Risk prediction model training method, device and equipment and risk prediction method

Abstract

The invention provides a risk prediction model training method and device, equipment, a medium and a program product, which can be applied to the field of artificial intelligence and the field of financial science and technology. The method comprises the steps that feature extraction is conducted on a historical risk prediction report and structured data associated with the historical risk prediction report, m initial features are obtained, the m initial features have a unified data format, and m is a positive integer; the m initial features are evaluated through a gradient lifting model, an evaluation result corresponding to each initial feature is obtained, and the evaluation results represent the importance degree of the initial features in the m initial features; the m initial features are screened according to the evaluation result, n target features are obtained, and n is a positive integer smaller than m; and training a risk prediction model by using the n target features to obtain a trained risk prediction model, the risk prediction model being a pre-trained natural language generation model.

Description

Technical Field

[0001] The present disclosure relates to the fields of artificial intelligence and financial technology, and more specifically to a training method, device, equipment, medium and program product for a risk prediction model, as well as a risk prediction method and device. Background Art

[0002] With the expansion of financial services, it is necessary to conduct risk assessments on users, such as whether the user has money laundering risks, etc., to avoid security risks in business processes.

[0003] Existing anti-money laundering detection technologies usually rely on rule-based systems and manual processes, and often generate compliance reports manually by analyzing suspicious behaviors. The above technologies are prone to generate a large number of false positives, increasing the burden of subsequent audits. That is, based on existing anti-money laundering detection technologies, it is difficult for the system to perform high-precision analysis of user-related information, and there is a problem of low system processing efficiency due to excessive data, and still requiring excessive system resources and human resources, which in turn leads to the problem that the system's processing efficiency and accuracy are difficult to adapt to the rapidly changing money laundering methods. Summary of the invention

[0004] In view of the above problems, the present disclosure provides a risk prediction model training method, device, equipment, medium and program product, as well as a risk prediction method and device.

[0005] According to a first aspect of the present disclosure, a method for training a risk prediction model is provided, comprising: performing feature extraction on historical risk prediction reports and structured data associated with the historical risk prediction reports to obtain m initial features, wherein the m initial features have a unified data format and m is a positive integer; evaluating the m initial features using a gradient boosting model to obtain an evaluation result corresponding to each initial feature, wherein the evaluation result characterizes the importance of the initial feature among the m initial features; screening the m initial features according to the evaluation result to obtain n target features, wherein n is a positive integer less than m; and training a risk prediction model using the n target features to obtain a trained risk prediction model, wherein the risk prediction model is a pre-trained natural language generation model.

[0006] According to an embodiment of the present disclosure, a gradient boosting model is used to evaluate m initial features to obtain an evaluation result corresponding to each initial feature, including: based on the business meaning of the initial features, a contribution evaluation is performed on the m initial features to obtain a contribution value corresponding to each initial feature; the m initial features are reordered to determine the reordered m initial features; the performance of the gradient boosting model is evaluated using the reordered m initial features, and the influence value corresponding to each initial feature is determined according to the performance evaluation result; and the evaluation result corresponding to each initial feature is determined according to the contribution value and the influence value.

[0007] According to an embodiment of the present disclosure, m initial features are screened according to an evaluation result to obtain n target features, including: according to preset conditions, m initial features are screened to obtain a first intermediate feature set, the first intermediate feature set includes at least one initial feature; according to the Pearson correlation coefficient between each initial feature of the first intermediate feature set and a preset feature, the first intermediate feature set is removed to obtain a second intermediate feature set, the second intermediate feature set includes at least one initial feature; each initial feature of the second intermediate feature set is removed for verification, and according to the verification result, the first intermediate feature set is removed to obtain a third intermediate feature set, the third intermediate feature set includes at least one initial feature; the initial features of the third intermediate feature set are determined as target features.

[0008] According to an embodiment of the present disclosure, the above method also includes: optimizing the gradient boosting model to obtain an optimized gradient boosting model, evaluating m initial features based on the optimized gradient boosting model to obtain an evaluation result corresponding to each initial feature.

[0009] According to an embodiment of the present disclosure, a gradient boosting model is optimized to obtain an optimized gradient boosting model, including: constructing a verification feature set; evaluating the gradient boosting model using the verification feature set to obtain an evaluation result; and adjusting the hyperparameters and training rounds of the gradient boosting model according to the evaluation result to obtain an optimized gradient boosting model.

[0010] According to an embodiment of the present disclosure, the above method also includes: adjusting the weights corresponding to the m initial features according to the evaluation results; based on the weights, performing new feature extraction on the historical risk prediction report and the structured data associated with the historical risk prediction report to obtain m new initial features, and using the optimized gradient boosting model to evaluate the m new initial features to obtain evaluation results corresponding to each initial feature.

[0011] According to an embodiment of the present disclosure, the above method also includes: performing data cleaning on the historical risk prediction report and the structured data associated with the historical risk prediction report to obtain the cleaned historical risk prediction report and the structured data associated with the historical risk prediction report, and performing feature extraction on the cleaned historical risk prediction report and the structured data associated with the historical risk prediction report to obtain m initial features.

[0012] A second aspect of the present disclosure provides a risk prediction method, including: obtaining user information; and processing the user information using the above-mentioned risk prediction model to obtain a risk prediction report.

[0013] The third aspect of the present disclosure provides a training device for a risk prediction model, comprising: an extraction module, used to perform feature extraction on historical risk prediction reports and structured data associated with the historical risk prediction reports, to obtain m initial features, wherein the m initial features have a unified data format and m is a positive integer; an evaluation module, used to evaluate the m initial features using a gradient boosting model, to obtain an evaluation result corresponding to each initial feature, wherein the evaluation result characterizes the importance of the initial feature among the m initial features; a screening module, used to screen the m initial features according to the evaluation results, to obtain n target features, where n is a positive integer less than m; and a training module, used to train the risk prediction model using the n target features, to obtain a trained risk prediction model, wherein the risk prediction model is a pre-trained natural language generation model.

[0014] A fourth aspect of the present disclosure provides a risk prediction device, including: an acquisition module for acquiring user information; and a processing module for processing the user information using a risk prediction model to obtain a risk prediction report.

[0015] A fifth aspect of the present disclosure provides an electronic device, comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the above method.

[0016] The sixth aspect of the present disclosure further provides a computer-readable storage medium having a computer program or instructions stored thereon, which implements the steps of the above method when the above computer program or instructions are executed by a processor.

[0017] The seventh aspect of the present disclosure further provides a computer program product, including a computer program or instructions, which implement the steps of the above method when the above computer program or instructions are executed by a processor.

[0018] According to the embodiments of the present disclosure, feature extraction is performed on historical risk prediction reports and structured data associated with historical risk prediction reports to obtain m initial features; the m initial features are evaluated using a gradient boosting model to obtain an evaluation result corresponding to each initial feature; the m initial features are screened according to the evaluation result to obtain n target features; the risk prediction model is trained using the n target features to obtain a trained risk prediction model. By extracting features from the initial data set, the risk prediction model can be easily processed due to the unified data format, and the gradient boosting model is used for evaluation to screen out features with high importance, thereby avoiding the problem of high false alarm rate and low true recognition rate in the trained risk prediction model due to too many useless features, thereby achieving the technical effect of improving the recognition accuracy of the risk prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The above contents and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0020] Figure 1 A diagram schematically illustrates an application scenario of a risk prediction model training method, apparatus, device, medium, and program product according to an embodiment of the present disclosure;

[0021] Figure 2 A flowchart schematically shows a method for training a risk prediction model according to an embodiment of the present disclosure;

[0022] Figure 3 A schematic diagram schematically illustrates a method for training a risk prediction model according to an embodiment of the present disclosure;

[0023] Figure 4 A flow chart of a risk prediction method according to an embodiment of the present disclosure is schematically shown;

[0024] Figure 5 The structure block diagram of the training device of the risk prediction model according to the embodiment of the present disclosure is schematically shown;

[0025] Figure 6 A schematic diagram of a risk prediction device according to an embodiment of the present disclosure is shown; and

[0026] Figure 7 A block diagram of an electronic device suitable for implementing a training method for a risk prediction model according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0027] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0028] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0029] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0030] When using expressions such as "at least one of A, B, and C, etc.", they should generally be interpreted according to the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0031] In the technical solution of the present disclosure, the user information (including but not limited to user personal information, user image information, user device information, such as location information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0032] In the scenario of using personal information for automated decision-making, the methods, devices, and systems provided by the embodiments of the present disclosure provide users with corresponding operation portals for users to choose to agree or reject the automated decision-making results; if the user chooses to reject, the expert decision-making process will be entered. The expression "automated decision-making" here refers to the activity of automatically analyzing and evaluating a person's behavioral habits, interests and hobbies, or economic, health, credit status, etc. through computer programs, and making decisions. The expression "expert decision-making" here refers to the activity of making decisions by people who specialize in a certain field, have specialized experience, knowledge and skills, and have reached a certain level of professionalism.

[0033] An embodiment of the present disclosure provides a method for training a risk prediction model, which performs feature extraction on historical risk prediction reports and structured data associated with the historical risk prediction reports to obtain m initial features, wherein the m initial features have a unified data format and m is a positive integer; the m initial features are evaluated using a gradient boosting model to obtain an evaluation result corresponding to each initial feature, wherein the evaluation result characterizes the importance of the initial feature among the m initial features; the m initial features are screened according to the evaluation results to obtain n target features, wherein n is a positive integer less than m; the risk prediction model is trained using the n target features to obtain a trained risk prediction model, wherein the risk prediction model is a pre-trained natural language generation model.

[0034] Figure 1 Schematic diagram showing the application scenario of the risk prediction model training method, device, equipment, medium and program product according to the embodiment of the present disclosure

[0035] like Figure 1 As shown, the application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 is used to provide a medium for a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired, wireless communication links, or optical fiber cables, etc.

[0036] The user can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 through the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only for example).

[0037] The first terminal device 101, the second terminal device 102, and the third terminal device 103 may be various electronic devices having display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0038] The server 105 may be a server that provides various services, such as a background management server (only as an example) that provides support for websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103. The background management server may analyze and process the received data such as user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal device.

[0039] It should be noted that the training method of the risk prediction model provided in the embodiment of the present disclosure can generally be executed by the server 105. Accordingly, the training device of the risk prediction model provided in the embodiment of the present disclosure can generally be set in the server 105. The training method of the risk prediction model provided in the embodiment of the present disclosure can also be executed by a server or a server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105. Accordingly, the training device of the risk prediction model provided in the embodiment of the present disclosure can also be set in a server or a server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105.

[0040] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0041] The following will be based on Figure 1 The scenario described is used to describe in detail the training method of the risk prediction model of the disclosed embodiment.

[0042] Figure 2 A flowchart of a risk prediction model training method according to an embodiment of the present disclosure is schematically shown.

[0043] like Figure 2 As shown, the training method of the risk prediction model of this embodiment includes operations S210 to S240.

[0044] In operation S210, feature extraction is performed on the historical risk prediction report and structured data associated with the historical risk prediction report to obtain m initial features, wherein the m initial features have a unified data format and m is a positive integer.

[0045] According to an embodiment of the present disclosure, historical risk prediction reports are automatically collected and integrated from multiple data sources, and the historical risk prediction reports also include but are not limited to large cash transactions, frequent cross-border transfers, transactions with high-risk areas, etc. The collected historical risk prediction reports are annotated, including key parts of the historical risk prediction reports, such as customer information, account information, transaction details, fund flows, descriptions of suspicious behaviors, analysis conclusions and recommended measures, etc.

[0046] According to an embodiment of the present disclosure, the report and structured data associated with the historical risk prediction report may be transaction records in a transaction database, user information in a user database, etc., which are used to train a model to understand and generate specific information mentioned in the report.

[0047] According to an embodiment of the present disclosure, feature extraction is performed on historical risk prediction reports and structured data associated with the historical risk prediction reports to extract initial features that are helpful in identifying suspicious transactions, such as transaction amount, transaction frequency, transaction location, IP, complexity of capital flow, basic information related to customers and main economic sources, counterparty information, etc.

[0048] In operation S220, the m initial features are evaluated using a gradient boosting model to obtain an evaluation result corresponding to each initial feature, wherein the evaluation result represents the importance of the initial feature among the m initial features.

[0049] According to an embodiment of the present disclosure, a gradient boosting model is used to evaluate m initial features, evaluate the importance of each initial feature among all initial features, analyze the contribution of each initial feature to the accuracy of the risk prediction model, and identify the most influential features.

[0050] In operation S230, the m initial features are screened according to the evaluation results to obtain n target features, where n is a positive integer less than m.

[0051] According to an embodiment of the present disclosure, according to the importance of each initial feature among all initial features, initial features with higher importance are screened out as target features.

[0052] In operation S240, the risk prediction model is trained using n target features to obtain a trained risk prediction model, wherein the risk prediction model is a pre-trained natural language generation model.

[0053] According to the embodiments of the present disclosure, the pre-trained natural language generation model already has powerful language understanding and generation capabilities and can generate coherent text based on a given context.

[0054] According to an embodiment of the present disclosure, the risk prediction model is fine-tuned using target features, and the key parts of the compliance report are used as training targets. The trained risk prediction model can generate corresponding text paragraphs based on given prompt words or structured data.

[0055] According to an embodiment of the present disclosure, after preliminary training, the generation effect of the risk prediction model is evaluated, and the risk prediction model parameters and training strategies are iteratively optimized to improve the quality and accuracy of the report.

[0056] According to the embodiments of the present disclosure, feature extraction is performed on historical risk prediction reports and structured data associated with historical risk prediction reports to obtain m initial features; the m initial features are evaluated using a gradient boosting model to obtain an evaluation result corresponding to each initial feature; the m initial features are screened according to the evaluation result to obtain n target features; the risk prediction model is trained using the n target features to obtain a trained risk prediction model. By extracting features from the initial data set, the risk prediction model can be easily processed due to the unified data format, and the gradient boosting model is used for evaluation to screen out features with high importance, thereby avoiding the problem of high false alarm rate and low true recognition rate in the trained risk prediction model due to too many useless features, thereby achieving the technical effect of improving the recognition accuracy of the risk prediction model.

[0057] According to an embodiment of the present disclosure, a gradient boosting model is used to evaluate m initial features to obtain an evaluation result corresponding to each initial feature, including: based on the business meaning of the initial features, a contribution evaluation is performed on the m initial features to obtain a contribution value corresponding to each initial feature; the m initial features are reordered to determine the reordered m initial features; the performance of the gradient boosting model is evaluated using the reordered m initial features, and the influence value corresponding to each initial feature is determined according to the performance evaluation result; and the evaluation result corresponding to each initial feature is determined according to the contribution value and the influence value.

[0058] According to an embodiment of the present disclosure, the SHAP value (SHapley AdditiveexPlanations) operation of the gradient boosting model can be specifically to calculate the marginal contribution of the initial features in various initial feature combinations and average them to obtain the SHAP value of the initial features. Based on the business meaning corresponding to each initial feature, the contribution of each initial feature to risk prediction is evaluated and interpreted to obtain the contribution value corresponding to each initial feature.

[0059] According to an embodiment of the present disclosure, the Permutation Importance operation of the gradient boosting model reorders the m initial features. Specifically, the original, unchanged m initial features are used to calculate the performance indicators of the gradient boosting model (such as accuracy, loss, etc.), and then the m initial features are randomly shuffled. The feature value corresponding to each initial feature in the reorganized feature set after the shuffle is changed, and the gradient boosting model is re-evaluated using the re-arranged m initial features to calculate the performance indicators as the performance evaluation results.

[0060] According to an embodiment of the present disclosure, based on the evaluation results, the difference between the original performance and the replaced performance is calculated, and the influence value corresponding to each initial feature is determined, wherein the larger the difference, the more important the initial feature.

[0061] According to an embodiment of the present disclosure, the contribution value and the influence value are taken together as the evaluation result.

[0062] According to an embodiment of the present disclosure, by training a gradient boosting model to evaluate the importance of initial features and analyzing the contribution of each initial feature to the model accuracy, the most influential initial features can be identified to improve the training accuracy of subsequent risk prediction models.

[0063] According to an embodiment of the present disclosure, m initial features are screened according to an evaluation result to obtain n target features, including: according to preset conditions, m initial features are screened to obtain a first intermediate feature set, the first intermediate feature set includes at least one initial feature; according to the Pearson correlation coefficient between each initial feature of the first intermediate feature set and a preset feature, the first intermediate feature set is removed to obtain a second intermediate feature set, the second intermediate feature set includes at least one initial feature; each initial feature of the second intermediate feature set is removed for verification, and according to the verification result, the first intermediate feature set is removed to obtain a third intermediate feature set, the third intermediate feature set includes at least one initial feature; the initial features of the third intermediate feature set are determined as target features.

[0064] According to an embodiment of the present disclosure, m initial features are screened according to preset conditions. For example, initial features whose contribution values ​​and influence values ​​in the evaluation results are both in the top 50% may be selected and used as the first intermediate feature set.

[0065] According to an embodiment of the present disclosure, the Pearson correlation coefficient is a statistic used to measure the degree of linear correlation between two variables. The second intermediate feature set is obtained by removing the initial features with low correlation with the preset features (such as the Pearson correlation coefficient is close to 0) and removing the strongly correlated features to avoid information redundancy between features, wherein the preset features can be set according to actual needs.

[0066] According to an embodiment of the present disclosure, each initial feature of the second intermediate feature set is gradually removed, and the change in the performance of the gradient boosting model is tested to obtain a verification result.

[0067] According to an embodiment of the present disclosure, based on the verification results, if the performance of the gradient boosting model does not significantly decrease after removing certain initial features, these initial features may be low-importance or noise features, which are removed to obtain the target features.

[0068] According to an embodiment of the present disclosure, by screening m initial features to avoid excessive useless features, leading to feature redundancy and other problems, the accuracy and efficiency of the risk prediction model in the training process are improved.

[0069] According to an embodiment of the present disclosure, the above method also includes: optimizing the gradient boosting model to obtain an optimized gradient boosting model, evaluating m initial features based on the optimized gradient boosting model to obtain an evaluation result corresponding to each initial feature.

[0070] According to an embodiment of the present disclosure, a gradient boosting model is optimized to obtain an optimized gradient boosting model, including: constructing a verification feature set; evaluating the gradient boosting model using the verification feature set to obtain an evaluation result; and adjusting the hyperparameters and training rounds of the gradient boosting model according to the evaluation result to obtain an optimized gradient boosting model.

[0071] According to an embodiment of the present disclosure, a validation feature set is used to evaluate a gradient boosting model. Specifically, an independent validation feature set may be used to comprehensively analyze model deficiencies in the following ways, for example, residual analysis between the validation feature set and the initial features; checking the contribution of the features to the model; for classification tasks, analyzing in which categories or specific feature ranges the gradient boosting model performs poorly; and comparing the performance between the validation feature set and the initial features, etc.

[0072] According to an embodiment of the present disclosure, the hyperparameters of the gradient boosting model may be the learning rate (learning_rate), which is used to reduce the learning rate to possibly improve the generalization ability; the depth of the tree (max_depth), which is used to appropriately reduce the depth of the tree to prevent overfitting; L1 / L2 regularization (reg_alpha, reg_lambda), which is used to increase the regularization term to reduce overfitting; and the sample and feature sampling ratio (subsample, colsample_bytree), which is used to control the sampling to increase the robustness of the model.

[0073] According to an embodiment of the present disclosure, overfitting may also be controlled by an early stopping method, and the number of training rounds may be dynamically adjusted according to the performance of the validation set.

[0074] According to an embodiment of the present disclosure, the performance of the improved gradient boosting model is re-evaluated using the same validation feature set and compared with the previous gradient boosting model. If the performance is improved, the improved gradient boosting model is saved.

[0075] According to an embodiment of the present disclosure, the above method also includes: adjusting the weights corresponding to the m initial features according to the evaluation results; based on the weights, performing new feature extraction on the historical risk prediction report and the structured data associated with the historical risk prediction report to obtain m new initial features, and using the optimized gradient boosting model to evaluate the m new initial features to obtain evaluation results corresponding to each initial feature.

[0076] According to an embodiment of the present disclosure, if the gradient boosting model performs poorly on certain categories or initial features in the evaluation results, these categories or initial features are given higher weights. Based on the weights, feature extraction is performed again to obtain initial features that are more suitable for the gradient boosting model.

[0077] According to the embodiments of the present disclosure, in addition to adjusting the weights, interactive features, nonlinear transformations or aggregation features may be added, and low-importance features with large noise may be removed. It is also possible to collect more relevant data based on the evaluation results and expand the number of initial features to improve the generalization ability.

[0078] According to an embodiment of the present disclosure, the above method also includes: performing data cleaning on the historical risk prediction report and the structured data associated with the historical risk prediction report to obtain the cleaned historical risk prediction report and the structured data associated with the historical risk prediction report, and performing feature extraction on the cleaned historical risk prediction report and the structured data associated with the historical risk prediction report to obtain m initial features.

[0079] According to an embodiment of the present disclosure, the collected historical risk prediction reports and structured data associated with the historical risk prediction reports are preprocessed, including text cleaning, data standardization, etc., to process missing values ​​and outliers and unify the data format.

[0080] Figure 3 A schematic diagram schematically illustrates a method for training a risk prediction model according to an embodiment of the present disclosure.

[0081] According to an embodiment of the present disclosure, the training method of the risk prediction model of the embodiment of the present disclosure is as follows: Figure 3 As shown, perform operation S310 to clean the historical risk prediction reports and the structured data associated with the historical risk prediction reports. Perform operation S320 to extract features from the historical risk prediction reports and the structured data associated with the historical risk prediction reports to obtain m initial features. Perform operation S330 to evaluate the m initial features using the gradient boosting model to obtain an evaluation result corresponding to each initial feature. Perform operation S340 to screen the m initial features according to the evaluation results to obtain n target features. Perform operation S350 to evaluate the gradient boosting model using the validation feature set to obtain an evaluation result. Perform operation S360 to adjust the hyperparameters and training rounds of the gradient boosting model to obtain an optimized gradient boosting model, and perform operations S330 and S340 using the optimized gradient boosting model to obtain new target features. Execute S370 to train the risk prediction model using n target features to obtain a trained risk prediction model, wherein during the training process, perform operation S360 according to the iterative training result to train using new target features to obtain a trained risk prediction model.

[0082] In some embodiments, the training method of the risk prediction model of the present disclosure can be, for example, to collect a certain number of historical risk prediction reports, which should cover different types of suspicious activity cases, including but not limited to large cash transactions, frequent cross-border transfers, transactions with high-risk areas, etc. The collected reports are labeled, including key parts of the reports, such as customer information, account information, transaction details, capital flows, descriptions of suspicious behaviors, analysis conclusions and recommended measures, etc. Prepare structured data related to the report, such as transaction records in the transaction database, customer information in the customer database, etc. These data will be used to train the model to understand and generate specific information mentioned in the report. In addition, external knowledge bases in the field of anti-money laundering can also be integrated, including laws and regulations, regulatory guidelines, typical cases, etc., to enrich the background knowledge of the model.

[0083] Subsequently, the risk prediction model is trained based on the specific requirements of the historical risk prediction report (such as the structured data mentioned above and the external knowledge base in the field of anti-money laundering, etc.) and the labeled data prepared above.

[0084] The training process includes preprocessing the historical risk prediction reports and structured data collected above, including text cleaning, data standardization, etc., so that they have a unified data format; then extract key features from the structured data, such as customer information (KYC data), transaction amount, transaction frequency, historical compliance time and penalty records, etc., and embed these features into the training data (including the above-mentioned annotated data) in an appropriate format. The pre-trained language model (i.e., risk prediction model) is fine-tuned using the prepared annotated data. In particular, the key parts of the compliance report are used as training targets, and the training model is able to generate corresponding text paragraphs based on given prompt words or structured data. After preliminary training, the generation effect of the model is evaluated, and the model parameters and training strategies are iteratively optimized to improve the quality and accuracy of the report, so that the trained model can generate specific parts or the entire content of the report by giving specific prompt words or sentences (such as "generate a compliance report on large cash transactions"), and take structured data (such as transaction records of specific customers) as input to guide the model to generate compliance report text reflecting the results of these data analysis.

[0085] In addition, the above steps are integrated into a system to ensure that the data flow from data preprocessing to report generation is seamless, and the model is deployed in the production environment to ensure that the system can process real-time data and generate compliance reports in a timely manner. At the same time, based on feedback and new data, the model performance and report quality are continuously optimized, and the model performance and report feedback are regularly reviewed to optimize the model and report generation process. The collection and feature extraction rules of historical risk prediction reports, model training sets, etc. can also be updated according to new regulatory requirements and market changes.

[0086] Figure 4 The flowchart of the risk prediction method according to the embodiment of the present disclosure is schematically shown.

[0087] like Figure 4 As shown, the risk prediction method of this embodiment includes operations S410 to S420.

[0088] In operation S410, user information is acquired.

[0089] In operation S420, the user information is processed using the risk prediction model to obtain a risk prediction report.

[0090] According to an embodiment of the present disclosure, a specific part or the entire content of a risk prediction report is generated by giving a specific prompt word or sentence (such as "generate a compliance report on large cash transactions").

[0091] According to an embodiment of the present disclosure, structured data (such as transaction records of a specific user) is used as input to guide the model to generate compliance report text that reflects the results of these data analyses.

[0092] According to the embodiments of the present disclosure, through the above-mentioned risk prediction method, it is possible to automatically generate a risk prediction report suitable for compliance anti-money laundering based on given prompt words or structured data, greatly improving the efficiency of compiling compliance reports while maintaining the professionalism and accuracy of the reports.

[0093] Based on the above risk prediction model training method, the present disclosure also provides a risk prediction model training device. Figure 5 The device is described in detail.

[0094] Figure 5 The structural block diagram of the risk prediction model training device according to the embodiment of the present disclosure is schematically shown.

[0095] like Figure 5 As shown, the risk prediction model training device 500 of this embodiment includes an extraction module 510, an evaluation module 520, a screening module 530 and a training module 540.

[0096] The extraction module 510 is used to extract features from the historical risk prediction report and the structured data associated with the historical risk prediction report to obtain m initial features, wherein the m initial features have a unified data format and m is a positive integer. In one embodiment, the extraction module 510 can be used to perform the operation S210 described above, which will not be repeated here.

[0097] The evaluation module 520 is used to evaluate the m initial features using the gradient boosting model to obtain an evaluation result corresponding to each initial feature, wherein the evaluation result represents the importance of the initial feature among the m initial features. In one embodiment, the evaluation module 520 can be used to perform the operation S220 described above, which will not be repeated here.

[0098] The screening module 530 is used to screen the m initial features according to the evaluation results to obtain n target features, where n is a positive integer less than m. In one embodiment, the screening module 530 can be used to perform the operation S230 described above, which will not be described in detail here.

[0099] The training module 540 is used to train the risk prediction model using the n target features to obtain a trained risk prediction model, wherein the risk prediction model is a pre-trained natural language generation model. In one embodiment, the training module 540 can be used to perform the operation S240 described above, which will not be repeated here.

[0100] According to the embodiments of the present disclosure, feature extraction is performed on historical risk prediction reports and structured data associated with historical risk prediction reports to obtain m initial features; the m initial features are evaluated using a gradient boosting model to obtain an evaluation result corresponding to each initial feature; the m initial features are screened according to the evaluation result to obtain n target features; the risk prediction model is trained using the n target features to obtain a trained risk prediction model. By extracting features from the initial data set, the risk prediction model can be easily processed due to the unified data format, and the gradient boosting model is used for evaluation to screen out features with high importance, thereby avoiding the problem of high false alarm rate and low true recognition rate in the trained risk prediction model due to too many useless features, thereby achieving the technical effect of improving the recognition accuracy of the risk prediction model.

[0101] According to an embodiment of the present disclosure, any multiple modules of the extraction module 510, the evaluation module 520, the screening module 530 and the training module 540 can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present disclosure, at least one of the extraction module 510, the evaluation module 520, the screening module 530 and the training module 540 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware and firmware or in any appropriate combination of any of them. Alternatively, at least one of the extraction module 510, the evaluation module 520, the screening module 530 and the training module 540 can be at least partially implemented as a computer program module, and when the computer program module is run, the corresponding function can be performed.

[0102] According to an embodiment of the present disclosure, the evaluation module 520 includes a first evaluation submodule, a first processing submodule, a second evaluation submodule and a first determination submodule.

[0103] The first evaluation submodule is used to perform contribution evaluation on the m initial features based on the business meaning of the initial features to obtain a contribution value corresponding to each initial feature.

[0104] The first processing submodule is used to reorder the m initial features to determine the reordered m initial features.

[0105] The second evaluation submodule is used to perform performance evaluation on the gradient boosting model using the reordered m initial features, and determine the influence value corresponding to each initial feature according to the performance evaluation result.

[0106] The first determination submodule is used to determine the evaluation result corresponding to each initial feature according to the contribution value and the influence value.

[0107] According to an embodiment of the present disclosure, the screening module 530 includes a first screening submodule, a first removing submodule, a second removing submodule, and a second determining submodule.

[0108] The first screening submodule is used to screen the m initial features according to preset conditions to obtain a first intermediate feature set, where the first intermediate feature set includes at least one initial feature.

[0109] The first removal submodule is used to remove the first intermediate feature set according to the Pearson correlation coefficient between each initial feature of the first intermediate feature set and the preset feature to obtain a second intermediate feature set, wherein the second intermediate feature set includes at least one initial feature.

[0110] The second removal submodule is used to perform removal verification on each initial feature of the second intermediate feature set, and according to the verification result, perform removal processing on the first intermediate feature set to obtain a third intermediate feature set, wherein the third intermediate feature set includes at least one initial feature.

[0111] The second determination submodule is used to determine the initial features of the third intermediate feature set as target features.

[0112] According to an embodiment of the present disclosure, the risk prediction model training device 500 also includes an optimization module.

[0113] The optimization module is used to optimize the gradient boosting model to obtain the optimized gradient boosting model, and to evaluate the m initial features based on the optimized gradient boosting model to obtain an evaluation result corresponding to each initial feature.

[0114] According to an embodiment of the present disclosure, the optimization module includes a construction submodule, an evaluation submodule and an adjustment submodule.

[0115] Build submodule for constructing validation feature set.

[0116] The evaluation submodule is used to evaluate the gradient boosting model using the validation feature set to obtain the evaluation results.

[0117] The adjustment submodule is used to adjust the hyperparameters and training rounds of the gradient boosting model according to the evaluation results to obtain the optimized gradient boosting model.

[0118] According to an embodiment of the present disclosure, the risk prediction model training device 500 also includes an adjustment module and an acquisition module.

[0119] The adjustment module is used to adjust the weights corresponding to the m initial features according to the evaluation results.

[0120] A module is obtained, which is used to extract new features from the historical risk prediction report and the structured data associated with the historical risk prediction report based on the weights, and obtain m new initial features, so as to evaluate the m new initial features using the optimized gradient boosting model, and obtain an evaluation result corresponding to each initial feature.

[0121] According to an embodiment of the present disclosure, the risk prediction model training device 500 also includes a cleaning module.

[0122] A cleaning module is used to clean the historical risk prediction report and the structured data associated with the historical risk prediction report to obtain the cleaned historical risk prediction report and the structured data associated with the historical risk prediction report, and to extract features from the cleaned historical risk prediction report and the structured data associated with the historical risk prediction report to obtain m initial features.

[0123] Figure 6 The structural block diagram of the risk prediction device according to an embodiment of the present disclosure is schematically shown.

[0124] like Figure 6 As shown, the risk prediction 600 of this embodiment includes an acquisition module 610 and a processing module 620 .

[0125] The acquisition module 610 is used to acquire user information.

[0126] The processing module 620 is used to process the user information using the risk prediction model to obtain a risk prediction report.

[0127] Figure 7 A block diagram of an electronic device suitable for implementing a training method for a risk prediction model according to an embodiment of the present disclosure is schematically shown.

[0128] like Figure 7As shown, the electronic device 700 according to an embodiment of the present disclosure includes a processor 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage part 708 to a random access memory (RAM) 703. The processor 701 may include, for example, a general-purpose microprocessor (such as a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (for example, an application-specific integrated circuit (ASIC)), etc. The processor 701 may also include an onboard memory for caching purposes. The processor 701 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0129] In RAM 703, various programs and data required for the operation of electronic device 700 are stored. Processor 701, ROM 702 and RAM 703 are connected to each other via bus 704. Processor 701 performs various operations of the method flow according to the embodiment of the present disclosure by executing the program in ROM 702 and / or RAM 703. It should be noted that the program can also be stored in one or more memories other than ROM 702 and RAM 703. Processor 701 can also perform various operations of the method flow according to the embodiment of the present disclosure by executing the program stored in the one or more memories.

[0130] According to an embodiment of the present disclosure, the electronic device 700 may further include an input / output (I / O) interface 705, which is also connected to the bus 704. The electronic device 700 may further include one or more of the following components connected to the input / output (I / O) interface 705: an input portion 706 including a keyboard, a mouse, etc.; an output portion 707 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 708 including a hard disk, etc.; and a communication portion 709 including a network interface card such as a LAN card, a modem, etc. The communication portion 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the input / output (I / O) interface 705 as needed. A removable medium 711, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 710 as needed, so that a computer program read therefrom is installed into the storage portion 708 as needed.

[0131] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present disclosure is implemented.

[0132] According to an embodiment of the present disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, it may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, an apparatus or a device. For example, according to an embodiment of the present disclosure, the computer-readable storage medium may include the ROM 702 and / or RAM 703 described above and / or one or more memories other than ROM 702 and RAM 703.

[0133] The embodiment of the present disclosure also includes a computer program product, which includes a computer program, and the computer program contains program code for executing the method shown in the flowchart. When the computer program product is run in a computer system, the program code is used to enable the computer system to implement the training method of the risk prediction model provided by the embodiment of the present disclosure.

[0134] The above functions defined in the system / device of the embodiment of the present disclosure are performed when the computer program is executed by the processor 701. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0135] In one embodiment, the computer program may rely on tangible storage media such as optical storage devices, magnetic storage devices, etc. In another embodiment, the computer program may also be transmitted and distributed in the form of signals on a network medium, and downloaded and installed through the communication part 709, and / or installed from the removable medium 711. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0136] In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 709, and / or installed from the removable medium 711. When the computer program is executed by the processor 701, the above functions defined in the system of the embodiment of the present disclosure are performed. According to the embodiment of the present disclosure, the system, device, means, module, unit, etc. described above can be implemented by a computer program module.

[0137] According to an embodiment of the present disclosure, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level process and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, Java, C++, python, "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect through the Internet).

[0138] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0139] It will be appreciated by those skilled in the art that the features described in the various embodiments of the present disclosure may be combined and / or combined in a variety of ways, even if such combinations or combinations are not explicitly described in the present disclosure. In particular, without departing from the spirit and teachings of the present disclosure, the features described in the various embodiments of the present disclosure may be combined and / or combined in a variety of ways. All of these combinations and / or combinations fall within the scope of the present disclosure.

[0140] The embodiments of the present disclosure are described above. However, these embodiments are only for illustrative purposes and are not intended to limit the scope of the present disclosure. Although the embodiments are described above, this does not mean that the measures in the various embodiments cannot be used in combination to advantage. Without departing from the scope of the present disclosure, those skilled in the art may make a variety of substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A training method for a risk prediction model, characterized in that: The method comprises: Performing feature extraction on the historical risk prediction report and structured data associated with the historical risk prediction report to obtain m initial features, wherein the m initial features have a unified data format, and m is a positive integer; Using a gradient boosting model to evaluate the m initial features, to obtain an evaluation result corresponding to each of the initial features, wherein the evaluation result represents the importance of the initial feature among the m initial features; The m initial features are screened according to the evaluation results to obtain n target features, where n is a positive integer less than m; The risk prediction model is trained using the n target features to obtain a trained risk prediction model, wherein the risk prediction model is a pre-trained natural language generation model.

2. The method according to claim 1, characterized in that The m initial features are evaluated by using the gradient boosting model to obtain an evaluation result corresponding to each of the initial features, including: Based on the business meaning of the initial features, performing contribution evaluation on the m initial features to obtain a contribution value corresponding to each of the initial features; Reordering the m initial features to determine the reordered m initial features; Using the reordered m initial features to perform performance evaluation on the gradient boosting model, and determining an influence value corresponding to each of the initial features according to the performance evaluation result; An evaluation result corresponding to each of the initial features is determined according to the contribution value and the influence value.

3. The method according to claim 1, characterized in that The m initial features are screened according to the evaluation results to obtain n target features, including: According to a preset condition, the m initial features are screened to obtain a first intermediate feature set, wherein the first intermediate feature set includes at least one of the initial features; According to the Pearson correlation coefficient between each of the initial features of the first intermediate feature set and a preset feature, the first intermediate feature set is removed to obtain a second intermediate feature set, where the second intermediate feature set includes at least one of the initial features; Performing removal verification on each of the initial features of the second intermediate feature set, and performing removal processing on the first intermediate feature set according to the verification result to obtain a third intermediate feature set, wherein the third intermediate feature set includes at least one of the initial features; An initial feature of the third intermediate feature set is determined as the target feature.

4. The method according to claim 1, characterized in that The method further comprises: The gradient boosting model is optimized to obtain an optimized gradient boosting model, and the m initial features are evaluated based on the optimized gradient boosting model to obtain an evaluation result corresponding to each of the initial features.

5. The method according to claim 4, characterized in that The step of optimizing the gradient boosting model to obtain an optimized gradient boosting model includes: Construct validation feature set; Using the validation feature set to evaluate the gradient boosting model, and obtain an evaluation result; According to the evaluation results, the hyperparameters and training rounds of the gradient boosting model are adjusted to obtain the optimized gradient boosting model.

6. The method according to claim 5, characterized in that The method further comprises: According to the evaluation results, adjusting the weights corresponding to the m initial features; Based on the weights, new features are extracted from the historical risk prediction report and the structured data associated with the historical risk prediction report to obtain m new initial features, and the optimized gradient boosting model is used to evaluate the m new initial features to obtain evaluation results corresponding to each of the initial features.

7. The method according to claim 1, characterized in that The method further comprises: Performing data cleaning on the historical risk prediction report and the structured data associated with the historical risk prediction report to obtain the cleaned historical risk prediction report and the structured data associated with the historical risk prediction report, and performing feature extraction on the cleaned historical risk prediction report and the structured data associated with the historical risk prediction report to obtain the m initial features.

8. A risk prediction method, characterized in that: The method comprises: Get user information; The user information is processed using the risk prediction model as described in claims 1-7 to obtain a risk prediction report.

9. A training device for a risk prediction model, characterized in that: The device comprises: An extraction module, used for performing feature extraction on the historical risk prediction report and structured data associated with the historical risk prediction report to obtain m initial features, wherein the m initial features have a unified data format and m is a positive integer; An evaluation module, used for evaluating the m initial features by using a gradient boosting model to obtain an evaluation result corresponding to each of the initial features, wherein the evaluation result represents the importance of the initial feature among the m initial features; a screening module, configured to screen the m initial features according to the evaluation result to obtain n target features, where n is a positive integer less than m; and A training module is used to train the risk prediction model using the n target features to obtain a trained risk prediction model, wherein the risk prediction model is a pre-trained natural language generation model.

10. A risk prediction device, characterized in that: The device comprises: An acquisition module, used to acquire user information; and A processing module is used to process the user information using the risk prediction model as described in claims 1-7 to obtain a risk prediction report.

11. An electronic device, comprising: one or more processors; a memory for storing one or more computer programs, It is characterized in that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 8.

12. A computer-readable storage medium having a computer program or instruction stored thereon, characterized in that: When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

13. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images