Cross-domain customer credit assessment method and related device

Abstract

The invention discloses a cross-domain customer credit assessment method and a related device, and relates to the technical field of software, and the method comprises the steps: obtaining target domain data, and extracting a first data feature corresponding to the target domain data; calling a credit evaluation model, wherein the credit evaluation model is constructed by learning data distribution of the first source domain sample and the target domain sample based on a naive Bayes algorithm; and predicting a credit risk level corresponding to the first data feature through the credit evaluation model. According to the method, the data distribution of the source domain sample and the target domain sample is learned based on the naive Bayesian algorithm, so that the credit evaluation model can be continuously attached to the real data distribution of the target domain, the classification uncertainty of the target domain is greatly reduced, and the credit risk level can be quickly and accurately identified during prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the software technical field, and particularly relates to a cross-domain customer credit evaluation method and related device. BACKGROUND

[0002] With the continuous expansion and integration of the financial market, the business of financial institutions widely involves bank credit, securities investment, insurance services and emerging financial technology fields and other aspects, and the customer group is increasingly complex and diverse.

[0003] Therefore, how to quickly and accurately identify the credit risk of cross-domain customers has become a problem that financial institutions urgently need to solve at the present stage. SUMMARY

[0004] In view of the above problems, the present application provides a cross-domain customer credit evaluation method and related device to achieve the purpose of quickly and accurately identifying the credit risk of cross-domain customers. The specific scheme is as follows:

[0005] The first aspect of the present application provides a cross-domain customer credit evaluation method, which comprises:

[0006] Obtaining target domain data and extracting first data features corresponding to the target domain data;

[0007] Calling a credit evaluation model, wherein the credit evaluation model is constructed based on learning data distribution of first source domain samples and target domain samples by using a naive Bayes algorithm;

[0008] Predicting a credit risk level corresponding to the first data features by using the credit evaluation model.

[0009] In a possible implementation, the extracting the first data features corresponding to the target domain data comprises:

[0010] Extracting the first data features from the target domain data by using a feature extractor, wherein the feature extractor is obtained by training a deep neural network using second source domain samples.

[0011] In a possible implementation, the target domain samples include labeled first target domain samples and unlabeled second target domain samples, and the construction process of the credit evaluation model comprises:

[0012] Calling a basic model constructed based on the naive Bayes algorithm;

[0013] Extracting second data features corresponding to the first source domain samples, and learning a first data distribution of the second data features under the label of the first source domain samples by using the basic model;

[0014] extracting third data features corresponding to the first target domain samples, and learning, by the base model, second data distribution of the third data features under labels of the first target domain samples;

[0015] selecting, from the second target domain samples, part of the target domain samples for this time of learning, extracting fourth data features of the part of the target domain samples, and predicting, by the base model, pseudo labels corresponding to the fourth data features by using the first data distribution and the second data distribution;

[0016] calculating average entropy of the base model in predicting the pseudo labels;

[0017] if the average entropy does not satisfy a corresponding convergence condition, grouping the first target domain samples and the second target domain samples with pseudo labels into new first target domain samples, grouping the third data features and the fourth data features into new third data features, and returning to execute the step of selecting, from the second target domain samples, part of the target domain samples for this time of learning;

[0018] if the average entropy satisfies the convergence condition, taking the base model after this time of learning as the credit evaluation model.

[0019] In a possible implementation, the predicting, by the base model, the pseudo labels corresponding to the fourth data features by using the first data distribution and the second data distribution comprises:

[0020] obtaining first labels and prediction probabilities of the first labels of the base model in predicting the fourth data features by using the first data distribution, and calculating overall entropy values of the first labels according to the prediction probabilities of the first labels;

[0021] obtaining second labels and prediction probabilities of the second labels of the base model in predicting the fourth data features by using the second data distribution, and calculating overall entropy values of the second labels according to the prediction probabilities of the second labels;

[0022] comparing the overall entropy values of the first labels with the overall entropy values of the second labels, and taking one of the labels with the minimum overall entropy value as the pseudo label.

[0023] In a possible implementation, the process of constructing the credit evaluation model further comprises:

[0024] if the average entropy does not satisfy the convergence condition, and the cumulative number of times of learning of the base model reaches a corresponding upper limit of the number of times, taking the base model after one time of learning with the minimum average entropy as the credit evaluation model.

[0025] The second aspect of the application provides a cross-domain customer credit evaluation device, the cross-domain customer credit evaluation device comprises:

[0026] a model construction module configured to construct a credit evaluation model based on learning data distribution of the first source domain sample and the target domain sample according to a naive Bayes algorithm;

[0027] a credit evaluation module configured to acquire target domain data, extract first data features corresponding to the target domain data, call the credit evaluation model, and predict a credit risk level corresponding to the first data features by using the credit evaluation model.

[0028] In a possible implementation, the credit evaluation module configured to extract the first data features corresponding to the target domain data is specifically configured to:

[0029] extract the first data features from the target domain data by using a feature extractor, and the feature extractor is obtained by training a deep neural network using second source domain samples.

[0030] The third aspect of the application provides a computer program product, comprising computer readable instructions, when the computer readable instructions run on an electronic device, the electronic device can implement the cross-domain customer credit evaluation method of the first aspect or any implementation manner of the first aspect.

[0031] The fourth aspect of the application provides an electronic device, comprising at least one processor and a memory connected with the processor, wherein:

[0032] the memory is configured to store a computer program;

[0033] the processor is configured to execute the computer program, so that the electronic device can implement the cross-domain customer credit evaluation method of the first aspect or any implementation manner of the first aspect.

[0034] The fifth aspect of the application provides a computer storage medium, the storage medium carries one or more computer programs, when the one or more computer programs are executed by an electronic device, the electronic device can implement the cross-domain customer credit evaluation method of the first aspect or any implementation manner of the first aspect.

[0035] By means of the technical scheme, the cross-domain customer credit evaluation method and the related device are provided, which comprises the following steps: obtaining target domain data and extracting first data features corresponding to the target domain data; calling a credit evaluation model, the credit evaluation model being constructed based on a naive Bayes algorithm learning data distribution of first source domain samples and target domain samples; and predicting a credit risk level corresponding to the first data features through the credit evaluation model. In the present application, the data distribution of the source domain samples and the target domain samples is learned based on the naive Bayes algorithm, so that the credit evaluation model can be continuously fitted to the real data distribution of the target domain, thereby greatly reducing the classification uncertainty of the target domain and ensuring that the credit risk level can be quickly and accurately identified during prediction. BRIEF DESCRIPTION OF DRAWINGS

[0036] The above and other features, advantages, and aspects of the present disclosure will become more apparent by describing in detail the following specific embodiments thereof with reference made to the accompanying drawings. Throughout the drawings, the same or similar reference numerals are used to represent the same or similar elements. It should be understood that the drawings are schematic and elements and features are not necessarily to scale.

[0037] Figure 1 A process schematic diagram of a cross-domain customer credit evaluation method provided by an embodiment of the present application;

[0038] Figure 2 A partial process schematic diagram of a cross-domain customer credit evaluation method provided by an embodiment of the present application;

[0039] Figure 3 Another partial process schematic diagram of a cross-domain customer credit evaluation method provided by an embodiment of the present application;

[0040] Figure 4 A structure schematic diagram of a cross-domain customer credit evaluation device provided by an embodiment of the present application;

[0041] Figure 5 A structure schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0042] The embodiments of the present application are described below in conjunction with the accompanying drawings. The terms used in the embodiment part of the present application are only used to explain the specific embodiments of the present application, and are not intended to limit the present application.

[0043] The embodiments of the present application are described below in conjunction with the accompanying drawings. It is known to those skilled in the art that, with the development of technology and the appearance of new scenarios, the technical solutions provided by the embodiments of the present application are also applicable to similar technical problems.

[0044] The terms "first", "second", and the like in the description and in the claims of the present application and above-described drawings are used to distinguish similar objects and are not necessarily used to describe a specific sequential or chronological order. It is to be understood that the terms so used are interchangeable under appropriate circumstances and that the embodiments of the present application are capable of functioning in other sequences than described or otherwise illustrated herein. Moreover, the terms "comprise", "have" and any variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of elements is not necessarily limited to those elements, but can include other elements not expressly listed or inherent to such process, method, system, article, or apparatus.

[0045] For the convenience of understanding the present application, the relevant concepts involved in the present application will be described first as follows:

[0046] Naive Bayes algorithm: a classification algorithm based on Bayes theorem and feature conditional independence assumption. It calculates the posterior probability through the prior probability and the conditional probability of each feature in the data set, and then determines the class to which the sample belongs.

[0047] Prior probability: subjective estimate of the probability of an event based on past experience or knowledge before obtaining new observation data. It does not depend on the current observation data, and reflects the initial cognition of the event.

[0048] Conditional probability: the probability of an event occurring given that another event has occurred. It represents the degree of association between two events, calculated by restricting the sample space.

[0049] Posterior probability: the probability of an event occurring after obtaining new observation data, calculated by Bayes theorem combining prior probability and observation data. It is an update and correction of the prior probability, and more accurately reflects the likelihood of the event occurring under the existing evidence.

[0050] Entropy: Entropy is a measure of the uncertainty or average amount of information contained in a random variable, calculated by probability distribution. The greater the entropy value, the higher the uncertainty.

[0051] Cross-entropy: used to measure the difference between two probability distributions, often used as a loss function in machine learning to measure the distance between model predictions and true distributions.

[0052] Gaussian distribution: also known as normal distribution, is a common continuous probability distribution. Its probability density function is a bell-shaped curve with symmetry, determined by mean and variance, widely used in natural science, social science and engineering fields.

[0053] Source domain and target domain: The source domain refers to a domain with a large amount of labeled data and known data characteristics and distribution, whose data and knowledge can be used as a reference and basis to help solve problems in the target domain. The target domain is the domain where the problem to be solved is located, and usually has insufficient or no labeled data, and the data characteristics and source domain are different, and the knowledge of the source domain needs to be used for analysis and processing.

[0054] In the scenario of customer credit risk assessment, commonly used customer data includes customer basic information (including age, occupation, income, etc.), credit record data (credit card repayment records, loan default situations, etc.), financial behavior data (involving transaction flow, account balance changes, etc.), financial data (such as income, assets, liabilities, etc.), and many other aspects. However, the distribution of data characteristics in the source domain and the target domain may be different. For example, in the source domain, most are traditional industry practitioners, while in the target domain, most are emerging industry practitioners. Due to differences in business types or market environments, the emerging business in the target domain may lack historical credit data, or the distribution of default rates is different from the source domain. In addition, the transaction mode in the target domain may change due to new financial products or services, and the transaction frequency and amount distribution are different from the source domain.

[0055] The risk status of customer groups is influenced by various factors in different fields, such as fluctuations in macroeconomic situation, changes in industry competition situation, instability of market price, and differences in personal consumption and credit behavior patterns. In this context, accurately and quickly identifying the risk of cross-domain customers is of great significance to financial institutions.

[0056] On the one hand, accurate credit risk assessment is the key to ensuring asset quality. If financial institutions cannot accurately identify the credit risk of customers, they may invest funds in high-risk customers, leading to an increase in non-performing assets, and thus affecting the financial health and stability of the institution. For example, in the credit business, if a high default risk is not identified for a corporate customer due to the recession of the industry in which it operates and its own poor management, and a large amount of loans are still provided to it, once the enterprise defaults, the bank will suffer huge losses and the asset quality will decrease significantly.

[0057] On the other hand, rapid credit risk identification helps optimize resource allocation and enhance market competitiveness. Financial institutions that can quickly assess customer credit risk can adjust credit and investment strategies in time, invest resources in high-quality customers and projects with low risk and high return, and improve resource utilization efficiency. At the same time, in the highly competitive financial market, the ability to quickly and accurately assess credit risk enables financial institutions to make decisions faster, provide better services to customers, attract more customers, and enhance their competitiveness in the market. In addition, accurate credit risk assessment can also maintain financial stability and lay a solid foundation for the stable operation of the financial system, avoiding systemic financial risks caused by risk events in individual institutions.

[0058] To solve the above problems, the embodiment of the application provides a cross-domain customer credit evaluation method. The cross-domain customer credit evaluation method of the embodiment of the application will be described in detail below with reference to the drawings.

[0059] Referring to Figure 1 , Figure 1 The flowchart of the cross-domain customer credit evaluation method provided by the embodiment of the application is shown in FIG. 1. As shown in FIG. 1, the cross-domain customer credit evaluation method provided by the embodiment of the application can include steps S101 to S103, which will be described in detail below. Figure 1

[0060] S101, obtaining target domain data and extracting first data features corresponding to the target domain data.

[0061] In the embodiment of the application, the target domain data is customer data to be evaluated in the target domain, which can include customer basic information, credit record data, financial behavior data, etc. Then, high-dimensional data features (i.e., first data features) contained in the target domain data are extracted.

[0062] In one possible implementation, feature extraction can be achieved by designing a feature extractor, which can be obtained by training a deep neural network using source domain samples. For this purpose, the cross-domain customer credit evaluation method provided by the embodiment of the application, wherein the step S101 of "extracting first data features corresponding to the target domain data" can include the following steps:

[0063] The first data features are extracted from the target domain data by the feature extractor, which is obtained by training a deep neural network using second source domain samples.

[0064] In the embodiment of the application, the deep neural network is trained on the source domain with labels, and the weight parameters of the deep neural network are frozen after training to solidify the feature extraction mode learned from the source domain to form the feature extractor. Specifically:

[0065] There are a large number of sample data with labels in the source domain, and part of the sample data is selected as the second source domain sample, and the deep neural network is trained in a supervised learning manner. During training, a suitable loss function is usually selected, such as the cross-entropy loss function commonly used for classification problems. Assuming that the number of second source domain samples is , the number of types of credit risk levels is , the second source domain sample belongs to the credit risk level , and the label is (taking 0 or 1), and the deep neural network predicts the second source domain sample ​The probability of belonging to the credit risk level The cross-entropy loss function may be expressed as Through optimization algorithms such as stochastic gradient descent (SGD), the weight parameters of the deep neural network are constantly adjusted, so that the loss function value is constantly reduced, that is, the predicted value of the model is as close to the true label as possible. When the training meets certain conditions, such as reaching the set accuracy requirement or the specified number of iterations, the training is ended, the weight parameters of the deep neural network are frozen, and they will not change in subsequent use, thereby forming a feature extractor.

[0066] After the feature extractor obtains the input data, it will go through complex calculations of multiple layers of neurons, such as first passing through a convolutional layer to preliminarily extract and screen features from the input data, and then passing through a fully connected layer to integrate and further convert these features. After such processing, deep and more representative features are extracted from the input data, which are generally presented in the form of a high-dimensional vector. For example, assuming that the input data is customer data, after a series of calculations of the neural network, a 128-dimensional feature vector is finally output, which contains key information of the input data as data features for subsequent operations.

[0067] S102, call the credit evaluation model, the credit evaluation model is constructed based on the data distribution of the first source domain sample and the target domain sample learned by the naive Bayes algorithm.

[0068] In the embodiments of the present application, a credit evaluation model is constructed in advance, which is constructed based on the data distribution of sample data in the source domain and the target domain learned by the naive Bayes algorithm, wherein the sample data in the source domain is labeled, and part of the sample data in the target domain is labeled and part is not labeled.

[0069] Referring to Figure 2 , Figure 2 is a partial flowchart of a cross-domain customer credit evaluation method provided by the embodiments of the present application. As shown in Figure 2 , the cross-domain customer credit evaluation method provided by the embodiments of the present application, wherein the target domain sample includes the first target domain sample with label and the second target domain sample without label, the construction process of the credit evaluation model can include steps S201 to S207, which will be described in detail below.

[0070] S201, call the basic model constructed based on the naive Bayes algorithm.

[0071] In the embodiments of the present application, a corresponding classification model is constructed based on the naive Bayes algorithm as a basic model.

[0072] ​S202, extract the second data features corresponding to the first source domain sample, and learn the first data distribution of the second data features under the label of the first source domain sample through the basic model.

[0073] In the embodiment of the application, there are a large number of sample data with labels in the source domain, part of which is selected as the first source domain sample, and high-dimensional data features (i.e. second data features) contained in the first source domain sample are extracted. Specifically, a feature extractor trained based on a deep neural network can be used to extract the second data features. Further, the basic model learns the data distribution (i.e. the first data distribution) of different labels on all dimensions of the second data features.

[0074] Naive Bayes algorithm has an important assumption that features are independent of each other, based on which the basic model analyzes the second data features in depth. For different labels, the parameters of the Gaussian distribution (normal distribution) on each dimension of the second data features are estimated, i.e. the mean and variance of each label on each feature dimension are calculated. For the Gaussian distribution of label on feature dimension , the probability density function is , where is the feature value of the first source domain sample on feature dimension . By calculating these parameters, the basic model can learn the data distribution of the second data features under each label, which provides important prior probability distribution information for predicting the pseudo label of the unlabeled sample data in the target domain subsequently.

[0075] S203, extract the third data features corresponding to the first target domain sample, and learn the second data distribution of the third data features under the label of the first target domain sample through the basic model.

[0076] In the embodiment of the application, there are a small amount of sample data with labels in the target domain, which are selected as the first target domain sample, and high-dimensional data features (i.e. third data features) contained in the first target domain sample are extracted. Specifically, a feature extractor trained based on a deep neural network can be used to extract the third data features. Further, the basic model learns the data distribution (i.e. the second data distribution) of different labels on all dimensions of the third data features.

[0077] It should be noted that although the data distribution of the target domain and the source domain may be different, sharing the same feature extractor can to some extent mine the common features between cross-domain data.

[0078] Similar to the learning process of the second data feature in step S202, when the base model deeply analyzes the third data feature, the parameters and probability density functions of the Gaussian distribution are estimated on each dimension of the third data feature for different labels, so that the base model can learn the data distribution of the third data feature under each label. Due to the difference in data distribution between the target domain and the source domain, this operation can make the base model better adapt to the data characteristics of the target domain, which is used for predicting the pseudo label of the unlabeled sample data in the target domain.

[0079] S204, selecting part of the target domain samples for this learning from the second target domain samples, extracting the fourth data features of the part of the target domain samples, and predicting the pseudo label corresponding to the fourth data features by the base model using the first data distribution and the second data distribution.

[0080] In the embodiment of the application, there are a large number of sample data without labels in the target domain, part of which is selected as the second target domain sample. Further, part of the target domain samples for this learning is selected from the second target domain sample, and the high-dimensional data features (i.e. the fourth data features) contained in the part of the target domain samples are extracted. Specifically, the feature extractor trained based on the deep neural network can be used to extract the fourth data features.

[0081] Further, the base model uses the first data distribution and the second data distribution to predict the label corresponding to the fourth data features as the pseudo label thereof.

[0082] In a possible implementation, the base model uses the first data distribution and the second data distribution to predict the label of the fourth data features, respectively, determines the more reliable data distribution according to the overall entropy value of the label, and takes the result given by the more reliable data distribution as the pseudo label. Referring to Figure 3 , Figure 3 Another part of the flowchart of the cross-domain customer credit evaluation method provided by the embodiment of the application. As shown in Figure 3 The cross-domain customer credit evaluation method provided by the embodiment of the application, wherein the step S204 of "predicting the pseudo label corresponding to the fourth data features by the base model using the first data distribution and the second data distribution" can include steps S301 to S303, which will be described in detail below.

[0083] S301, obtaining the first label corresponding to the fourth data features and the prediction probability thereof predicted by the base model using the first data distribution; and calculating the overall entropy value of the first label according to the prediction probability of the first label.

[0084] In the embodiment of the application, the base model uses the first data distribution to predict the label (i.e. the first label) corresponding to the fourth data feature. For some of the samples in the part of the target domain samples , the base model can give the probability that the fourth data belongs to the label , wherein, represents the corresponding feature of the sample in the fourth data feature, represents the feature value of the sample in the feature dimension , represents the total number of dimensions of the fourth data feature, represents the number of labels in the first label (i.e., the number of types of credit risk levels). Further, the overall entropy value of the first label is calculated according to the prediction probability , wherein, represents the number of samples in the part of the target domain samples.

[0085] S302, obtaining the base model to predict the second label corresponding to the fourth data feature and the prediction probability thereof using the second data distribution; and calculating the overall entropy value of the second label according to the prediction probability of the second label.

[0086] In the embodiments of the present application, the base model predicts the label (i.e., the second label) corresponding to the fourth data feature using the second data distribution. Similar to step S301, the base model gives the probability that the part of the target domain samples belongs to the second label, and the overall entropy value of the second label is calculated accordingly.

[0087] S303, comparing the overall entropy value of the first label with the overall entropy value of the second label, and taking the label with the smallest overall entropy value as the pseudo label.

[0088] In the embodiments of the present application, the smaller the overall entropy value is, the lower the uncertainty of the data distribution is, which means that the prediction result given by the data distribution is more reliable. Therefore, by comparing the overall entropy value of the first label with the overall entropy value of the second label, the label with the smallest overall entropy value can be selected as the pseudo label, which provides more label information for subsequent training.

[0089] S205, calculating the average entropy of the base model predicting the pseudo label.

[0090] ​In an embodiment of the present application, if the pseudo-label is a first label, the ratio of the overall entropy value of the first label to the number of samples in some target domain samples can be used as the average entropy of the pseudo-label; if the pseudo-label is a second label, the ratio of the overall entropy value of the second label to the number of samples in some target domain samples can be used as the average entropy of the pseudo-label. The average entropy can reflect the degree of uncertainty in the model prediction results. If the average entropy is less than or equal to the corresponding threshold, it means that the prediction uncertainty of the basic model for the unlabeled samples in the target domain is small enough, and a good fitting state is achieved on the target domain, and the convergence condition is met at this time. On the contrary, if the average entropy is greater than the corresponding threshold, it means that the basic model has not converged to the ideal state.

[0091] S206: If the average entropy does not meet the corresponding convergence condition, the first target domain sample and the second target domain sample with the pseudo label are combined into a new first target domain sample, and the third data feature and the fourth data feature are combined into a new third data feature, and the process returns to step S204.

[0092] In an embodiment of the present application, if the average entropy is greater than the corresponding threshold, the first target domain sample and the second target domain sample with pseudo-label are combined into a new first target domain sample, the third data feature and the fourth data feature are combined into a new third data feature, and the process returns to step S204 to enter the next learning phase and relearn the data distribution of the basic model in the target domain.

[0093] S207: If the average entropy meets the convergence condition, the basic model after this learning is used as the credit assessment model.

[0094] In the embodiment of the present application, if the average entropy is less than or equal to the corresponding threshold, the learning is terminated and the basic model after this learning is used as the credit assessment model.

[0095] In one possible implementation, if the basic model cannot be fitted after multiple learning cycles, the data distribution learned once with the smallest average entropy value can be used as the final distribution. In this regard, an embodiment of the present application provides a cross-domain customer credit assessment method, wherein the credit assessment model construction process also includes the following steps:

[0096] If the average entropy does not meet the convergence conditions and the cumulative number of learning times of the basic model reaches the corresponding upper limit, the basic model after one learning with the smallest average entropy will be used as the credit assessment model.

[0097] In the embodiment of the present application, the base model records the first data distribution and the second data distribution learned this time after each learning, and records the cumulative learning times. If the average entropy after this learning is greater than the corresponding threshold, and the cumulative learning times reach the corresponding upper limit of times, the base model after the learning with the minimum average entropy is taken as the credit evaluation model, and the first data distribution and the second data distribution learned at the time of the minimum average entropy are taken as the final distribution of the information evaluation model. Since the smaller the average entropy is, the smaller the uncertainty of the prediction result of the model is, and relatively speaking, it is more stable and reliable, the data distribution with the lowest average entropy is taken as the final target domain distribution, so that the base model can still give a relatively optimal credit evaluation model in the case of unable to converge ideally, and provide a relatively reliable basis for subsequent customer credit risk assessment.

[0098] In S103, the credit risk level corresponding to the first data feature is predicted by the credit evaluation model.

[0099] In the embodiment of the present application, the first data feature is input into the credit evaluation model, and the credit risk level corresponding to the first data feature is predicted by the credit evaluation model based on the data distribution learned by the credit evaluation model.

[0100] Through the above description, the cross-domain customer credit evaluation method provided by the embodiment of the present application can make the credit evaluation model continuously fit the real data distribution of the target domain based on the learning of the data distribution of the source domain sample and the target domain sample by the naive Bayes algorithm, so that the classification uncertainty of the target domain is greatly reduced, and the credit risk level can be quickly and accurately identified during prediction.

[0101] The above introduces the cross-domain customer credit evaluation method provided by the embodiment of the present application, and the device for executing the above cross-domain customer credit evaluation method will be introduced below.

[0102] Referring to Figure 4 , Figure 4 The structure diagram of the cross-domain customer credit evaluation device provided by the embodiment of the present application is shown in FIG. 1. As shown in FIG. 1, the cross-domain customer credit evaluation device provided by the embodiment of the present application comprises: Figure 4

[0103] The model construction module 401 is configured to construct a credit evaluation model based on the naive Bayes algorithm learning the data distribution of the first source domain sample and the target domain sample.

[0104] The credit evaluation module 402 is configured to acquire target domain data, extract the first data feature corresponding to the target domain data, call the credit evaluation model, and predict the credit risk level corresponding to the first data feature by the credit evaluation model.

[0105] ​In a possible implementation, the credit assessment module 402 configured to extract the first data features corresponding to the target domain data, specifically configured to:

[0106] extract the first data features from the target domain data by a feature extractor, the feature extractor being obtained by training a deep neural network using the second source domain samples.

[0107] In a possible implementation, the target domain samples include labeled first target domain samples and unlabeled second target domain samples, and the model construction module 401 is specifically configured to:

[0108] invoke a basic model constructed based on a Naive Bayes algorithm; extract second data features corresponding to the first source domain samples, and learn a first data distribution of the second data features under the labels of the first source domain samples by the basic model; extract third data features corresponding to the first target domain samples, and learn a second data distribution of the third data features under the labels of the first target domain samples by the basic model; select part of the target domain samples from the second target domain samples for this learning, extract fourth data features of the part of the target domain samples, and predict pseudo labels corresponding to the fourth data features by the basic model using the first data distribution and the second data distribution; calculate an average entropy of the pseudo labels predicted by the basic model; if the average entropy does not satisfy a corresponding convergence condition, group the first target domain samples and the second target domain samples with the pseudo labels into new first target domain samples, group the third data features and the fourth data features into new third data features, and return to execute the step of selecting part of the target domain samples from the second target domain samples for this learning; if the average entropy satisfies the convergence condition, take the basic model after this learning as the credit assessment model.

[0109] In a possible implementation, the model construction module 401 configured to predict the pseudo labels corresponding to the fourth data features by the basic model using the first data distribution and the second data distribution, is specifically configured to:

[0110] obtain a first label and a prediction probability thereof predicted by the basic model using the first data distribution corresponding to the fourth data features, and calculate an overall entropy value of the first label according to the prediction probability; obtain a second label and a prediction probability thereof predicted by the basic model using the second data distribution corresponding to the fourth data features, and calculate an overall entropy value of the second label according to the prediction probability; compare the overall entropy values of the first label and the second label, and take the label with the minimum overall entropy value as the pseudo label.

[0111] In a possible implementation, the model construction module 401 is further configured to:

[0112] If the average entropy does not satisfy the convergence condition, and the cumulative learning number of the base model reaches the corresponding upper limit, the base model after the learning with the minimum average entropy is taken as the credit evaluation model.

[0113] It should be noted that the detailed functions of the modules in the embodiments of the present application can be referred to the corresponding disclosure part of the cross-field customer credit evaluation method embodiments described above, which will not be repeated here.

[0114] The embodiments of the present application also provide an electronic device. Referring to Figure 5 , Figure 5 A structural schematic diagram of an electronic device provided by the embodiments of the present application. The electronic device in the embodiments of the present application can include but is not limited to fixed terminals such as mobile phones, notebook computers, PDAs (personal digital assistants), PADs (tablet computers), desktop computers, etc. Figure 5 The electronic device shown is only an example and should not impose any limitation on the functions and use range of the embodiments of the present application.

[0115] As shown in Figure 5 , the electronic device can include a processing device (such as a central processor, a graphics processor, etc.) 501, which can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 502 or programs loaded from a storage device 508 into a random access memory (RAM) 503. In the state that the electronic device is powered on, various programs and data required for the operation of the electronic device are also stored in the RAM 503. The processing device 501, the ROM 502, and the RAM 503 are connected to each other through a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0116] Generally, the following devices can be connected to the I / O interface 505: input devices 506 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 507 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 508 including, for example, a memory card, a hard disk, etc.; and communication devices 509. The communication devices 509 can allow the electronic device to communicate with other devices wirelessly or by wire to exchange data. Although Figure 5 The electronic device with various devices is shown, but it should be understood that it is not required to implement or have all the devices shown. More or fewer devices can be alternatively implemented or possessed.

[0117] The embodiments of the present application also provide a computer program product including computer readable instructions, which, when running on an electronic device, enable the electronic device to implement any cross-field customer credit evaluation method provided by the embodiments of the present application.

[0118] The embodiment of the present application further provides a computer readable storage medium, the storage medium carries one or more computer programs, when the one or more computer programs are executed by an electronic device, the electronic device can realize any cross-field customer credit evaluation method provided by the embodiment of the present application.

[0119] It can be understood that, before using the technical solutions disclosed in the embodiments of the present application, the type of personal information involved in the present application, the use range, the use scene and the like should be informed to the user and the authorization of the user is obtained according to relevant laws and regulations.

[0120] It can also be understood that the data involved in the present application (including but not limited to the data itself, the acquisition or use of the data) should comply with the requirements of relevant laws and regulations and relevant provisions.

[0121] In addition, it should be noted that the apparatus embodiments described above are only schematic, and the units shown as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the present embodiment. In addition, in the apparatus embodiment provided by the present application, the connection relationship between the modules indicates that there is a communication connection between them, which can be realized as one or more communication buses or signal lines.

[0122] Through the above description of the embodiments, those skilled in the art can clearly understand that the present application can be realized by means of software and necessary general hardware, and of course can also be realized by special hardware including special integrated circuits, special CPUs, special memories, special components and the like. Generally, functions completed by computer programs can be easily realized by corresponding hardware, and specific hardware structures for realizing the same function can also be various, such as analog circuits, digital circuits or special circuits. However, for the present application, software program implementation is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer floppy disk, U disk, mobile hard disk, ROM, RAM, magnetic disk or optical disk, etc., including a plurality of instructions to make a computer device (which can be a personal computer, training device or network device, etc.) execute the methods described in the embodiments of the present application.

[0123] In the above embodiments, all or part can be realized by software, hardware, firmware or any combination thereof. When realized by software, it can be realized in the form of a computer program product in whole or in part.

[0124] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from one website, computer, training device or data center to another website, computer, training device or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be stored by the computer or a data storage device such as a training device, a data center, etc. integrated with one or more available media sets. The available media can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid state disk (SSD)), etc.

Claims

1. A cross-domain customer credit assessment method, characterized in that: The cross-domain customer credit assessment method includes: Acquire target domain data, and extract a first data feature corresponding to the target domain data; Retrieving a credit assessment model, where the credit assessment model is constructed by learning data distribution of first source domain samples and target domain samples based on a naive Bayes algorithm; The credit risk level corresponding to the first data feature is predicted using the credit assessment model.

2. The cross-domain customer credit assessment method according to claim 1, characterized in that: The extracting the first data feature corresponding to the target domain data includes: The first data feature is extracted from the target domain data by a feature extractor, wherein the feature extractor is obtained by training a deep neural network using second source domain samples.

3. The cross-domain customer credit assessment method according to claim 1 or 2, characterized in that: The target domain samples include labeled first target domain samples and unlabeled second target domain samples. The credit assessment model construction process includes: Retrieve a basic model constructed based on the naive Bayes algorithm; Extracting a second data feature corresponding to the first source domain sample, and learning a first data distribution of the second data feature under the label of the first source domain sample through the basic model; Extracting a third data feature corresponding to the first target domain sample, and learning a second data distribution of the third data feature under the label of the first target domain sample through the basic model; Selecting a portion of target domain samples for current learning from the second target domain samples, extracting a fourth data feature from the portion of target domain samples, and predicting a pseudo label corresponding to the fourth data feature using the first data distribution and the second data distribution through the basic model; Calculating the average entropy of the pseudo labels predicted by the basic model; If the average entropy does not meet the corresponding convergence condition, the first target domain sample and the second target domain sample with the pseudo label are combined into a new first target domain sample, the third data feature and the fourth data feature are combined into a new third data feature, and the process returns to the step of selecting some target domain samples for current learning from the second target domain sample. If the average entropy satisfies the convergence condition, the basic model after this learning is used as the credit assessment model.

4. The cross-domain customer credit assessment method according to claim 3, characterized in that: The predicting the pseudo label corresponding to the fourth data feature by using the first data distribution and the second data distribution through the basic model includes: Obtaining the first label and its predicted probability corresponding to the fourth data feature predicted by the basic model using the first data distribution; and calculating the overall entropy value of the first label based on the predicted probability of the first label; Obtaining the second label and its predicted probability corresponding to the fourth data feature predicted by the basic model using the second data distribution; and calculating the overall entropy value of the second label based on the predicted probability of the second label; The overall entropy value of the first label is compared with the overall entropy value of the second label, and the label with the smallest overall entropy value is used as the pseudo label.

5. The cross-domain customer credit assessment method according to claim 3, characterized in that: The credit assessment model construction process also includes: If the average entropy does not meet the convergence condition and the cumulative learning times of the basic model reaches the corresponding upper limit, the basic model after one learning with the minimum average entropy is used as the credit assessment model.

6. A cross-domain customer credit assessment device, characterized in that: The cross-domain customer credit assessment device includes: A model building module is used to build a credit assessment model by learning the data distribution of the first source domain samples and the target domain samples based on the Naive Bayes algorithm; The credit assessment module is configured to obtain target domain data and extract a first data feature corresponding to the target domain data; retrieve the credit assessment model; and predict a credit risk level corresponding to the first data feature using the credit assessment model.

7. The cross-domain customer credit assessment device according to claim 6, characterized in that: The credit assessment module for extracting the first data feature corresponding to the target domain data is specifically configured to: The first data feature is extracted from the target domain data by a feature extractor, wherein the feature extractor is obtained by training a deep neural network using second source domain samples.

8. A computer program product, characterized in that The method comprises computer-readable instructions, which, when executed on an electronic device, enable the electronic device to implement the cross-domain customer credit assessment method according to any one of claims 1 to 5.

9. An electronic device, characterized in that: comprising at least one processor and a memory connected to the processor, wherein: The memory is used to store computer programs; The processor is configured to execute the computer program so as to enable the electronic device to implement the cross-domain customer credit assessment method according to any one of claims 1 to 5.

10. A computer storage medium, characterized in that The storage medium carries one or more computer programs, which, when executed by an electronic device, enable the electronic device to implement the cross-domain customer credit assessment method according to any one of claims 1 to 5.

Images