholds).

[0178] The application unit 940 is configured to apply the optimized rule parameters (such as thresholds) to the rule engine to obtain new rule configuration information.

[0179] In the embodiments of the present disclosure, the device embodiments may refer to the method embodiments and may implement the steps described in the method embodiments, and the features of the device embodiments may be incorporated into the method embodiments in a non-contradictory manner, and vice versa.

[0180] In some embodiments of the present disclosure, Fig.10 As shown, a schematic diagram of an electronic device 1000 that can implement the method of the disclosed embodiment is provided, and in some embodiments, more or fewer electronic devices than shown may be included. In some embodiments, it can be implemented using a single or multiple electronic devices. In some embodiments, it can be implemented using cloud or distributed electronic devices.

[0181] like Fig.10 As shown, the electronic device 1000 includes a processor 1001, which can perform various appropriate operations and processes according to the programs and / or data stored in the read-only memory (ROM) 1002 or the programs and / or data loaded from the storage part 1008 to the random access memory (RAM) 1003. CPU1001 can be a multi-core processor, or it can include multiple processors. In some embodiments, the processor 1001 can include a general-purpose main processor and one or more special coprocessors, such as a graphics processing unit (GPU), a neural network processor (NPU), a digital signal processor (DSP), etc. In RAM 1003, various programs and data required for the operation of the electronic device 1000 are also stored. CPU 1001, ROM 1002 and RAM 1003 are connected to each other via bus 1004. Input / output (I / O) interface 1005 is also connected to bus 1004.

[0182] The following components are connected to the I / O interface 1005: an input section 1006 including a keyboard, a mouse, a touch screen, etc.; an output section 1007 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1008 including a hard disk, etc.; and a communication section 1009 including a network interface card such as a LAN card, a modem, etc. The communication section 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to the I / O interface 1005 as needed. A removable medium 1011, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1010 as needed, so that a computer program read therefrom is installed into the storage section 1008 as needed. Fig.10 Only some components are schematically shown in the figure, which does not mean that the computer system 1000 only includes Fig.10 Components shown.

[0183] The systems, devices, modules or units described in the above embodiments may be implemented by a computer or its associated components. The computer may be, for example, a mobile terminal, a smart phone, a personal computer, a laptop computer, a vehicle-mounted human-computer interaction device, a personal digital assistant, a media player, a navigation device, a game console, a tablet computer, a wearable device, a smart TV, an Internet of Things system, a smart home, an industrial computer, a server or a combination thereof.

[0184] Although not shown, in an embodiment of the present disclosure, a program product is provided, including a computer program, and when the computer program is executed by a processor, the method of any one of the embodiments of the present disclosure is implemented.

[0185] Although not shown, in an embodiment of the present disclosure, a storage medium is provided, the storage medium storing a computer program, the computer program being configured to implement any method of the embodiments of the present disclosure when executed.

[0186] The storage media in the embodiments of the present disclosure include permanent and non-permanent, removable and non-removable items that can be implemented by any method or technology to store information. Examples of storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission medium that can be used to store information that can be accessed by a computing device.

[0187] The methods, programs, systems, devices, etc. of the embodiments of the present disclosure may be executed or implemented in a single or multiple networked computers, or may be practiced in a distributed computing environment. In the embodiments of the present specification, in these distributed computing environments, tasks may be performed by remote processing devices connected via a communication network.

[0188] Those skilled in the art should understand that the embodiments of the present specification may be provided as methods, systems or computer program products. Therefore, those skilled in the art may imagine that the functional modules / units or controllers and related method steps described in the above embodiments may be implemented in software, hardware or a combination of software / hardware.

[0189] Unless explicitly stated, the actions or steps of the methods, programs, and embodiments of the present disclosure do not have to be performed in a specific order and can still achieve the desired results. In some implementations, multitasking and parallel processing are also possible or may be advantageous.

[0190] In this article, multiple embodiments of the present disclosure are described, but for the sake of brevity, the description of each embodiment is not exhaustive, and the same or similar features or parts between the embodiments may be omitted. In this article, "one embodiment", "some embodiments", "example", "specific example", or "some examples" are meant to be applicable to at least one embodiment or example according to the present disclosure, rather than all embodiments. The above terms do not necessarily mean to refer to the same embodiment or example. In the absence of contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples.

[0191] The exemplary systems and methods of the present disclosure have been specifically shown and described with reference to the above embodiments, which are merely examples of the best modes for implementing the present systems and methods. It will be appreciated by those skilled in the art that various changes may be made to the embodiments of the systems and methods described herein when implementing the present systems and / or methods without departing from the spirit and scope of the present disclosure as defined in the appended claims.

Claims

1. A rule parameter adjustment method, characterized in that: include: Acquire current rule configuration information from the rule engine, wherein the rule configuration information includes the rule configuration to be processed and the corresponding current rule parameters; Acquire user history data associated with the current rule parameters; Based on the preset tuning target of the indicator to be optimized, the user historical data is used to perform dynamic optimization processing to obtain optimized rule parameters; as well as The optimized rule parameters are applied to the rule engine to obtain new rule configuration information.

2. The method according to claim 1, characterized in that The preset optimization target based on the indicator to be optimized uses the user historical data to perform dynamic optimization processing to obtain optimized rule parameters, including: Based on the user historical data, using a preset algorithm to construct at least one association model between a feature and at least one indicator, wherein the user historical data includes user historical feature data and historical indicators, and the at least one indicator includes the indicator to be optimized; and Determine the correlation model corresponding to the indicator to be optimized, and determine the feature to be optimized and its corresponding rule parameters according to the corresponding correlation model; Based on the preset tuning target of the indicator to be optimized, the corresponding association model is used to dynamically optimize the rule parameters corresponding to the feature to be optimized to obtain the optimized rule parameters.

3. The method according to claim 2, characterized in that The preset algorithm is the XGBoost algorithm. The step of constructing at least one association model between a feature and at least one indicator based on the user historical data using a preset algorithm includes: Obtaining training sample data for training the at least one association model according to the user history feature data and the history execution result of the user history data, wherein the training sample data includes a training set and a validation set; and Based on the training set, using the XGBoost algorithm to iteratively construct a decision tree sequence to obtain at least one trained association model, each association model including a decision tree set; For each iteration, a new decision tree is constructed based on the prediction error of the previous decision tree, and the performance index of the association model is evaluated using the validation set until the performance index meets the preset iteration termination condition.

4. The method according to claim 2, characterized in that: The step of determining the association model corresponding to the indicator to be optimized, and using the corresponding association model to determine the feature to be optimized and its corresponding rule parameter includes: Determine the correlation model corresponding to the indicator to be optimized; Determine the influence of each feature in the decision tree set of the corresponding association model on the indicator to be optimized; and The feature to be optimized is determined according to the influence, and a rule parameter corresponding to the feature to be optimized is obtained.

5. The method according to claim 2, characterized in that: Based on the preset tuning target of the indicator to be optimized, the rule parameters corresponding to the feature to be optimized are dynamically optimized using the corresponding association model to obtain the optimized rule parameters, including: Based on the association model corresponding to the indicator to be optimized, determining the correlation between each feature to be optimized and the indicator to be optimized; Determining, based on the correlation, an adjustment direction and / or adjustment range of the rule parameters corresponding to each feature to be optimized; and The rule parameters corresponding to each feature to be optimized are dynamically adjusted along the adjustment direction and / or within the adjustment range to obtain the optimized rule parameters that meet the preset tuning objectives.

6. The method according to claim 5, characterized in that The dynamically adjusting the rule parameters corresponding to each feature to be optimized along the adjustment direction and / or within the adjustment range includes: Determine the updated rule parameters corresponding to each feature to be optimized; Based on the updated rule parameters, using the association model to perform prediction analysis on the historical sample data to obtain a prediction result; If the prediction result does not meet the preset tuning target, the rule parameter is further adjusted along the adjustment direction and / or within the adjustment range according to a preset step size to obtain a further updated rule parameter.

7. The method according to any one of claims 2 to 6, characterized in that Also includes: Acquire user performance data corresponding to the rule engine to which the optimized rule parameters are applied; Evaluate whether the actual indicator value corresponding to the indicator to be optimized in the user performance data meets the preset tuning target; as well as When the actual indicator value does not meet the preset optimization target, the dynamic optimization process is re-performed using the corresponding association model based on the preset optimization target of the indicator to be optimized.

8. A rule parameter adjustment device, characterized in that: include: A configuration acquisition unit, used to acquire current rule configuration information from a rule engine, wherein the rule configuration information includes current rule parameters; A data acquisition unit, used to acquire user history data associated with the current rule parameters; An optimization unit, configured to perform dynamic optimization processing using the user historical data based on a preset optimization target of the indicator to be optimized, to obtain optimized rule parameters; as well as The application unit is used to apply the optimized rule parameters to the rule engine to obtain new rule configuration information.

9. An electronic device, characterized in that: include: A processor and a memory storing a computer program, wherein the processor is configured to implement the method according to any one of claims 1 to 7 when executing the computer program.

10. A program product comprising a computer program, wherein: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images