Customer risk report generation method and device, electronic equipment, medium and product

Abstract

The invention provides a customer risk report generation method and device, electronic equipment, a medium and a product, and the method comprises the steps: obtaining a plurality of pieces of feature information of a customer, and determining a prediction risk level of the customer according to the plurality of pieces of feature information; according to the multiple pieces of feature information and the predicted risk level, determining the contribution degree of each piece of feature information to the predicted risk level; for any risk feature information, determining risk description corresponding to the risk feature information based on a natural language model according to the contribution degree corresponding to the risk feature information; wherein the risk feature information is screened from multiple pieces of feature information; and generating the first customer risk report according to the risk description corresponding to each piece of risk feature information, so that a clear risk profile and detailed risk description can be provided for the customer, and the efficiency of writing the customer risk report and the accuracy of the confirmed customer risk point can be improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method, device, electronic device, medium and product for generating a customer risk report. Background Art

[0002] Risk rating is a key component of a bank's risk management system, helping banks to effectively manage risks, optimize resource allocation, and provide better customer service in a complex financial environment.

[0003] At present, in the customer risk rating system, the system calculates the customer's quantitative score and qualitative score, and gives the customer an appropriate system score and system level. The score source and system assessment level provided by the rating system are the main reference for business personnel to write customer risk reports. In the process of writing reports, business personnel also need to read information from multiple sources, such as additional lists provided by other business lines, or analysis and summary of the customer's transactions in the past year, etc., resulting in low writing efficiency. Moreover, under the premise that the risk level determined by the system is low risk or medium risk, the final risk level assessment and risk point determination are heavily dependent on the risk capture acumen of the business personnel. Business personnel with low acumen may not be able to fully identify the customer's potential risks, resulting in inaccurate risk ratings and risk points. Summary of the invention

[0004] The present application provides a method, device, electronic device, medium and product for generating a customer risk report, so as to improve the efficiency of writing a customer risk report.

[0005] In a first aspect, an embodiment of the present application provides a method for generating a customer risk report, comprising:

[0006] Acquire multiple characteristic information of the customer, and determine the predicted risk level of the customer based on the multiple characteristic information;

[0007] Determining, based on the plurality of feature information and the predicted risk level, a contribution of each feature information to the predicted risk level;

[0008] For any risk characteristic information, according to the contribution degree corresponding to the risk characteristic information, based on the natural language model, determine the risk description corresponding to the risk characteristic information; wherein the risk characteristic information is selected from the multiple characteristic information;

[0009] A first customer risk report is generated according to the risk descriptions corresponding to each risk characteristic information.

[0010] Optionally, the plurality of feature information includes: a plurality of current feature information, a plurality of historical feature information, and a historical predicted risk level; wherein the plurality of historical feature information is used to indicate the feature information used when the predicted risk level of the customer was determined last time, and the historical risk level is used to indicate the predicted risk level determined last time;

[0011] The categories corresponding to the multiple current feature information include: basic information category, extended information category, qualitative information category, quantitative information category, transaction information category, management and control information category, and large amount information category.

[0012] Optionally, determining the predicted risk level of the customer according to the plurality of feature information includes:

[0013] Determine the predicted risk level of the customer based on the trained risk level prediction model according to the plurality of feature information;

[0014] The method further comprises:

[0015] Constructing a training sample set, wherein each training sample in the training sample set includes: a plurality of feature information, and a label corresponding to the training sample is: an actual risk level of the customer;

[0016] Inputting the training sample set into the initial risk level prediction model to obtain the predicted risk level corresponding to each training sample;

[0017] For each training sample, according to the actual risk level and the predicted risk level respectively corresponding to the training sample, the loss value is determined by the following loss function;

[0018]

[0019] Among them, the actual risk level Y = [y1, y2, y3, y4], and the predicted risk level is is the loss value, λ is a constant;

[0020] According to the loss values ​​corresponding to each training sample, the parameters of the initial risk level prediction model are adjusted to obtain a trained risk level prediction model.

[0021] Optionally, the actual risk level of the customer includes: low risk, medium risk, relatively high risk and high risk, and each feature information includes: a feature field and a feature value corresponding to the feature field;

[0022] The method further comprises:

[0023] Obtain multiple secondary customer risk reports written by business personnel for higher risk and high risk customers;

[0024] Merging the contents of multiple second customer risk reports to obtain a target customer risk report;

[0025] Arrange the words in the target customer risk report from high to low according to the frequency of occurrence to obtain a word frequency sequence;

[0026] At least one feature field is determined according to the vocabulary frequency sequence.

[0027] Optionally, each feature information includes: a feature field and a feature value corresponding to the feature field;

[0028] The method further comprises:

[0029] For any characteristic information, if the contribution degree corresponding to the characteristic information is a positive value, the characteristic information is determined to be risk characteristic information;

[0030] Accordingly, for any risk characteristic information, according to the contribution degree corresponding to the risk characteristic information, determining the risk description corresponding to the risk characteristic information includes:

[0031] For any risk characteristic information, if the characteristic value of the risk characteristic information is a numerical value, the description level corresponding to the characteristic information is determined according to the contribution degree corresponding to the risk characteristic information, and the risk description corresponding to the risk characteristic information is determined based on the first mapping relationship table according to the characteristic field and the description level corresponding to the risk characteristic information, wherein the first mapping relationship table is used to indicate the mapping relationship between the risk description and the characteristic field and the description level respectively; the description level includes: low, medium, and high;

[0032] If the characteristic value of the risk characteristic information is not a numerical value, the characteristic field and characteristic value corresponding to the risk characteristic information are used to determine the risk description corresponding to the risk characteristic information based on a second mapping relationship table, wherein the second mapping relationship table is used to indicate the mapping relationship between the risk characteristic and the characteristic field and characteristic value respectively.

[0033] Optionally, a first customer risk report is generated according to the risk descriptions corresponding to the risk feature information, including:

[0034] The risk description, contribution, category and weight corresponding to each risk feature information are input into the risk report generation model to obtain the first customer risk report, wherein the order in which the risk feature information appears in the first customer risk report is related to the weight corresponding to the category.

[0035] In a second aspect, an embodiment of the present application provides a device for generating a customer risk report, including:

[0036] An acquisition module, used to acquire multiple characteristic information of a customer and determine the predicted risk level of the customer based on the multiple characteristic information;

[0037] A first determination module, configured to determine, based on a plurality of feature information and the predicted risk level, a contribution of each feature information to the predicted risk level;

[0038] A second determination module is used to determine, for any risk feature information, a risk description corresponding to the risk feature information according to the contribution corresponding to the risk feature information and based on a natural language model; wherein the risk feature information is selected from the plurality of feature information;

[0039] The generation module is used to generate a first customer risk report according to the risk descriptions corresponding to each risk feature information.

[0040] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a processor, and a memory communicatively connected to the processor;

[0041] The memory stores computer-executable instructions;

[0042] The processor executes the computer-executable instructions stored in the memory to implement the method described in the first aspect above.

[0043] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, they are used to implement the method described in the first aspect above.

[0044] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, which implements the method described in the first aspect when executed by a processor.

[0045] The present application provides a method, device, electronic device, medium and product for generating a customer risk report, the method comprising: obtaining multiple feature information of a customer, and determining the predicted risk level of the customer according to the multiple feature information; determining the contribution of each feature information to the predicted risk level according to the multiple feature information and the predicted risk level; for any risk feature information, determining the risk description corresponding to the risk feature information based on a natural language model according to the contribution corresponding to the risk feature information; wherein the risk feature information is screened out from the multiple feature information; generating a first customer risk report according to the risk descriptions corresponding to each risk feature information, so that the predicted risk level of the customer is first determined according to the multiple feature information, and then the contribution of each feature information to the predicted risk level is determined. The method can identify which features have the greatest impact on the risk level, and then use the natural language model to generate the risk description corresponding to the risk feature information, so that the risk report is easier to understand. The natural language description can convert complex risk data into easy-to-understand text information. Finally, the first customer risk report generated according to the risk descriptions corresponding to each risk feature information can provide customers with a clear risk overview and detailed risk description, which can improve the efficiency of writing customer risk reports and the accuracy of confirmed customer risk points. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0047] Figure 1 A schematic diagram of score calculation for a customer risk rating system provided in an embodiment of the present application;

[0048] Figure 2 An application scenario diagram provided for an embodiment of the present application;

[0049] Figure 3 A schematic diagram of a process for generating a customer risk report provided in an embodiment of the present application;

[0050] Figure 4 A schematic diagram of the contribution ratio of each feature information provided in an embodiment of the present application;

[0051] Figure 5 A schematic diagram of the structure of a customer risk report generating device provided in an embodiment of the present application;

[0052] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application.

[0053] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0054] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0055] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards, and corresponding operation entrances shall be provided for users to choose to authorize or refuse.

[0056] Risk rating is a key component of a bank's risk management system, helping banks to effectively manage risks, optimize resource allocation, and provide better customer service in a complex financial environment.

[0057] Currently, in the customer risk rating system, the system calculates the customer's quantitative score through a quantitative model, and then calculates the qualitative score based on a qualitative model. Finally, the system sums up the quantitative score and the qualitative score to give the customer a suitable system score and system grade.

[0058] Figure 1 A schematic diagram of score calculation for a customer risk rating system provided in an embodiment of the present application is shown in FIG. Figure 1As shown in the figure, the quantitative model is captured based on the given expert model, and different score ranges are given to customers who hit the corresponding scenarios. According to the difference between new and existing customers, the quantitative model also has different classifications, which covers four dimensions: customer characteristics, business characteristics, geographical factors, and industry factors. The degree of customer hit is calculated in each dimension to obtain a quantitative score. The qualitative model depends on whether the customer hits the external risk customer list and the internal risk customer list. Different lists correspond to different scores, and the score of the qualitative list accounts for a large proportion, so not all customers can hit it. Generally speaking, customers who hit the qualitative list will be assigned to a higher risk level or above. After integrating the quantitative score and the qualitative score, it is divided into 4 levels according to the threshold, and the 4 levels correspond to low risk, medium risk, higher risk, and high risk respectively.

[0059] The score source and system assessment level provided by the rating system are the main references for business personnel to write customer risk reports. However, business personnel will not invariably follow the scores and levels given by the system rating. In the process of writing reports, business personnel also need to read information from multiple sources, such as additional lists provided by other business lines, or analysis and summary of the customer's transactions in the past year, etc., and give a comprehensive rating result on this basis. However, the disadvantage of this method is that the writing efficiency is low, and under the premise that the risk level determined by the system is low risk or medium risk, the final risk level assessment and risk point determination are heavily dependent on the risk capture acumen of the business personnel. Business personnel with low acuity may not be able to fully identify the customer's potential risks, resulting in inaccurate risk ratings and risk points.

[0060] In view of this, the present application provides a method for generating a customer risk report, which can obtain multiple feature information of a customer and determine the customer's predicted risk level based on the multiple feature information; determine the contribution of each feature information to the predicted risk level based on the multiple feature information and the predicted risk level; for any risk feature information, determine the risk description corresponding to the risk feature information based on a natural language model according to the contribution corresponding to the risk feature information; wherein the risk feature information is screened out from multiple feature information; and generate a first customer risk report based on the risk description corresponding to each risk feature information. In this way, the customer's predicted risk level is first determined based on multiple feature information, and then the contribution of each feature information to the predicted risk level is determined. The method can identify which features have the greatest impact on the risk level, and then use the natural language model to generate the risk description corresponding to the risk feature information, so that the risk report is easier to understand. The natural language description can convert complex risk data into easy-to-understand text information. Finally, the first customer risk report generated based on the risk description corresponding to each risk feature information can provide customers with a clear risk overview and detailed risk description, which can improve the efficiency of writing customer risk reports and the accuracy of confirmed customer risk points.

[0061] Figure 2 An application scenario diagram provided for an embodiment of the present application, such as Figure 2 As shown, the business personnel input multiple characteristic information of the customer on the input interface of the business end, and / or automatically extract multiple characteristic information of the customer from a preset file. After the input is completed, click the confirmation button, the business end sends the multiple characteristic information of the customer to the server, the server obtains the multiple characteristic information of the customer, and determines the customer's predicted risk level based on the multiple characteristic information; determines the contribution of each characteristic information to the predicted risk level based on the multiple characteristic information and the predicted risk level; for any risk characteristic information, determines the risk description corresponding to the risk characteristic information based on the natural language model according to the contribution corresponding to the risk characteristic information; wherein the risk characteristic information is screened out from multiple characteristic information; generates a first customer risk report based on the risk descriptions corresponding to each risk characteristic information, and sends the first customer risk report to the business end, and the business end displays the obtained first customer risk report on the display interface.

[0062] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0063] Figure 3A schematic diagram of a method for generating a customer risk report provided in an embodiment of the present application. The execution subject of the present embodiment can be any device with a data processing function. The present application is specifically described with a server as the execution subject. Figure 3 As shown, a method for generating a customer risk report provided in an embodiment of the present application may include:

[0064] Step 301: Acquire multiple feature information of a customer, and determine the predicted risk level of the customer based on the multiple feature information.

[0065] Among them, the characteristic information is used to indicate various data points or attributes related to the customer. Any characteristic information includes a characteristic field and a corresponding characteristic value. For example, the characteristic field corresponding to a certain characteristic information of a customer is the number of dependent children, and the characteristic value is 3.

[0066] After obtaining multiple characteristic information of the customer, the server can clean the obtained multiple characteristic information. In the process of processing customer information characteristics, it is found that some basic information of some customers is missing or abnormal, so it is necessary to process abnormal data or missing values. For example, the customer's date of birth is 1900-01-01. This part of abnormal data can be calculated and repaired by intercepting the birthday information in the identity document. For the processing of missing values ​​in text type fields, for example, data with NULL (empty) customer transaction information can be modified to 0.

[0067] Different processing methods can be used for different types of fields depending on the specific situation. Here are some processing methods for some common types of fields:

[0068] (1) Enumeration fields: such as nationality, gender, education, marital status, etc., are mapped to numbers;

[0069] (2) Numeric fields: such as resident ID, personal banking customer level, and amount fields, retain the original data format;

[0070] (3) Basic information fields: fields that do not affect the customer’s risk level, such as name and mobile phone number, are removed;

[0071] (4) Date-related and Chinese word fields without enumeration values: such as work unit, organization name, investigation department, etc., are vectorized using a word-level representation method based on the Bert pre-training model;

[0072] (5) Long text fields: Natural language processing technology is used. First, the TF-IDF (Term Frequency-Inverse Document Frequency) algorithm is used to extract multiple keywords for each long text. Then, each keyword is vectorized using the Bert (Bidirectional Encoder Representations from Transformers) pre-trained model. Finally, feature fusion is performed using matrix addition, so that the vectorized features of the long text contain the semantic features of all keywords.

[0073] After cleaning, the multiple feature information is the original feature information. Then, the chi-square distribution is used to perform preliminary feature screening on the multiple original feature information. The chi-square test is used to evaluate the independence between each original feature information and the target variable (risk level). The specific steps are as follows: First, for each original feature information, the chi-square statistic and p-value are calculated to measure the correlation strength between the original feature information and the target variable, and the original feature information with a significantly low p-value is selected because the relationship between the original feature information and the target variable is more significant. In order to ensure that the screened original feature information has practical significance, the p-value threshold is set to 0.05, and the original feature information below this threshold will be retained. In addition, considering the business significance of the original feature information, the original feature information selected here is not only statistically significant, but also retains the reasonable features commonly used in the business. This step reduces the amount of original feature information, removes redundant and irrelevant original feature information, and improves the efficiency of subsequent feature dimensionality reduction and model training. After this step, the amount of original feature information is reduced by about 20%.

[0074] After completing the preliminary feature screening, for the multiple target feature information screened out from the multiple original feature information, principal component analysis (PCA) is applied to the target feature information with numerical eigenvalues ​​to further reduce the dimension. The target feature information with numerical eigenvalues ​​accounts for more than 82% of all target feature information. The PCA approach is: first, the target feature information with numerical eigenvalues ​​is standardized so that its mean is 0 and its variance is 1. After standardization, the PCA algorithm is applied to convert the original features into principal components. The goal of PCA is to generate a set of new unrelated features (principal components) by linearly combining the original features, and these principal components can explain most of the variance of the model. According to the cumulative variance contribution rate graph (scree plot), selecting the principal components that can explain most of the variance (95%) can reduce the number of target feature information and reduce the computational complexity.

[0075] The server can determine the customer's predicted risk level based on multiple feature information after feature screening and dimensionality reduction.

[0076] Optionally, the plurality of feature information includes: a plurality of current feature information, a plurality of historical feature information, and a historical predicted risk level; wherein the plurality of historical feature information is used to indicate the feature information used when the predicted risk level of the customer was determined last time, and the historical risk level is used to indicate the predicted risk level determined last time;

[0077] The categories corresponding to the multiple current feature information include: basic information category, extended information category, qualitative information category, quantitative information category, transaction information category, management and control information category, and large amount information category.

[0078] Among them, the basic information class is used to indicate the basic information of the user. For example, it can be information extracted from the basic information table of private customers. The basic information class can include characteristic information including: customer assets and liabilities, employment status, education level, real estate information, dependent children, number of valid / invalid accounts under the name, and the region where the commonly used telephone number is located.

[0079] The extended information class is used to indicate the user's extended information. For example, it can be information extracted from the ECIF (Enterprise Customer Information File) customer relationship table, customer basic information zipper table, and public related person table. The extended information class can include characteristic information including: the latest historical combined customer number rating, the combined customer number with the highest historical risk level, whether the customer has been cancelled, the number of cancellations, the highest risk level before cancellation, the number of entities served as a public related person, the risk level of serving as a public related person, etc.

[0080] The qualitative information class is used to indicate the hit status of multiple daily latest qualitative lists. For the qualitative lists that the customer hits, the feature value corresponding to this field can be assigned 1, and for the qualitative lists that the customer does not hit, the feature value corresponding to this field can be assigned 0.

[0081] The quantitative information class is used to indicate the information extracted from the latest daily quantitative indicator result table. According to the quantitative indicators involved by the customer, each field is filled with the corresponding result mapping value.

[0082] The transaction information class is used to indicate the user's transaction information. For example, the information can be extracted from the four private transaction tables (transfer, foreign exchange, deposits and withdrawals) and the password reset register. The transaction information class can include characteristic information such as: the customer's preferred transaction time period, the customer's preferred transaction channel, all-weather transaction points, the tendency to use integer multiples of amounts, the region where the customer's preferred transaction IP (Internet Protocol) belongs, the customer's preferred transaction summary, sensitive remarks, sensitive counterparty name, high and relatively high risk of counterparties, non-bank counterparties, and the number of password resets in the past year.

[0083] The control information class is used to indicate the user's account control information. For example, the risk view customer account control status can be extracted, and the customer's valid and non-valid hit status can be calculated for each control module.

[0084] The large amount information class is used to indicate the user's large amount information. For example, information in the large amount customer table and the large amount rule table can be extracted. The feature information included in the large amount information class includes whether the customer hits the large amount rule and whether the customer is a private banking customer.

[0085] In this way, by combining current feature information and historical feature information, the method can dynamically evaluate the user's risk status. Historical feature information and historical risk levels provide a time dimension perspective, so that risk assessment is not only based on the current status, but also takes into account the changing trend of user risk. The current feature information is subdivided into multiple categories, including basic information, extended information, qualitative information, quantitative information, transaction information, management and control information, and large-amount information. This classification method ensures comprehensive coverage of information, so that risk assessment can be analyzed from multiple angles, thereby improving the accuracy of the determined risk level.

[0086] Optionally, the actual risk level of the customer includes: low risk, medium risk, relatively high risk and high risk, and each feature information includes: a feature field and a feature value corresponding to the feature field;

[0087] The method for generating a customer risk report provided in the embodiment of the present application also includes:

[0088] Obtain multiple secondary customer risk reports written by business personnel for higher risk and high risk customers;

[0089] Merging the contents of multiple second customer risk reports to obtain a target customer risk report;

[0090] Arrange the words in the target customer risk report from high to low according to the frequency of occurrence to obtain a word frequency sequence;

[0091] At least one feature field is determined according to the vocabulary frequency sequence.

[0092] Specifically, the server can obtain multiple second customer risk reports written by business personnel for high-risk and high-risk customers. The second customer risk report is a real risk report written manually by business personnel. The contents of multiple second customer risk reports are merged to obtain the target customer risk report; extract text content from the target customer risk report, remove punctuation, special characters and extra spaces, and use different word segmentation tools (such as NLTK for English and Jieba for Chinese) to segment the text content into separate words according to different languages, traverse the word list after word segmentation, count the frequency of occurrence of each word, and sort the words from high to low according to their frequency of occurrence to obtain a word frequency sequence, which can be achieved using the `sorted()` function in Python or the sorting function of the `pandas` library. Finally, the top N words with the highest frequency in the word frequency sequence can be selected as feature fields.

[0093] In this way, by performing frequency analysis on the words in the target customer risk report, the fields that appear most frequently in high-risk customers can be identified. These high-frequency words may represent the most critical factors in risk assessment and can improve the efficiency and accuracy of selecting feature fields.

[0094] Optionally, determining the predicted risk level of the customer according to the plurality of feature information includes:

[0095] Determine the predicted risk level of the customer based on the trained risk level prediction model according to the plurality of feature information;

[0096] The customer risk report generation method provided by this application also includes:

[0097] Constructing a training sample set, wherein each training sample in the training sample set includes: a plurality of feature information, and a label corresponding to the training sample is: an actual risk level of the customer;

[0098] Inputting the training sample set into the initial risk level prediction model to obtain the predicted risk level corresponding to each training sample;

[0099] For each training sample, according to the actual risk level and the predicted risk level respectively corresponding to the training sample, the loss value is determined by the following loss function;

[0100]

[0101] Among them, the actual risk level Y = [y1, y2, y3, y4], and the predicted risk level is is the loss value, λ is a constant;

[0102] According to the loss values ​​corresponding to each training sample, the parameters of the initial risk level prediction model are adjusted to obtain a trained risk level prediction model.

[0103] The trained risk level prediction model can be any model that can predict the risk level of a user, and can be, for example, a lightGBM decision tree classifier model. The input of the lightGBM decision tree classifier model is a vector corresponding to multiple feature information or a vector corresponding to feature information after cleaning, screening and dimensionality reduction, and the output of the lightGBM decision tree classifier model can be the predicted risk level of the customer.

[0104] In the process of training the risk level prediction model, a custom loss function based on logarithm is used, in which the labels corresponding to the samples, that is, the actual risk levels, are processed into one-hot encodings of length 4*1. Therefore, the actual risk level Y = [y1, y2, y3, y4], and the predicted risk level is

[0105] The loss function is:

[0106] For each category i, the gradient is The second-order derivative Hessian corresponding to the loss function also changes accordingly, and the preferred λ is 0.1.

[0107] In this way, the optimized loss function adds a regularization term To control overfitting, and to adjust λ to adapt to different data distributions and task requirements, better capture the relationship between different categories, and thus improve the accuracy of the determined user risk level.

[0108] Step 302: Determine the contribution of each feature information to the predicted risk level based on multiple feature information and the predicted risk level.

[0109] Specifically, the server calculates the contribution of each feature information to the predicted risk level by calculating the Shapley value based on the multiple feature information and the predicted risk level.

[0110] Calculate the contribution of feature information j to the predicted risk level φ j The formula for (val) is as follows:

[0111]

[0112] Among them, S represents the feature set consisting of multiple feature information after removing feature information j, p is the number of multiple feature information, p! represents the factorial of the number of multiple feature information, which is all possible arrangements of feature information, and val(S) represents the output of the risk level prediction model when the input is feature set S.

[0113] Step 303: for any risk characteristic information, determine the risk description corresponding to the risk characteristic information according to the contribution degree corresponding to the risk characteristic information and based on the natural language model; wherein the risk characteristic information is screened out from the multiple characteristic information.

[0114] Specifically, for any risk characteristic information, the server determines the risk description corresponding to the risk characteristic information according to the contribution degree corresponding to the risk characteristic information and based on the natural language model; wherein the risk characteristic information is screened out from multiple characteristic information.

[0115] Optionally, each feature information includes: a feature field and a feature value corresponding to the feature field;

[0116] The method for generating a customer risk report provided in the embodiment of the present application also includes:

[0117] For any characteristic information, if the contribution degree corresponding to the characteristic information is a positive value, the characteristic information is determined to be risk characteristic information;

[0118] Accordingly, for any risk characteristic information, according to the contribution degree corresponding to the risk characteristic information, determining the risk description corresponding to the risk characteristic information includes:

[0119] For any risk characteristic information, if the characteristic value of the risk characteristic information is a numerical value, the description level corresponding to the characteristic information is determined according to the contribution degree corresponding to the risk characteristic information, and the risk description corresponding to the risk characteristic information is determined based on the first mapping relationship table according to the characteristic field and the description level corresponding to the risk characteristic information, wherein the first mapping relationship table is used to indicate the mapping relationship between the risk description and the characteristic field and the description level respectively; the description level includes: low, medium, and high;

[0120] If the characteristic value of the risk characteristic information is not a numerical value, the characteristic field and characteristic value corresponding to the risk characteristic information are used to determine the risk description corresponding to the risk characteristic information based on a second mapping relationship table, wherein the second mapping relationship table is used to indicate the mapping relationship between the risk characteristic and the characteristic field and characteristic value respectively.

[0121] Specifically, for any characteristic information, if the contribution degree corresponding to the characteristic information is a positive value, the server determines that the characteristic information is risk characteristic information; if the contribution degree corresponding to the characteristic information is a negative value or 0, the server determines that the characteristic information is not risk characteristic information.

[0122] For any risk characteristic information, if the characteristic value of the risk characteristic information is a numerical value, the server determines the description level corresponding to the characteristic information according to the contribution degree corresponding to the risk characteristic information. For example, when the contribution degree corresponding to a certain characteristic information is between A1 and A2, the corresponding description level is low, when the contribution degree corresponding to a certain characteristic information is between A2 and A3, the corresponding description level is medium, and when the contribution degree corresponding to a certain characteristic information is between A3 and A4, the corresponding description level is high, wherein A1, A2, A3, and A4 are all numerical values, and A1 is <A2<A3<A4。

[0123] Afterwards, the server determines the risk description corresponding to the risk characteristic information based on the first mapping relationship table according to the characteristic fields and description levels corresponding to the risk characteristic information, wherein the first mapping relationship table is used to indicate the mapping relationship between the risk description and the characteristic fields and description levels respectively; the description levels include: low, medium, and high.

[0124] Exemplarily, the characteristic field corresponding to a certain risk characteristic information is the number of abnormal accounts, the corresponding description level is high, and the corresponding risk description in the first mapping relationship table is that there are many abnormal accounts under the name of the customer.

[0125] If the characteristic value of the risk characteristic information is not a numerical value, the characteristic field and characteristic value corresponding to the risk characteristic information are used to determine the risk description corresponding to the risk characteristic information based on the second mapping relationship table, wherein the second mapping relationship table is used to indicate the mapping relationship between the risk characteristic and the characteristic field and characteristic value respectively.

[0126] For example, the characteristic field corresponding to a certain risk characteristic information is the account opening certificate type, and the corresponding characteristic value is a temporary ID card. Then the corresponding risk description in the second mapping relationship table is that the customer's account opening certificate is a temporary ID card.

[0127] In this way, by distinguishing between numerical and non-numerical risk feature information and using different mapping tables for processing, this flexibility enables the method to adapt to various types of data, improves the accuracy and applicability of risk descriptions, and by using mapping tables to determine risk descriptions, this method can standardize the risk assessment process, help maintain consistency in different risk assessment scenarios, and facilitate comparison and analysis of risk features in different customers or situations.

[0128] Step 304: Generate a first customer risk report based on the risk descriptions corresponding to the risk feature information.

[0129] Specifically, the server generates a first customer risk report according to the risk descriptions corresponding to each risk feature information.

[0130] The method for generating a customer risk report provided by the present application can obtain multiple characteristic information of a customer, and determine the predicted risk level of the customer according to the multiple characteristic information; determine the contribution of each characteristic information to the predicted risk level according to the multiple characteristic information and the predicted risk level; for any risk characteristic information, determine the risk description corresponding to the risk characteristic information based on the natural language model according to the contribution corresponding to the risk characteristic information; wherein the risk characteristic information is screened from the multiple characteristic information; and generate a first customer risk report according to the risk description corresponding to each risk characteristic information. In this way, the predicted risk level of the customer is first determined according to the multiple characteristic information, and then the contribution of each characteristic information to the predicted risk level is determined. The method can identify which characteristics have the greatest impact on the risk level, and then use the natural language model to generate the risk description corresponding to the risk characteristic information, so that the risk report is easier to understand. The natural language description can convert complex risk data into easy-to-understand text information. Finally, according to the risk description corresponding to each risk characteristic information, the first customer risk report generated can provide customers with a clear risk profile and detailed risk description, which can improve the efficiency of writing customer risk reports and the accuracy of confirmed customer risk points.

[0131] Optionally, the customer generates a first customer risk report according to the risk descriptions corresponding to each risk feature information, including:

[0132] The risk description, contribution, category and weight corresponding to each risk feature information are input into the risk report generation model to obtain the first customer risk report, wherein the order in which the risk feature information appears in the first customer risk report is related to the weight corresponding to the category.

[0133] The risk report generation model may be any model that can generate a risk report. For example, the risk report generation model may be an open source large model Qwen2-72b.

[0134] The server inputs the risk description, contribution, category and weight corresponding to each risk feature information into the risk report generation model to obtain the first customer risk report. The order in which each risk feature information appears in the first customer risk report is related to the weight corresponding to the category. The higher the weight corresponding to the risk feature information, the higher the corresponding order of appearance.

[0135] Exemplarily, the output first customer risk report is: the customer is a credit card customer, has many abnormal accounts in his name, has small debts, etc., the account opening document is a temporary ID card, the risk level of the customer number before the merger is general risk, was an unidentified individual customer, and had many round-the-clock trading days in a year.

[0136] The first customer risk report can also include the user's predicted risk level and an intuitive chart of the proportion of each feature information in the risk level generation. Figure 4 A schematic diagram of the contribution ratio of each feature information provided in an embodiment of the present application is shown in FIG. Figure 4 As shown in the figure, the contribution percentage of each feature information is marked. Business personnel can refer to them one by one in order, and for any feature information, when the mouse moves to Figure 4 When you click the rectangular area corresponding to the feature information, you can also view the specific value of the contribution corresponding to the feature information. As can be seen from the figure, the customer's last risk level has a negative impact on the current risk level assessment. The feature value shows that the customer's last risk level was 2 (general risk). The last risk level was assessed when the customer opened an account, which has little reference value for this risk level 4. This feature that has a negative impact on the current risk level assessment is only used in feature impact comparison and developer debugging. The specific speech generation report does not display such features.

[0137] In this way, by inputting the risk description, contribution, category and weight of each risk feature information into the risk report generation model, a personalized risk report for each customer can be quickly generated. In the generated risk report, the order of appearance of each risk feature information is related to the weight of the category to which it belongs. This means that more important risk features will be displayed more prominently in the report, helping business personnel quickly identify and focus on the most critical risk factors, thereby improving the readability and practicality of the report.

[0138] Corresponding to the above-mentioned customer risk report generation method, the embodiment of the present application also provides a customer risk report generation device, Figure 5 This is a schematic diagram of the structure of a customer risk report generation device provided in an embodiment of the present application. Figure 5 As shown, the device comprises:

[0139] The acquisition module 501 is used to acquire multiple characteristic information of the customer and determine the predicted risk level of the customer according to the multiple characteristic information;

[0140] A first determination module 502, configured to determine, based on a plurality of feature information and the predicted risk level, a contribution of each feature information to the predicted risk level;

[0141] The second determination module 503 is used to determine, for any risk feature information, a risk description corresponding to the risk feature information according to the contribution corresponding to the risk feature information and based on a natural language model; wherein the risk feature information is selected from the plurality of feature information;

[0142] The generating module 504 is used to generate a first customer risk report according to the risk descriptions corresponding to each risk characteristic information.

[0143] Optionally, the plurality of feature information includes: a plurality of current feature information, a plurality of historical feature information, and a historical predicted risk level; wherein the plurality of historical feature information is used to indicate the feature information used when the predicted risk level of the customer was determined last time, and the historical risk level is used to indicate the predicted risk level determined last time;

[0144] The categories corresponding to the multiple current feature information include: basic information category, extended information category, qualitative information category, quantitative information category, transaction information category, management and control information category, and large amount information category.

[0145] Optionally, when determining the predicted risk level of the customer according to the plurality of feature information, the acquisition module 501 is specifically configured to:

[0146] Determine the predicted risk level of the customer based on the trained risk level prediction model according to the plurality of feature information;

[0147] The acquisition module 501 is also used for:

[0148] Constructing a training sample set, wherein each training sample in the training sample set includes: a plurality of feature information, and a label corresponding to the training sample is: an actual risk level of the customer;

[0149] Inputting the training sample set into the initial risk level prediction model to obtain the predicted risk level corresponding to each training sample;

[0150] For each training sample, according to the actual risk level and the predicted risk level respectively corresponding to the training sample, the loss value is determined by the following loss function;

[0151]

[0152] Among them, the actual risk level Y = [y1, y2, y3, y4], and the predicted risk level is is the loss value, λ is a constant;

[0153] According to the loss values ​​corresponding to each training sample, the parameters of the initial risk level prediction model are adjusted to obtain a trained risk level prediction model.

[0154] Optionally, the actual risk level of the customer includes: low risk, medium risk, relatively high risk and high risk, and each feature information includes: a feature field and a feature value corresponding to the feature field;

[0155] The acquisition module 501 is also used for:

[0156] Obtain multiple secondary customer risk reports written by business personnel for higher risk and high risk customers;

[0157] Merging the contents of multiple second customer risk reports to obtain a target customer risk report;

[0158] Arrange the words in the target customer risk report from high to low according to the frequency of occurrence to obtain a word frequency sequence;

[0159] At least one feature field is determined according to the vocabulary frequency sequence.

[0160] Optionally, each feature information includes: a feature field and a feature value corresponding to the feature field;

[0161] The second determining module 503 is further used for:

[0162] For any characteristic information, if the contribution degree corresponding to the characteristic information is a positive value, the characteristic information is determined to be risk characteristic information;

[0163] Accordingly, the second determining module 503 is specifically configured to:

[0164] For any risk characteristic information, if the characteristic value of the risk characteristic information is a numerical value, the description level corresponding to the characteristic information is determined according to the contribution degree corresponding to the risk characteristic information, and the risk description corresponding to the risk characteristic information is determined based on the first mapping relationship table according to the characteristic field and the description level corresponding to the risk characteristic information, wherein the first mapping relationship table is used to indicate the mapping relationship between the risk description and the characteristic field and the description level respectively; the description level includes: low, medium, and high;

[0165] If the characteristic value of the risk characteristic information is not a numerical value, the characteristic field and characteristic value corresponding to the risk characteristic information are used to determine the risk description corresponding to the risk characteristic information based on a second mapping relationship table, wherein the second mapping relationship table is used to indicate the mapping relationship between the risk characteristic and the characteristic field and characteristic value respectively.

[0166] Optionally, the generating module 504 is specifically used for:

[0167] The risk description, contribution, category and weight corresponding to each risk feature information are input into the risk report generation model to obtain the first customer risk report, wherein the order in which the risk feature information appears in the first customer risk report is related to the weight corresponding to the category.

[0168] The specific implementation principles and effects of the customer risk report generation device provided in the embodiment of the present application can be found in the aforementioned embodiment and will not be repeated here.

[0169] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 6 As shown, the electronic device of this embodiment may include:

[0170] at least one processor 601; and

[0171] A memory 602 communicatively connected to the at least one processor;

[0172] The memory 602 stores instructions that can be executed by the at least one processor 601, and the instructions are executed by the at least one processor 601 to enable the electronic device to execute the method described in any of the above embodiments.

[0173] Optionally, the memory 602 may be independent or integrated with the processor 601 .

[0174] The implementation principle and technical effects of the electronic device provided in this embodiment can be found in the aforementioned embodiments and will not be described in detail here.

[0175] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions. When a processor executes the computer-executable instructions, the method described in any of the above embodiments is implemented.

[0176] An embodiment of the present application further provides a computer program product, including a computer program, which implements the method described in any of the above embodiments when executed by a processor.

[0177] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0178] The above-mentioned integrated module implemented in the form of a software function module can be stored in a computer-readable storage medium. The above-mentioned software function module is stored in a storage medium, including a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to perform some steps of the method described in each embodiment of the present application.

[0179] It should be understood that the above-mentioned processor can be a processing unit (Central Processing Unit, CPU), or other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application-specific integrated circuits (Application Specific Integrated Circuit, ASIC), etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the application can be directly embodied as being executed by a hardware processor, or executed by a combination of hardware and software modules in the processor. The memory may include a high-speed random access memory (Random Access Memory, RAM), and may also include non-volatile memory (Non-Volatile Memory, NVM), such as at least one disk storage, and can also be a USB flash drive, a mobile hard disk, a read-only memory, a disk or an optical disk, etc.

[0180] The above storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The storage medium can be any available medium that can be accessed by a general or special-purpose computer.

[0181] An exemplary storage medium is coupled to a processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the storage medium can also exist as discrete components in an electronic device or a main control device.

[0182] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0183] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0184] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0185] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for generating a customer risk report, characterized in that: include: Acquire multiple characteristic information of the customer, and determine the predicted risk level of the customer based on the multiple characteristic information; Determining, based on the plurality of feature information and the predicted risk level, a contribution of each feature information to the predicted risk level; For any risk characteristic information, according to the contribution degree corresponding to the risk characteristic information, based on the natural language model, determine the risk description corresponding to the risk characteristic information; wherein the risk characteristic information is selected from the multiple characteristic information; A first customer risk report is generated according to the risk descriptions corresponding to each risk characteristic information.

2. The method according to claim 1, characterized in that The plurality of characteristic information includes: a plurality of current characteristic information, a plurality of historical characteristic information, and a historical predicted risk level; wherein the plurality of historical characteristic information is used to indicate the characteristic information used when the predicted risk level of the customer was determined last time, and the historical risk level is used to indicate the predicted risk level determined last time; The categories corresponding to the multiple current feature information include: basic information category, extended information category, qualitative information category, quantitative information category, transaction information category, management and control information category, and large amount information category.

3. The method according to claim 1, characterized in that Determining the predicted risk level of the customer according to the plurality of feature information includes: Determine the predicted risk level of the customer based on the trained risk level prediction model according to the plurality of feature information; The method further comprises: Constructing a training sample set, wherein each training sample in the training sample set includes: a plurality of feature information, and a label corresponding to the training sample is: an actual risk level of the customer; Inputting the training sample set into the initial risk level prediction model to obtain the predicted risk level corresponding to each training sample; For each training sample, according to the actual risk level and the predicted risk level respectively corresponding to the training sample, the loss value is determined by the following loss function; Among them, the actual risk level Y = [y1, y2, y3, y4], and the predicted risk level is is the loss value, λ is a constant; According to the loss values ​​corresponding to each training sample, the parameters of the initial risk level prediction model are adjusted to obtain a trained risk level prediction model.

4. The method according to claim 1, characterized in that: The actual risk level of the customer includes: low risk, medium risk, relatively high risk and high risk. Each feature information includes: a feature field and a feature value corresponding to the feature field; The method further comprises: Obtain multiple secondary customer risk reports written by business personnel for higher risk and high risk customers; Merging the contents of multiple second customer risk reports to obtain a target customer risk report; Arrange the words in the target customer risk report from high to low according to the frequency of occurrence to obtain a word frequency sequence; At least one feature field is determined according to the vocabulary frequency sequence.

5. The method according to any one of claims 1 to 4, characterized in that: Each feature information includes: a feature field and a feature value corresponding to the feature field; The method further comprises: For any characteristic information, if the contribution degree corresponding to the characteristic information is a positive value, the characteristic information is determined to be risk characteristic information; Accordingly, for any risk characteristic information, according to the contribution degree corresponding to the risk characteristic information, determining the risk description corresponding to the risk characteristic information includes: For any risk characteristic information, if the characteristic value of the risk characteristic information is a numerical value, the description level corresponding to the characteristic information is determined according to the contribution degree corresponding to the risk characteristic information, and the risk description corresponding to the risk characteristic information is determined based on the first mapping relationship table according to the characteristic field and the description level corresponding to the risk characteristic information, wherein the first mapping relationship table is used to indicate the mapping relationship between the risk description and the characteristic field and the description level respectively; the description level includes: low, medium, and high; If the characteristic value of the risk characteristic information is not a numerical value, the characteristic field and characteristic value corresponding to the risk characteristic information are used to determine the risk description corresponding to the risk characteristic information based on a second mapping relationship table, wherein the second mapping relationship table is used to indicate the mapping relationship between the risk characteristic and the characteristic field and characteristic value respectively.

6. The method according to claim 5, characterized in that Generate the first customer risk report based on the risk descriptions corresponding to each risk feature information, including: The risk description, contribution, category and weight corresponding to each risk feature information are input into the risk report generation model to obtain the first customer risk report, wherein the order in which the risk feature information appears in the first customer risk report is related to the weight corresponding to the category.

7. A customer risk report generating device, characterized in that: include: An acquisition module, used to acquire multiple characteristic information of a customer and determine the predicted risk level of the customer based on the multiple characteristic information; A first determination module, configured to determine, based on a plurality of feature information and the predicted risk level, a contribution of each feature information to the predicted risk level; A second determination module is used to determine, for any risk feature information, a risk description corresponding to the risk feature information according to the contribution degree corresponding to the risk feature information and based on a natural language model; wherein the risk feature information is selected from the plurality of feature information; The generation module is used to generate a first customer risk report according to the risk descriptions corresponding to each risk feature information.

8. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 6 when executed by a processor.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images