Risk analysis model training method, risk data analysis method, device, equipment, medium and product

Abstract

The invention provides a risk analysis model training method and device, a risk data analysis method and device, equipment, a medium and a product, and can be applied to the technical field of artificial intelligence and big data. The training method comprises the steps that target sample risk data are input into a first risk analysis model, a sample risk analysis result is output, and the sample risk data comprise labels corresponding to the sample risk data; training a first risk analysis model by using the sample risk analysis result and a label corresponding to the sample risk analysis result to obtain a second risk analysis model; and updating the second risk analysis model by using a target evaluation model and a preset updating strategy to generate a third risk analysis model, the target evaluation model being used for scoring based on data composed of the sample risk data and the sample risk analysis result.

Description

Technical Field

[0001] The present disclosure relates to the fields of artificial intelligence and big data technology, and in particular to a risk analysis model training method, risk data analysis method, device, equipment, medium and product. Background Art

[0002] With the digital transformation of financial institutions, financial institutions need to use risk prevention and control analysis models to integrate and analyze relevant risk data (such as market risk data, operational risk data, credit risk data, etc.) to obtain analysis results of relevant risk data.

[0003] In the process of conceiving the present disclosure, the inventors discovered that the analysis of risk data in the relevant technology has the following defects: the data quality of the data to be analyzed is poor, which affects the accuracy of the risk analysis results; in traditional risk data analysis, the output risk analysis results are fixedly generated only based on the input data (data to be analyzed), and there is a lack of consideration of actual business needs and different user preferences, resulting in poor risk analysis results, poor user experience, and difficulty in meeting actual business needs. Summary of the invention

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

[0005] According to a first aspect of the present disclosure, a method for training a risk analysis model is provided, comprising: inputting target sample risk data into a first risk analysis model, and outputting a sample risk analysis result, wherein the target sample risk data includes a label corresponding to the target sample risk data;

[0006] Using the sample risk analysis results and the labels corresponding to the sample risk analysis results to train the first risk analysis model to obtain a second risk analysis model; and

[0007] The second risk analysis model is updated using a target assessment model and a preset update strategy to generate a third risk analysis model, wherein the target assessment model is used to score data based on the sample risk data and the sample risk analysis results.

[0008] According to an embodiment of the present disclosure, the method for obtaining the above target evaluation model includes:

[0009] Obtaining the above sample risk data and the above sample risk analysis results; and

[0010] The initial evaluation model is trained based on the above sample risk data, the above sample risk analysis results and the feedback data corresponding to the above sample risk analysis results to obtain the above target evaluation model.

[0011] According to an embodiment of the present disclosure, the above-mentioned feedback data is associated with the user's needs, and includes multiple feedback sub-data, the above-mentioned feedback sub-data includes a data pair consisting of the above-mentioned target sample risk data and the above-mentioned sample risk analysis result, and feedback data corresponding to the above-mentioned data pair, and the above-mentioned feedback data is used to characterize the positive feedback or negative feedback of the above-mentioned sample risk analysis result generated relative to the same sample risk data.

[0012] According to an embodiment of the present disclosure, the above-mentioned preset update strategy includes a proximal strategy optimization algorithm;

[0013] The method of updating the second risk analysis model using the target assessment model and the preset update strategy to generate a third risk analysis model includes:

[0014] The target evaluation model and the proximal strategy optimization algorithm are used to update the parameters of the second risk analysis model to generate a third risk analysis model.

[0015] According to an embodiment of the present disclosure, the cumulative feedback data obtained by the third risk analysis model in the result sequence for realizing the risk analysis result meets the preset conditions, and the cumulative feedback data is obtained based on the feedback data corresponding to each stage in the risk analysis result.

[0016] According to an embodiment of the present disclosure, the above-mentioned proximal strategy optimization algorithm includes a behavior optimization algorithm and a judgment optimization algorithm, and the above-mentioned result sequence includes at least one stage;

[0017] Each stage in the above result sequence includes:

[0018] According to the target sample risk data and the behavior optimization algorithm, generating a risk analysis result corresponding to the target sample risk data; and

[0019] Based on the above target sample risk data and the above risk analysis results, feedback data for the current stage is generated.

[0020] According to an embodiment of the present disclosure, the above method further includes:

[0021] Upon receiving the first sample risk data sent from the target data source, screening the first risk data to obtain second sample risk data; and

[0022] The second sample risk data is processed to obtain target sample risk data, wherein the target sample risk data is represented as sample risk word vectors corresponding to multiple risk types.

[0023] A second aspect of the present disclosure provides a risk data analysis method, comprising:

[0024] The obtained initial risk data is input into a third risk analysis model, and a target risk analysis result is output, wherein the third risk analysis model is trained based on the above method.

[0025] A third aspect of the present disclosure provides a risk analysis model training device, comprising:

[0026] A sample risk analysis result output module, used to input the target sample risk data into the first risk analysis model and output the sample risk analysis result, wherein the target sample risk data includes a label corresponding to the target sample risk data;

[0027] A first risk analysis model training module, configured to train the first risk analysis model using the sample risk analysis results and labels corresponding to the sample risk analysis results to obtain a second risk analysis model; and

[0028] The third risk analysis model generation module is used to update the above-mentioned second risk analysis model using a target assessment model and a preset update strategy to generate a third risk analysis model, wherein the above-mentioned target assessment model is used to score based on data consisting of the above-mentioned sample risk data and the above-mentioned sample risk analysis results.

[0029] A fourth aspect of the present disclosure provides a risk data analysis device, comprising:

[0030] The target risk analysis result output module is used to input the obtained initial risk data into the third risk analysis model and output the target risk analysis result, wherein the third risk analysis model is trained based on the above method.

[0031] A fifth aspect of the present disclosure provides an electronic device, comprising: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the above method.

[0032] The sixth aspect of the present disclosure also provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, causes the processor to execute the above method.

[0033] The seventh aspect of the present disclosure also provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0034] According to the risk analysis model training method, risk data analysis method, device, equipment, medium and product provided by the present invention, by inputting the target sample risk data into the first risk analysis model and outputting the sample risk analysis result, the sample risk analysis result and the label corresponding to the sample risk analysis result can be used to train the first risk analysis model to obtain the second risk analysis model, and then the target assessment model can be used to fine-tune the second risk analysis model using the data pair consisting of the sample risk data scored by the user and the sample risk analysis result, thereby improving the data quality of the training data and the overall performance of the risk analysis model. At the same time, the policy gradient algorithm is improved by pre-setting the update strategy, so that the update of the gradient and step size in the risk analysis model training is more reasonable and scientific. The third risk analysis model generated on this basis takes into account the preferences and needs of different users in the analysis process of the risk data, thereby making the risk analysis results more accurate, improving the user experience, and better meeting the needs of actual business. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] 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:

[0036] Figure 1 A diagram schematically illustrates an application scenario of a risk analysis model training method, risk data analysis method, device, equipment, medium, and product according to an embodiment of the present disclosure;

[0037] Figure 2 A flowchart schematically illustrates a method for training a risk analysis model according to an embodiment of the present disclosure;

[0038] Figure 3 A flowchart schematically shows a method for analyzing risk data according to an embodiment of the present disclosure;

[0039] Figure 4 A system architecture diagram schematically illustrates a method for training a risk analysis model according to an embodiment of the present disclosure;

[0040] Figure 5 The structure block diagram of the training device of the risk analysis model according to the embodiment of the present disclosure is schematically shown;

[0041] Figure 6 A block diagram schematically shows a structure of a risk data analysis device according to an embodiment of the present disclosure; and

[0042] Figure 7 A block diagram of an electronic device suitable for implementing a risk analysis model training method and a risk data analysis method according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0043] 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.

[0044] 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.

[0045] 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.

[0046] 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.).

[0047] 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.

[0048] In the scenario of using personal information for automated decision-making, the methods, devices, and systems provided by the embodiments of the present invention 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 an individual'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.

[0049] The inventor found that in the related art, the risk prevention and control assessment model has the following defects: the data quality of the data to be analyzed is poor, which affects the accuracy of the risk analysis results; in the traditional risk data analysis process, the output risk analysis results are fixedly generated based on the input data (data to be analyzed), lacking consideration of actual business needs and user preferences, resulting in poor risk analysis results, poor user experience, and difficulty in meeting actual business needs. In view of this, the present disclosure inputs the target sample risk data into the first risk analysis model and outputs the sample risk analysis results, so that the sample risk analysis results and the labels corresponding to the sample risk analysis results can be used to train the first risk analysis model to obtain the second risk analysis model, and then the target assessment model can be used to fine-tune the second risk analysis model using the data pairs consisting of the sample risk data and the sample risk analysis results scored by the user, thereby improving the data quality of the training data and the overall performance of the risk analysis model, and at the same time improving the policy gradient algorithm by presetting the update strategy, so that the update of the gradient and step size in the risk analysis model training is more reasonable and scientific, and the third risk analysis model generated on this basis takes into account the preferences and needs of different users in the analysis process of risk data, thereby making the risk analysis results more accurate and better meeting the needs of actual business.

[0050] Embodiments of the present disclosure provide a risk analysis model training method, risk data analysis method, device, equipment, medium and product. The training method includes: inputting target sample risk data into a first risk analysis model and outputting sample risk analysis results, wherein the sample risk data includes a label corresponding to the sample risk data; training the first risk analysis model using the sample risk analysis results and the label corresponding to the sample risk analysis results to obtain a second risk analysis model; and updating the second risk analysis model using a target assessment model and a preset update strategy to generate a third risk analysis model, wherein the target assessment model is used to score based on data consisting of the sample risk data and the sample risk analysis results.

[0051] Figure 1The application scenario diagram of the risk analysis model training method, risk data analysis method, device, equipment, medium and product according to the embodiment of the present disclosure is schematically shown.

[0052] 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.

[0053] The user may use at least one of 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 may 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 as examples).

[0054] 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.

[0055] 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.

[0056] It should be noted that the training method of the risk analysis model and the analysis method of risk data provided in the embodiment of the present disclosure can generally be executed by the server 105. Accordingly, the training device of the risk analysis model and the analysis device of risk data provided in the embodiment of the present disclosure can generally be set in the server 105. The training method of the risk analysis model and the analysis method of risk data provided in the embodiment of the present disclosure can also be executed by a server or 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 analysis model and the analysis device of risk data provided in the embodiment of the present disclosure can also be set in a server or 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.

[0057] 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.

[0058] Figure 2 The flowchart of the risk analysis model training method according to the embodiment of the present disclosure is schematically shown.

[0059] like Figure 2 As shown, the risk analysis model training method of this embodiment includes operations S210 to S230.

[0060] In operation S210 , target sample risk data is input into a first risk analysis model, and a sample risk analysis result is output, wherein the target sample risk data includes a label corresponding to the target sample risk data.

[0061] According to an embodiment of the present disclosure, the target sample risk data may be data related to the risk type of a financial institution, and a method for acquiring the target sample risk data may include: obtaining the second sample risk data by performing data screening and filtering on the first sample risk data, and performing word segmentation processing, word segmentation cleaning, and feature word extraction on the second sample risk data using relevant methods to obtain the target sample risk data, i.e., a risk feature vector.

[0062] According to an embodiment of the present disclosure, a clustering algorithm can also be used to cluster the target sample risk data to obtain multiple types of target sample risk data. The types of target sample risk data include: operational risk type data, market risk type data, compliance risk type data, and credit risk type data, and the target sample risk data corresponding to each type can include multiple types of sub-data. The label corresponding to the target sample risk data can correspond to the type of the target sample risk data.

[0063] In operation S220, the first risk analysis model is trained using the sample risk analysis results and labels corresponding to the sample risk analysis results to obtain a second risk analysis model.

[0064] According to an embodiment of the present disclosure, after obtaining the sample risk analysis results, the sample risk analysis results and the corresponding labels can be used to train the first risk analysis model to obtain a trained model (i.e., the second risk analysis model), and the labels corresponding to the sample risk analysis results can correspond to the types of the sample risk analysis results.

[0065] In operation S230, the second risk analysis model is updated using a target assessment model and a preset update strategy to generate a third risk analysis model, wherein the target assessment model is used to score based on a data pair consisting of sample risk data and sample risk analysis results.

[0066] According to the embodiments of the present disclosure, inputting risk data into the second risk analysis model can output corresponding sample risk analysis results, that is, different types of risk directions. However, in the actual business process, the sample risk analysis results output by the second risk analysis model are inconsistent with the goals and intentions of the business personnel. It is necessary to set up a target evaluation model to score the sample risk analysis results and then generate feedback data (reward signal). The reward is designed based on user feedback data, and the reward signal evaluates / scores the generated sample risk analysis results from the perspective of the business personnel.

[0067] According to an embodiment of the present disclosure, the target evaluation model may be a reward model (RM), and business personnel may rank the generated sample risk analysis results from best to worst according to actual conditions and business personnel preferences, and the reward signal may include positive signals and negative signals.

[0068] According to an embodiment of the present disclosure, the preset update strategy can be to update the second risk analysis model through a reinforcement learning algorithm. Specifically in the present disclosure, the proximal policy optimization algorithm (PPO) in the reinforcement learning algorithm can be used to introduce a reward signal to adjust the behavior of the second risk analysis model, so that the sample risk analysis results generated by the model are more in line with the preferences of business personnel or the needs of actual business. That is, the reinforcement learning algorithm can be used to adjust the strategy of the second risk analysis model by maximizing the expected reward, so that the model is more inclined to choose behaviors that can get higher rewards when selecting behaviors, and the reinforcement learning algorithm is used to adjust the objective function gradient and step size update strategy.

[0069] In a feasible embodiment, in order to prevent the output strategy from changing too drastically, a regularization term may be introduced into the reinforcement learning algorithm to limit the range of changes in the strategy, for example, by limiting it through a specific proximal norm.

[0070] According to an embodiment of the present disclosure, by inputting target sample risk data into a first risk analysis model and outputting sample risk analysis results, the sample risk analysis results and labels corresponding to the sample risk analysis results can be used to train the first risk analysis model to obtain a second risk analysis model. Then, the target assessment model can be used to fine-tune the second risk analysis model using data pairs consisting of sample risk data scored by users and sample risk analysis results, thereby improving the data quality of the training data and the overall performance of the risk analysis model. At the same time, the policy gradient algorithm is improved by presetting an update strategy, so that the update of gradients and step sizes in risk analysis model training is more reasonable and scientific. The third risk analysis model generated on this basis takes into account the preferences and needs of different users in the analysis process of risk data, thereby making the risk analysis results more accurate and better meeting the needs of actual business.

[0071] According to an embodiment of the present disclosure, a method for acquiring a target assessment model includes: acquiring sample risk data and sample risk analysis results; and training an initial assessment model based on the sample risk data, the sample risk analysis results, and feedback data corresponding to the sample risk analysis results to obtain a target assessment model.

[0072] According to an embodiment of the present disclosure, the initial evaluation model may be an untrained evaluation model, and the target evaluation model may be a pre-trained reward model. The target evaluation model is obtained by training the initial evaluation model with the acquired sample risk data, the sample risk analysis results corresponding to the sample risk data output by the second risk analysis model, and the scores of the sample risk analysis results by business personnel based on actual conditions or preferences.

[0073] According to an embodiment of the present disclosure, feedback data is associated with user needs, and includes multiple feedback sub-data, the feedback sub-data including a data pair consisting of target sample risk data and a sample risk analysis result, and feedback data corresponding to the data pair, the feedback data is used to characterize positive feedback or negative feedback of the sample risk analysis result generated relative to the same sample risk data.

[0074] According to an embodiment of the present disclosure, the feedback data may be composed of a plurality of feedback sub-data, including a data pair consisting of target sample risk data and a sample risk analysis result, and feedback data corresponding to the data pair. For example, the feedback data D1 includes the target sample risk data k, the sample risk analysis result k', the data pair consisting of the target sample risk data and the sample risk analysis result (k, k'), and the feedback data (reward signal) m corresponding to the data pair (k, k').

[0075] In a feasible embodiment, the reward model can be used to score the sample risk analysis results, and the scoring range is an integer between 0 and 10, where 0 represents the lowest reward and 10 represents the highest reward. For example, based on the input text a (target sample risk data), there are three sample risk analysis results X, Y and Z. The analysis results are ranked according to the feedback data of the business personnel: Z>Y>X, that is, for the input text a, based on the comprehensive cognition of the business personnel, the sample risk analysis result Z is better than the sample risk analysis result Y.

[0076] According to an embodiment of the present disclosure, by having users score sample risk analysis results, the second risk analysis model can better understand the quality of the generated sample risk analysis results and improve the performance of the model in subsequent model adjustments.

[0077] According to an embodiment of the present disclosure, the preset update strategy includes a proximal strategy optimization algorithm; wherein, the second risk analysis model is updated using the target evaluation model and the preset update strategy to generate a third risk analysis model, including: using the target evaluation model and the proximal strategy optimization algorithm to update the parameters of the second risk analysis model to generate a third risk analysis model.

[0078] According to an embodiment of the present disclosure, the preset update strategy can be a reinforcement learning algorithm, the core of which is the intelligence obtained by the subject in the interaction with the environment, and is based on the environment, subject, behavior and reward. Specifically in the present disclosure, the subject can be the second risk analysis model, the environment can be the input text (target sample risk data) corresponding to the second risk analysis model and the generated sample risk analysis results, the behavior can be the noise data output by the second risk analysis model, and the reward can be designed according to the preferences of business personnel or actual business needs. In this process, the step of denoising the second risk analysis model to generate sample risk analysis results is used as the trajectory of reinforcement learning, and the reward signal is used to control the entire generation process, so that the second risk analysis model is optimized in the direction of high cumulative rewards.

[0079] There are many different environments in the reinforcement learning algorithm, and each environment corresponds to many different feasible solutions. Therefore, each time the environment → behavior → reward cycle is performed, the model parameters are updated, which makes it difficult for the model to converge, resulting in the model being able to cope with the "current situation" but not with a larger number of environments. Therefore, the proximal policy optimization algorithm does not achieve the optimality of a single or a few behaviors, but takes the optimality of the sequence as the ultimate goal.

[0080] According to an embodiment of the present disclosure, a proximal policy optimization algorithm (PPO) can use a clipping function to limit the difference between the new and old strategies during the optimization process. The clipping function is used to calculate the ratio of the new and old strategies on each sample and compare it with a preset range. The clipping function uses a clipping ratio (a numerical value between 0 and 1). For example, if the clipping ratio is set to 0.1, then when calculating the ratio of the new and old strategies, the ratio can be limited to between 0.9 and 1.1.

[0081] According to the embodiments of the present disclosure, by presetting the update strategy and using the cutting function to ensure that the strategy is within a predetermined range when updating, excessive strategy changes and strategy instability are avoided, thereby ensuring the convergence of the training model. At the same time, by comparing the proportions of the new and old strategies on each target sample risk data and selecting the smaller one, the part of the old strategy that has performed well can be retained, thereby improving the stability and performance of the strategy.

[0082] According to an embodiment of the present disclosure, the cumulative feedback data obtained by the third risk analysis model in the result sequence for realizing the risk analysis result meets the preset conditions, and the cumulative feedback data is obtained according to the feedback data corresponding to each stage in the risk analysis result.

[0083] According to an embodiment of the present disclosure, the result sequence may be composed of the risk analysis results of each stage. At the same time, the risk analysis results of each stage correspond to the feedback data of each stage to form cumulative feedback data. The cumulative rewards obtained based on the user feedback data in the result sequence may include a total score and a depreciation value. The total score is obtained by the pre-trained reward model according to the input data and output results of the result sequence. The depreciation value is the product of the reward of the last stage in the result sequence and the depreciation coefficient. According to actual business needs, the cumulative feedback data should meet the preset conditions, that is, the cumulative feedback data should be greater than the corresponding preset threshold. The preset threshold can be determined according to the actual situation, and is not specifically limited here.

[0084] According to an embodiment of the present disclosure, the proximal strategy optimization algorithm includes a behavior optimization algorithm and a judgment optimization algorithm, and the result sequence includes at least one stage; wherein each stage in the result sequence includes: generating a risk analysis result corresponding to the target sample risk data based on the target sample risk data and the behavior optimization algorithm; and generating feedback data for the current stage based on the target sample risk data and the risk analysis result.

[0085] According to an embodiment of the present disclosure, the behavior optimization algorithm in the proximal policy optimization algorithm can be the risk analysis result of the corresponding noisy data generated by the second risk analysis model according to the target sample risk data; the judgment optimization algorithm can be the reward signal of the current stage output according to the risk analysis result of the target sample risk data and the noisy data. For example, two sample risk analysis results a1 and a2 can be generated based on the same target sample risk data A, a1 comes from the second risk analysis model, and a2 comes from the current iterative risk analysis model after the proximal policy optimization algorithm. The risk analysis results generated by the two models can be compared to calculate the difference reward signal. It can be understood that the reward signal can be expressed as a positive value or a negative value.

[0086] According to an embodiment of the present disclosure, the method also includes: upon receiving first sample risk data sent from a target data source, screening the first risk data to obtain second sample risk data; and processing the second sample risk data to obtain target sample risk data, wherein the target sample risk data is represented as sample risk word vectors corresponding to multiple risk types.

[0087] According to an embodiment of the present disclosure, the first sample risk data may be risk data of different fields or types obtained from a risk database in a user business system, wherein the risk database may be a risk data set defined by the user business system, or may be a storage space connected to the user business system for storing risk data generated by the institution to be analyzed, such as cloud storage, mobile hard disk, etc., and no specific restrictions are made here. At the same time, the risk data type, data size, and data format in the risk database are not restricted here. The second sample risk data may be sample risk data obtained by filtering the first sample risk data using preset filtering rules, and the filtering rules include filtering high-frequency and inefficient words.

[0088] According to an embodiment of the present disclosure, natural language text analysis technology, such as Neuro-Linguistic Programming (NLP), can be used to perform semantic analysis on the second sample risk data and output an initial word segmentation list; for the initial word segmentation, high-frequency and inefficient vocabulary filtering and consistency processing can be performed, and similar word segmentations can be replaced with standard word segmentations. For example, high-frequency and inefficient words such as "financial institutions" are filtered out, and "users", "clients" and "customers" are all replaced with "users", and a cleaned and filtered word segmentation list is output; all sentences that have been segmented and filtered are input, and the TF-IDF (term frequency-inverse document frequency) method is used to output a vectorized representation of the text (second sample risk data).

[0089] According to an embodiment of the present disclosure, the TF-IDF method is used to output the vectorized representation of the second sample risk data. The specific operation of the term frequency (TF) calculation method is shown in formula (1):

[0090]

[0091] Among them, tf i,j is the frequency of the word in the text, n i，j is the frequency of the word appearing in the above text, and the denominator is the sum of the frequencies of all words appearing in the above text.

[0092] Inverse document frequency (IDF) is a measure of the general importance of a word. The specific operation of the calculation method is shown in formula (2):

[0093]

[0094] The numerator is the total number of texts, and the denominator is the number of texts containing the word. The larger the calculation result is, the more distinguishing the word is. Multiply the results of the above formulas (1) and (2). If the result is higher than the threshold, it corresponds to an important word. It should be noted that the threshold can be determined according to the actual situation and is not limited here.

[0095] In a feasible embodiment, high-frequency and low-efficiency words commonly used in actual business, such as "according to", "carry out", "work", etc., can be accumulated and input to form a high-frequency and low-efficiency word library. The words in the word library are used to filter and select the first sample risk data to obtain the second sample risk data.

[0096] In a feasible embodiment, after the second sample risk data is screened and filtered, an unsupervised learning method, such as a k-means clustering algorithm, may be used to cluster the data to obtain multiple types of target sample risk data.

[0097] Figure 3 The flowchart of the risk data analysis method according to the embodiment of the present disclosure is schematically shown.

[0098] like Figure 3 As shown, the risk analysis model training method of this embodiment includes operation S310.

[0099] In operation S310, the obtained initial risk data is input into a third risk analysis model, and a target risk analysis result is output, wherein the third risk analysis model is trained based on the above method.

[0100] According to the embodiments of the present disclosure, the data to be input (initial risk data) can be input into the trained third risk analysis model to obtain the target risk analysis result, that is, the direction of intelligent risk control. Specifically in the present disclosure, Table 1 is used as an example for explanation. Table 1 is the relevant content of multiple target risk analysis results corresponding to various types of initial risk data according to the embodiments of the present disclosure.

[0101] Table 1

[0102]

[0103]

[0104] Figure 4 The system architecture diagram of the risk analysis model training method according to the embodiment of the present disclosure is schematically shown.

[0105] like Figure 4As shown, the first risk analysis model 410 is trained with the target sample risk data to obtain the sample risk analysis result 420 (i.e., multiple types of intelligent directions), and the sample risk analysis result 420 and the corresponding label are used to train the first risk analysis model 410, thereby generating a second risk analysis model 430, and then the second risk analysis model 430 is updated in combination with the target evaluation model (i.e., reward model) 440 and the preset update strategy 450 to generate a third risk analysis model 460, and the third risk analysis model 460 needs to obtain the cumulative feedback data obtained in the result sequence for realizing the risk analysis result. If the preset conditions are met, and the cumulative feedback data of the risk analysis results obtained by using the third risk analysis model 460 meets the preset conditions, the model can be used to perform risk analysis on the initial risk data, that is, model application 470; and if the cumulative feedback data of the risk analysis results obtained by using the third risk analysis model 460 does not meet the preset conditions, the target evaluation model (that is, the reward model) 440 and the preset update strategy 450 can be used to update the second risk analysis model 430. It can be understood that the number of updates can be determined based on actual conditions, and is not specifically limited here.

[0106] Based on the above risk analysis model training method, the present disclosure also provides a risk analysis model training device. Figure 5 The device is described in detail.

[0107] Figure 5 The structural block diagram of the risk analysis model training device according to the embodiment of the present disclosure is schematically shown.

[0108] like Figure 5 As shown, the risk analysis model training device 500 of this embodiment includes a sample risk analysis result output module 510 , a first risk analysis model training module 520 and a third risk analysis model generating module 530 .

[0109] The sample risk analysis result output module 510 is used to input the target sample risk data into the first risk analysis model and output the sample risk analysis result, wherein the sample risk data includes a label corresponding to the sample risk data. In one embodiment, the sample risk analysis result output module 510 can be used to perform the operation S210 described above, which will not be repeated here.

[0110] The first risk analysis model training module 520 is used to train the first risk analysis model using the sample risk analysis results and the labels corresponding to the sample risk analysis results to obtain the second risk analysis model. In one embodiment, the first risk analysis model training module 520 can be used to perform the operation S220 described above, which will not be repeated here.

[0111] The third risk analysis model generation module 530 is used to update the second risk analysis model using the target assessment model and the preset update strategy to generate a third risk analysis model, wherein the target assessment model is used to score based on the data consisting of the sample risk data and the sample risk analysis results. In one embodiment, the third risk analysis model generation module 530 can be used to perform the operation S230 described above, which will not be repeated here.

[0112] According to an embodiment of the present disclosure, through the sample risk analysis result output module 510, the first risk analysis model training module 520 and the third risk analysis model generation module 530 in the risk analysis model training device 500, by inputting the target sample risk data into the first risk analysis model, the sample risk analysis result is output, so that the first risk analysis model can be trained using the sample risk analysis result and the label corresponding to the sample risk analysis result to obtain the second risk analysis model, and then the second risk analysis model can be fine-tuned and trained through the target evaluation model using the data pair consisting of the sample risk data scored by the user and the sample risk analysis result, thereby improving the data quality of the training data and the overall performance of the risk analysis model, and at the same time, the policy gradient algorithm is improved by presetting the update strategy, so that the update of the gradient and step size in the risk analysis model training is more reasonable and scientific, and the third risk analysis model generated on this basis takes into account the preferences and needs of different users in the analysis process of the risk data, thereby making the risk analysis results more accurate, improving the user experience, and better meeting the needs of actual business.

[0113] According to an embodiment of the present disclosure, a method for acquiring a target assessment model includes: acquiring sample risk data and sample risk analysis results; and training an initial assessment model based on the sample risk data, the sample risk analysis results, and feedback data corresponding to the sample risk analysis results to obtain a target assessment model.

[0114] According to an embodiment of the present disclosure, feedback data is associated with user needs, and includes multiple feedback sub-data, the feedback sub-data including a data pair consisting of target sample risk data and a sample risk analysis result, and feedback data corresponding to the data pair, the feedback data is used to characterize positive feedback or negative feedback of the sample risk analysis result generated relative to the same sample risk data.

[0115] According to an embodiment of the present disclosure, the third risk analysis model generation module includes: a third risk analysis model generation submodule, which is used to update the parameters of the second risk analysis model using the target assessment model and the proximal strategy optimization algorithm to generate a third risk analysis model.

[0116] According to an embodiment of the present disclosure, the cumulative feedback data obtained by the third risk analysis model in the result sequence for realizing the risk analysis result meets the preset conditions, and the cumulative feedback data is obtained according to the feedback data corresponding to each stage in the risk analysis result.

[0117] According to an embodiment of the present disclosure, the proximal strategy optimization algorithm includes a behavior optimization algorithm and a judgment optimization algorithm, and the result sequence includes at least one stage; wherein each stage in the result sequence includes: generating a risk analysis result corresponding to the target sample risk data based on the target sample risk data and the behavior optimization algorithm; and generating feedback data for the current stage based on the target sample risk data and the risk analysis result.

[0118] According to an embodiment of the present disclosure, the device further includes: a first risk data screening module and a second sample risk data processing module.

[0119] The first risk data screening module is used to screen the first risk data to obtain second sample risk data when receiving the first sample risk data sent from the target data source.

[0120] The second sample risk data processing module is used to process the second sample risk data to obtain target sample risk data, wherein the target sample risk data is represented as sample risk word vectors corresponding to multiple risk types.

[0121] Based on the above risk data analysis method, the present disclosure also provides a risk data analysis device. Figure 6 The device is described in detail.

[0122] Figure 6 The structure block diagram of the risk data analysis device according to the embodiment of the present disclosure is schematically shown.

[0123] like Figure 6 As shown, the risk data analysis device 600 of this embodiment includes a target risk analysis result output module 610. It is used to input the obtained initial risk data into the third risk analysis model and output the target risk analysis result, wherein the third risk analysis model is trained based on the above method. In one embodiment, the target risk analysis result output module 610 can be used to perform the operation S310 described above, which will not be repeated here.

[0124] According to an embodiment of the present disclosure, any multiple modules of the sample risk analysis result output module 510, the first risk analysis model training module 520 and the third risk analysis model generation module 530 can be combined in 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 sample risk analysis result output module 510, the first risk analysis model training module 520 and the third risk analysis model generation module 530 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 an appropriate combination of any of them. Alternatively, at least one of the sample risk analysis result output module 510, the first risk analysis model training module 520 and the third risk analysis model generation module 530 may be at least partially implemented as a computer program module, which may perform corresponding functions when executed.

[0125] Figure 7 A block diagram of an electronic device suitable for implementing a risk analysis model training method and a risk data analysis method according to an embodiment of the present disclosure is schematically shown.

[0126] like Figure 7 As 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 into a random access memory (RAM) 703. The processor 701 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., 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.

[0127] 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.

[0128] 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.

[0129] 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.

[0130] According to an embodiment of the present disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, 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 of the above. 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, a 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.

[0131] The embodiment of the present disclosure also includes a computer program product, which includes a computer program, and the computer program includes a 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 risk analysis model training method and risk data analysis method provided in the embodiment of the present disclosure.

[0132] 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.

[0133] 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.

[0134] 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.

[0135] 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).

[0136] 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.

[0137] It will be appreciated by those skilled in the art that the features described in the various embodiments and / or claims of the present disclosure may be combined and / or combined in a variety of ways, even if such combinations and / or combinations are not explicitly described in the present disclosure. In particular, the features described in the various embodiments and / or claims of the present disclosure may be combined and / or combined in a variety of ways without departing from the spirit and teachings of the present disclosure. All of these combinations and / or combinations fall within the scope of the present disclosure.

[0138] 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 separately, this does not mean that the measures in the various embodiments cannot be used in combination to advantage. The scope of the present disclosure is defined by the attached claims and their equivalents. 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 method for training a risk analysis model, characterized in that: The method comprises: Inputting the target sample risk data into the first risk analysis model, and outputting the sample risk analysis result, wherein the target sample risk data includes a label corresponding to the target sample risk data; Using the sample risk analysis result and the label corresponding to the sample risk analysis result to train the first risk analysis model to obtain a second risk analysis model; and The second risk analysis model is updated using a target assessment model and a preset update strategy to generate a third risk analysis model, wherein the target assessment model is used to score based on data consisting of the sample risk data and the sample risk analysis results.

2. The method according to claim 1, characterized in that in, The target evaluation model is obtained by: Acquiring the sample risk data and the sample risk analysis result; and An initial evaluation model is trained according to the sample risk data, the sample risk analysis result and feedback data corresponding to the sample risk analysis result to obtain the target evaluation model.

3. The method according to claim 2, characterized in that in, The feedback data is associated with the user's needs, and includes multiple feedback sub-data, wherein the feedback sub-data includes a data pair consisting of the target sample risk data and the sample risk analysis result, and feedback data corresponding to the data pair, and the feedback data is used to characterize positive feedback or negative feedback of the sample risk analysis result generated relative to the same sample risk data.

4. The method according to claim 1, characterized in that: in, The preset update strategy includes a proximal strategy optimization algorithm; The method of updating the second risk analysis model by using the target assessment model and the preset update strategy to generate a third risk analysis model includes: The target assessment model and the proximal strategy optimization algorithm are used to update the parameters of the second risk analysis model to generate a third risk analysis model.

5. The method according to any one of claims 1 to 4, characterized in that: in, The accumulated feedback data obtained by the third risk analysis model in the result sequence for realizing the risk analysis result meets the preset condition, and the accumulated feedback data is obtained according to the feedback data corresponding to each stage in the risk analysis result.

6. The method according to claim 5, characterized in that in, The proximal strategy optimization algorithm includes a behavior optimization algorithm and a judgment optimization algorithm, and the result sequence includes at least one stage; Wherein, each stage in the result sequence includes: generating a risk analysis result corresponding to the target sample risk data according to the target sample risk data and the behavior optimization algorithm; and Feedback data for the current stage is generated based on the target sample risk data and the risk analysis results.

7. The method according to claim 1, characterized in that The method further comprises: Upon receiving the first sample risk data sent from the target data source, screening the first risk data to obtain second sample risk data; and The second sample risk data is processed to obtain target sample risk data, wherein the target sample risk data is represented as sample risk word vectors corresponding to multiple risk types.

8. A method for analyzing risk data, characterized in that: The method comprises: The obtained initial risk data is input into a third risk analysis model, and a target risk analysis result is output, wherein the third risk analysis model is trained based on the method according to any one of claims 1 to 7.

9. A training device for a risk analysis model, characterized in that: The device comprises: A sample risk analysis result output module, used to input the target sample risk data into the first risk analysis model and output the sample risk analysis result, wherein the target sample risk data includes a label corresponding to the target sample risk data; A first risk analysis model training module, configured to train the first risk analysis model using the sample risk analysis result and a label corresponding to the sample risk analysis result to obtain a second risk analysis model; and The third risk analysis model generation module is used to update the second risk analysis model using a target assessment model and a preset update strategy to generate a third risk analysis model, wherein the target assessment model is used to score based on data consisting of the sample risk data and the sample risk analysis results.

10. A risk data analysis device, characterized in that: The device comprises: A target risk analysis result output module is used to input the obtained initial risk data into a third risk analysis model and output a target risk analysis result, wherein the third risk analysis model is trained based on the method described in any one of claims 1 to 7.

11. An electronic device, comprising: one or more processors; a storage device for storing one or more computer programs, The method 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 stored thereon, characterized in that: When the computer program 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, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images