Online credit intelligent risk control method and device, storage medium and electronic equipment

Abstract

The invention provides an online credit intelligent risk control method and device, a storage medium and electronic equipment, and is applied to the technical field of credit risk control. According to the invention, multi-modal data is integrated in credit risk assessment, image and audio features are respectively extracted by using the convolutional neural network and the long-short term memory network, and the image and audio features are fused with financial credit data to generate a comprehensive credit feature vector, so that richer information input is provided for a risk control model, and the risk control efficiency is improved. Therefore, the intelligent risk control model based on the random forest algorithm can more accurately evaluate the default probability of the customer, so that the comprehensiveness and accuracy of risk evaluation are remarkably improved, and more reliable support is provided for credit decision.

Description

Technical Field

[0001] The present invention relates to the technical field of credit risk control, and in particular to an online credit intelligent risk control method, device, storage medium and electronic equipment. Background Art

[0002] In recent years, the diversified development of the financial sector has put forward higher requirements for risk control management. The rise of digital financial services has enabled banks to launch convenient online credit loans and micro-loan products, and users can apply for loans through WeChat mini-programs, apps, etc. The risk control of online business brought about by digital transformation needs to be addressed urgently, especially in the pre-loan approval stage.

[0003] Traditional risk control mainly relies on expert experience and basic data analysis. Due to insufficient standardization and limited number of experts, it has limited effect under large-scale credit. To this end, statistical modeling such as linear discriminant analysis, linear regression and logistic regression are adopted to achieve automated approval, reduce labor costs and unify standards. However, these linear models are insufficient when dealing with nonlinear factors and feature interactions. Although nonlinear models such as decision trees and neural networks have enhanced complexity and flexibility, they still rely on structured data, resulting in bottlenecks in the development of risk control modeling.

[0004] With the digital transformation of finance, the ability to utilize multimodal data has been improved, providing a new development path for risk control models. Therefore, how to effectively integrate multimodal data to achieve a more comprehensive credit risk assessment has become a technical problem that technicians in this field urgently need to solve. Summary of the invention

[0005] In view of the above problems, the present invention provides an online credit intelligent risk control method, device, storage medium and electronic device that overcome the above problems or at least partially solve the above problems. The technical solution is as follows:

[0006] An online credit intelligent risk control method, comprising:

[0007] Obtain image data, audio data, and financial credit data collected when target customers apply for credit services online;

[0008] Using a convolutional neural network to extract features from the image data to obtain an image feature vector;

[0009] Using a long short-term memory network to extract features from the audio data to obtain an audio feature vector;

[0010] Performing feature fusion on the image feature vector, the audio feature vector and the financial credit data to obtain a credit feature vector of the target customer;

[0011] The credit feature vector is input into a pre-trained credit intelligent risk control model to obtain the default probability of the target customer output by the credit intelligent risk control model, wherein the credit intelligent risk control model is constructed based on a random forest algorithm.

[0012] Optionally, the training process of the credit intelligent risk control model includes:

[0013] Collect image sample data, audio sample data, and financial credit sample data collected from multiple sample customers when they handle credit business online;

[0014] For any of the sample customers: using the convolutional neural network to perform feature extraction on the image sample data of the sample customer to obtain an image feature sample vector, using the long short-term memory network to perform feature extraction on the audio sample data to obtain an audio feature sample vector, and performing feature fusion on the image feature sample vector, the audio feature sample vector and the financial credit sample data to obtain a credit feature sample vector of the sample customer;

[0015] According to whether the sample customer is overdue or in default on the credit business, adding a corresponding sample identifier to the credit feature sample vector of each sample customer;

[0016] Dividing the credit feature sample vectors of the plurality of sample customers to which the sample identifiers are added into a training set and a test set;

[0017] The credit intelligent risk control model is trained using the credit feature sample vectors in the training set. After each round of training, the performance of the trained credit intelligent risk control model is verified using the credit feature sample vectors in the test set to obtain performance indicator data of the credit intelligent risk control model, and determine whether the performance indicator data reaches a preset performance indicator threshold. If not, based on the gap between the performance indicator data and the preset performance indicator threshold, the hyperparameters of the credit intelligent risk control model are adjusted, and the step of training the credit intelligent risk control model using the credit feature sample vectors in the training set is returned to be executed;

[0018] When the performance indicator data reaches the preset performance indicator threshold, or the training round reaches the preset number of training iterations, the trained credit intelligent risk control model is obtained.

[0019] Optionally, the image data includes a customer selfie collected when the target customer handles a credit business online, and the using a convolutional neural network to extract features from the image data to obtain an image feature vector includes:

[0020] The image background of the customer's selfie is classified using a ResNet convolutional neural network to extract a structured image feature vector, wherein the image feature vector includes a probability distribution of the customer's selfie in different image background categories.

[0021] Optionally, the image background categories include home, outdoor, company, car and bank.

[0022] Optionally, the audio data includes environmental background sound collected when the target customer handles credit business online, and the step of extracting features from the audio data using a long short-term memory network to obtain an audio feature vector includes:

[0023] Extracting audio features of the ambient background sound using a preset audio processing technology;

[0024] The audio features are input into a long short-term memory network for audio background classification, and a structured audio feature vector is extracted, wherein the audio feature vector includes the probability distribution of the environmental background sound in different audio background categories.

[0025] Optionally, the audio background categories include noise, music and silence.

[0026] Optionally, the financial credit data includes the target customer's credit data, bank internal source data, and designated external source data.

[0027] An online credit intelligent risk control device comprises: a data acquisition unit, an image feature vector acquisition unit, an audio feature vector acquisition unit, a credit feature vector acquisition unit and a default probability acquisition unit.

[0028] The data acquisition unit is used to obtain image data, audio data and financial credit data collected by the target customer when handling credit business online;

[0029] The image feature vector obtaining unit is used to extract features from the image data using a convolutional neural network to obtain an image feature vector;

[0030] The audio feature vector obtaining unit is used to extract features from the audio data using a long short-term memory network to obtain an audio feature vector;

[0031] The credit feature vector obtaining unit is used to perform feature fusion on the image feature vector, the audio feature vector and the financial credit data to obtain the credit feature vector of the target customer;

[0032] The default probability obtaining unit is used to input the credit feature vector into a pre-trained credit intelligent risk control model to obtain the default probability of the target customer output by the credit intelligent risk control model, wherein the credit intelligent risk control model is constructed based on a random forest algorithm.

[0033] A computer-readable storage medium having a program stored thereon, wherein when the program is executed by a processor, the online credit intelligent risk control method described in any one of the above items is implemented.

[0034] An electronic device, comprising at least one processor, and at least one memory and a bus connected to the processor; wherein the processor and the memory communicate with each other via the bus; and the processor is used to call program instructions in the memory to execute any of the above-mentioned online credit intelligent risk control methods.

[0035] By means of the above technical scheme, the present invention provides an online credit intelligent risk control method, device, storage medium and electronic device, which obtains image data, audio data and financial credit data collected by the target customer when handling credit business online; uses a convolutional neural network to extract features from the image data to obtain an image feature vector; uses a long short-term memory network to extract features from the audio data to obtain an audio feature vector; performs feature fusion on the image feature vector, the audio feature vector and the financial credit data to obtain the credit feature vector of the target customer; inputs the credit feature vector into a pre-trained credit intelligent risk control model to obtain the default probability of the target customer output by the credit intelligent risk control model, wherein the credit intelligent risk control model is constructed based on a random forest algorithm. The present invention integrates multimodal data in credit risk assessment, extracts image and audio features respectively using a convolutional neural network and a long short-term memory network, and fuses them with financial credit data to generate a comprehensive credit feature vector, which provides a richer information input for the risk control model, so that the intelligent risk control model based on the random forest algorithm can more accurately assess the customer's default probability, thereby significantly improving the comprehensiveness and accuracy of risk assessment and providing more reliable support for credit decision-making.

[0036] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0038] Figure 1 A schematic diagram showing a flow chart of an implementation of an online credit intelligent risk control method provided by an embodiment of the present invention;

[0039] Figure 2 A schematic diagram showing the training process of the credit intelligent risk control model provided by an embodiment of the present invention;

[0040] Figure 3 A logic block diagram of a process for extracting audio features using Mel-cepstral coefficients provided in an embodiment of the present invention is shown;

[0041] Figure 4 A logical block diagram of a data set construction provided by an embodiment of the present invention is shown;

[0042] Figure 5 A logical block diagram of the classification process of the random forest algorithm provided by an embodiment of the present invention is shown;

[0043] Figure 6 The following is a logical block diagram of an online credit intelligent risk control system provided by an embodiment of the present invention;

[0044] Figure 7 A schematic diagram showing the structure of an online credit intelligent risk control device provided by an embodiment of the present invention is shown;

[0045] Figure 8 A schematic structural diagram of an electronic device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0046] The exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present invention and to enable the scope of the present invention to be fully communicated to those skilled in the art.

[0047] In recent years, the diversified development of the financial sector has put forward higher requirements for risk control management. With the rise of digital financial services such as cloud outlets and cloud counters, banks have launched convenient online credit loans and micro-loan products. Users can conveniently apply for loans through channels such as WeChat applets and APP clients. In the process of financial digital transformation, risk control of online business has become an urgent problem to be solved, especially the pre-loan approval of users who apply for loans online requires stricter control.

[0048] Traditionally, loan risk control relies mainly on expert experience and basic data analysis. These methods identify risks through the experience of financial professionals or use a small amount of data analysis in decision-making. However, the low degree of standardization of expert experience and the limited number of experts restrict its effectiveness under large-scale credit. To solve these problems, statistical modeling methods such as linear discriminant analysis, linear regression and logistic regression have emerged. These linear models realize automated approval, reduce labor costs and unify approval standards, and are therefore widely adopted by credit institutions.

[0049] However, these linear models are insufficient in dealing with nonlinear factors and the interaction between features. To this end, the industry has introduced nonlinear models such as decision trees and neural networks. Although these new methods have enhanced the complexity and flexibility of the models, they still rely on structured data, and the data source is single and fixed, resulting in bottlenecks in risk control modeling.

[0050] With the deepening of financial digital transformation, the ability to utilize multimodal data has been greatly improved, providing a new development path for risk control models. By integrating different types of data such as images, text, and sound, the information input of the model can be enriched, thereby conducting a more comprehensive risk assessment. This progress has brought new opportunities and challenges to the risk control field and opened up a new exploration direction for credit risk control.

[0051] In the online credit business of the banking industry, risk identification and anti-fraud usually rely on the analysis and prediction of the multi-dimensional data of loan applicants. However, the current mainstream models mainly use the credit information of the People's Bank of China, internal and external structured data as features, and have limited use of unstructured data such as images and voice obtained in the online credit process. Therefore, the present invention proposes a method to extract features from images, audio and structured data respectively, and then fuse multimodal features and input them into a prediction model to calculate the default probability of loan applicants, thereby achieving more effective risk control.

[0052] First, the present invention innovatively proposes a framework, which collects unstructured data from customers during online business processing by designing a data collector to form an unstructured data set. Then, the neural network model is used to extract the image and audio features of the customer as unstructured features of the credit scoring model. Finally, these unstructured features are combined with structured features and input into the credit scoring model to automatically predict the credit score of the borrower. It can be seen that the present invention is suitable for the online credit business of the banking industry. Through automated risk identification and intelligent early warning, the behavior and default probability of loan applicants are predicted, which significantly improves the initiative and effectiveness of risk management, reduces the risk of loan default, and improves the quality of loans.

[0053] like Figure 1 As shown, a flowchart of an implementation of an online credit intelligent risk control method provided by an embodiment of the present invention is provided. The method may include:

[0054] S100: Obtain image data, audio data, and financial credit data collected when the target customer handles credit business online.

[0055] Image data refers to visual information collected when customers apply for credit online. Image data may include ID photos, selfies, or other relevant image materials uploaded by customers when applying for credit online. Image data can be used for identity verification, fraud prevention, or analysis of customer emotions and behavioral characteristics.

[0056] Audio data refers to the sound information collected from the interaction between customers and banks. Audio data can include recordings, phone call records, or other voice communications provided by customers during voice verification. Audio data can be used for voice recognition, verifying customer identity, or assessing customers’ emotional state and integrity by analyzing voice characteristics.

[0057] Financial credit data includes traditional structured data, which involves the customer's financial history and credit record. Specifically, it may include the customer's credit score, loan record, repayment history, account balance, income, and reports from credit reporting agencies. These data are used to assess the customer's credit risk and repayment ability.

[0058] Optionally, the financial credit data includes the target customer's credit data, internal bank source data, and designated external source data.

[0059] S110. Use a convolutional neural network to extract features from the image data to obtain an image feature vector.

[0060] The image feature vector is a numerical representation extracted by processing image data, which aims to capture the important features of the image. The image feature vector is a multidimensional array, in which each element represents a certain feature of the image. This numerical representation can effectively summarize and condense the key information of the image, which is convenient for use in subsequent default probability prediction tasks.

[0061] When the embodiment of the present invention uses a convolutional neural network to extract features from image data, the convolutional neural network extracts multi-level feature information from the image data through a series of operations such as convolutional layers, activation functions, and pooling layers. Specifically, the convolutional neural network learns features including edges, textures, shapes, and more complex features at the pixel level of the image. Through layer-by-layer extraction and combination, these low-level features gradually form high-level features, which are ultimately output as image feature vectors.

[0062] Optional, in the above Figure 1 On the basis of one or more corresponding embodiments, in another optional embodiment provided by the embodiment of the present invention, the image data includes a customer selfie collected when the target customer handles a credit business online, and a convolutional neural network is used to extract features from the image data to obtain an image feature vector, which may specifically include:

[0063] The ResNet convolutional neural network is used to classify the image background of customer selfies and extract structured image feature vectors, where the image feature vectors include the probability distribution of customer selfies in different image background categories.

[0064] Among them, ResNet convolutional neural network is a deep residual network, which is famous for its outstanding performance in ImageNet image classification, detection and positioning tasks. It implements residual learning by introducing skip connections in the network, effectively solving the performance degradation problem caused by the increase of network layers.

[0065] Optionally, image background categories include home, outdoor, business, car, and bank.

[0066] The embodiment of the present invention utilizes the ability of the ResNet convolutional neural network to effectively solve the gradient vanishing problem in deep neural networks, and can extract accurate image background features from customer selfies, providing a reliable data basis for subsequent default probability prediction tasks.

[0067] S120. Use a long short-term memory network to extract features from the audio data to obtain an audio feature vector.

[0068] Among them, the Long Short-Term Memory network (LSTM) is a special recursive neural network designed specifically for analyzing time series data. Compared with traditional models such as time recursive neural networks and hidden Markov models, the LSTM network shows better performance in applications such as speech recognition. This is because the LSTM network can effectively capture and utilize long-term dependencies in sequence data, thus having significant advantages in processing complex temporal dynamic features.

[0069] The audio feature vector is a set of values ​​extracted from audio data, which aims to capture the key features of the audio signal. The audio feature vector usually includes information about the frequency, amplitude change, timbre, etc. of the audio signal. This numerical representation can effectively summarize the key information of the audio and is convenient for use in subsequent default probability prediction tasks.

[0070] When the embodiment of the present invention uses the long short-term memory network to extract features from audio data, the long short-term memory network processes the time series of the audio signal and gradually extracts the dynamic patterns and time-related features related to the sound. By using the long short-term memory network to process audio data, the embodiment of the present invention can better capture the temporal dynamic features in the audio, thereby generating accurate audio feature vectors, providing a more reliable data basis for subsequent predictions.

[0071] Optional, in the above Figure 1 On the basis of one or more corresponding embodiments, in another optional embodiment provided by the embodiment of the present invention, the audio data includes the ambient background sound collected when the target customer handles the credit business online, and the audio data is feature extracted using the long short-term memory network to obtain the audio feature vector, which may specifically include:

[0072] The audio features of the ambient background sound are extracted using a preset audio processing technology; the audio features are input into a long short-term memory network for audio background classification, and a structured audio feature vector is extracted, wherein the audio feature vector includes the probability distribution of the ambient background sound in different audio background categories.

[0073] The preset audio processing technology may be Mel-Frequency Cepstral Coefficients (MFCC).

[0074] Optionally, audio background categories include Noise, Music, and Quiet.

[0075] The embodiment of the present invention utilizes MFCC+LSTM to classify audio features, and can accurately output the probability distribution of environmental background sounds in different audio background categories, thereby providing a reliable data basis for subsequent default probability prediction tasks.

[0076] S130, performing feature fusion on the image feature vector, the audio feature vector and the financial credit data to obtain the credit feature vector of the target customer.

[0077] Specifically, the embodiments of the present invention can combine image feature vectors, audio feature vectors, and financial credit data from different sources or in different forms to form a credit feature vector with unified feature representation. Furthermore, the embodiments of the present invention use simple feature concatenation or advanced deep learning methods (such as ensemble learning or multimodal neural networks) to capture and integrate the potential relationship between each feature in the image feature vector, audio feature vector, and financial credit data to obtain a credit feature vector that more comprehensively describes the credit status of the customer, thereby facilitating a more accurate assessment of the customer's credit risk in the future.

[0078] S140. Input the credit feature vector into a pre-trained credit intelligent risk control model to obtain the default probability of the target customer output by the credit intelligent risk control model, wherein the credit intelligent risk control model is constructed based on a random forest algorithm.

[0079] Among them, the random forest algorithm is a method that integrates multiple decision trees based on the bagging idea of ​​ensemble learning. It combines the output results of multiple weak classifiers (i.e., single decision trees) and adopts a voting mechanism to form a stronger classifier. The random forest algorithm is an ensemble learning method consisting of multiple decision trees. Each tree processes a randomly sampled subset of training data and makes a prediction. The random forest aggregates the prediction results of these trees, usually by voting or averaging, to obtain the final prediction. The advantages of this method include: high accuracy, good ability to handle nonlinear feature relationships, and resistance to overfitting.

[0080] Among them, the credit intelligent risk control model is a machine learning model used to assess the credit risk of borrowers. Its purpose is to help financial institutions make wise credit decisions, reduce default risks, and optimize the loan approval process. The credit intelligent risk control model uses the input credit feature vector to predict the customer's credit behavior. In this process, the model will comprehensively consider various features and identify factors that affect credit risk.

[0081] Among them, the probability of default is a key indicator output by the credit risk control model, which is used to quantify the possibility that the customer will not be able to repay the debt on time in a certain period of time in the future. A high probability of default indicates that the customer has a high credit risk, while a low probability of default indicates that the customer is more reliable.

[0082] The present invention provides an online credit intelligent risk control method, which includes: obtaining image data, audio data and financial credit data collected by target customers when they handle credit business online; using convolutional neural networks to extract features from image data to obtain image feature vectors; using long short-term memory networks to extract features from audio data to obtain audio feature vectors; performing feature fusion on image feature vectors, audio feature vectors and financial credit data to obtain credit feature vectors of target customers; inputting the credit feature vectors into a pre-trained credit intelligent risk control model to obtain the default probability of the target customer output by the credit intelligent risk control model, wherein the credit intelligent risk control model is constructed based on a random forest algorithm. The present invention integrates multimodal data in credit risk assessment, extracts image and audio features respectively using convolutional neural networks and long short-term memory networks, and fuses them with financial credit data to generate a comprehensive credit feature vector, thereby providing a richer information input for the risk control model, so that the intelligent risk control model based on the random forest algorithm can more accurately assess the default probability of the customer, thereby significantly improving the comprehensiveness and accuracy of risk assessment, and providing more reliable support for credit decision-making.

[0083] Optional, based on Figure 1 The method shown, such as Figure 2 As shown, a schematic diagram of the training process of the credit intelligent risk control model provided by an embodiment of the present invention is shown. The training process may include:

[0084] S200: Collect image sample data, audio sample data, and financial credit sample data collected when multiple sample customers handle credit business online.

[0085] The embodiment of the present invention collects image sample data and audio sample data of multiple sample customers through a data collector when the customer handles credit business online, and processes it as an unstructured data set. In addition, the collector also obtains the customer's financial credit sample data, covering the bank's online credit business information in the past two years. The collection process includes first confirming the user's identity, and then using sensors such as the camera and microphone of the customer terminal to collect the user's image and sound. This information is classified and stored. The financial credit sample data used in the present invention is a structured data, which includes the customer's loan application and transaction records in the bank in the past two years. Financial credit sample data is divided into two categories: good samples (not overdue) and bad samples (overdue default). The three types of data are constructed and stored separately to facilitate subsequent feature extraction and fusion.

[0086] S210. For any sample customer: use a convolutional neural network to perform feature extraction on the sample customer's image sample data to obtain an image feature sample vector; use a long short-term memory network to perform feature extraction on the audio sample data to obtain an audio feature sample vector; perform feature fusion on the image feature sample vector, the audio feature sample vector and the financial credit sample data to obtain a credit feature sample vector for the sample customer.

[0087] The feature extraction of image sample data in the embodiment of the present invention is achieved through the ResNet convolutional neural network. The network consists of a convolution layer, a pooling layer, a batch normalization (BN) layer, a Dropout layer, and a fully connected layer. In the convolution calculation, the feature map of the previous layer is convolved by the convolution kernel, and nonlinearly mapped by the activation function to generate a convolution feature map. In order to avoid overfitting, the BN layer standardizes the output, and the Dropout layer randomly "discards" some neurons during the training process. The convolution feature calculation formula is:

[0088] x i+1 =f(∑ i w i x i +b)

[0089] Among them, x i+1 represents the output value of the next layer of neurons or the current node; f is the activation function; w i is the weight parameter; x i represents the input data of the current layer; b is the bias term.

[0090] The feature extraction of audio sample data in the embodiment of the present invention is completed by long short-term memory network. Figure 3 As shown in the figure, audio feature extraction uses Mel cepstral coefficients, including preprocessing, fast Fourier transform, Mel filter bank, logarithmic operation and discrete cosine transform. Mel cepstral coefficient is a feature widely used in the field of audio processing. It is a cepstral parameter extracted in the Mel scale frequency domain, which is used to effectively represent the characteristics of audio signals.

[0091] The long short-term memory network is good at processing time series data and can store relevant important information, which is particularly suitable for speech recognition tasks. Its loss function is cross entropy, which can effectively improve the training efficiency of the model. The expression of the cross entropy loss function is:

[0092]

[0093] Among them, C represents the loss value, which indicates the degree of difference between the model prediction and the true label. n represents the total number of samples; y(x i ) represents the true label of the i-th sample; a(x i) represents the probability estimate of the model output, that is, the model predicts the probability that the i-th sample belongs to label 1. The cross entropy loss function has the characteristic of dynamically adjusting the weight update speed according to the error size. When the error is large, it will quickly adjust the weight to speed up the convergence speed; when the error is small, the weight update is slower. This mechanism effectively improves the processing efficiency of the system.

[0094] After extracting the image feature vector and the audio feature vector, the embodiment of the present invention performs feature fusion on these features and the customer's financial credit sample data (such as loan application, transaction record, etc.). All feature vectors are spliced ​​together to expand the feature dimension. The fused feature vector uses the original good and bad sample labels as the input of the prediction model to evaluate the default probability of the loan applicant.

[0095] S220. Add a corresponding sample identifier to the credit feature sample vector of each sample customer based on whether the sample customer is overdue or in default in the credit business.

[0096] S230, dividing the credit feature sample vectors of multiple sample customers with added sample identifiers into a training set and a test set.

[0097] Figure 4 The figure shows the logical block diagram of the dataset construction. After the data collection and storage are completed, the data collector first adds the corresponding sample identifier to the credit feature sample vector of each sample customer based on the overdue default of the sample customer in the credit business. Next, the credit feature sample vectors of multiple sample customers with sample identifiers are divided into training sets and test sets in a ratio of 8:2. These data are then input into the deep neural network for feature extraction. At the same time, the data collector can also collect real-time image and audio data of the current loan customers and input them into the trained prediction model for classification prediction.

[0098] S240. Train the credit intelligent risk control model using the credit feature sample vectors in the training set.

[0099] The credit intelligent risk control model of the present invention provided by the embodiment of the present invention is constructed based on the random forest algorithm. The random forest is composed of multiple decision tree classifiers, which are generated by independent and identically distributed random vectors. The core idea of ​​the model is: from the credit feature sample vectors in the training set, multiple samples are extracted using the Bootstrap resampling method. A decision tree model is established for each sample to generate different classification results. Then, the optimal credit risk classification is determined by voting on these classification results. In the classification problem, the combination model of the random forest can be expressed as:

[0100]

[0101] Among them, H(X) represents the final output decision function, which classifies the input X by selecting the category Y that maximizes the number of votes for a certain item; argmax Y In other words, it selects a Y that maximizes the subsequent summation result; k represents the number of base classifiers used; I(h i (X) = Y) is the indicator function, indicating that if the base classifier h i If the predicted result on input X is Y, the value of I is 1, otherwise it is 0.

[0102] A key feature of the random forest algorithm is the Out of Bag (OOB) estimate, which is used to evaluate the classification performance of the model and fine-tune the initial parameters. In the bootstrap sampling process, if there is an initial data set containing m samples, then the probability of each sample not being selected is When m is large enough:

[0103]

[0104] This means that about 37% of the samples will not be selected, which is called "out-of-bag data". These data are not in the training set and can be used to evaluate the performance of the model, thereby partially replacing the role of the test set.

[0105] Figure 5 The figure shows the classification process of the random forest algorithm. The credit intelligent risk control model based on the random forest algorithm processes these credit feature sample vectors and outputs the assessed credit risk category, thereby realizing intelligent classification of credit risk.

[0106] S250. After each round of training cycle, the performance of the trained credit intelligent risk control model is verified using the credit feature sample vectors in the test set to obtain performance indicator data of the credit intelligent risk control model.

[0107] S260: Determine whether the performance indicator data reaches a preset performance indicator threshold; if not, execute step S270.

[0108] S270. Based on the gap between the performance indicator data and the preset performance indicator threshold, adjust the hyperparameters of the credit intelligent risk control model, and return to execute step S240.

[0109] When the performance indicator data reaches a preset performance indicator threshold, or the training round reaches a preset number of training iterations, step S280 is executed.

[0110] S280. Obtain a trained credit intelligent risk control model.

[0111] The embodiment of the present invention collects image, audio and financial credit data, uses convolutional neural network (CNN) and long short-term memory network (LSTM) to extract different types of features respectively, fuses the features of multimodal data, and fully utilizes the complementarity of various types of data to train and optimize the credit intelligent risk control model, thereby obtaining a credit intelligent risk control model that can accurately assess the customer's default probability.

[0112] like Figure 6 As shown, the logical block diagram of the online credit intelligent risk control system provided by an embodiment of the present invention includes a data set construction and processing module, a data feature extraction and fusion module, and an intelligent analysis and prediction module. The data set construction and processing module is used to collect data of multiple modes using sensors of client terminal devices to construct a multimodal data set. By fusing structured and unstructured data, the recognition ability of the risk prediction model is enhanced. The data feature extraction and fusion module is used to use the credit data of the People's Bank of China, internal data and external data directly as structured data of the input features of the model. Features are extracted from image data through the ResNet deep convolutional network. MFCC (Mel-frequency cepstral coefficients) and LSTM (long short-term memory network) are used to analyze the spectral features of audio data. The intelligent analysis and prediction module is used to combine the features extracted from different modes and input them into the random forest prediction model to output prediction results, including overdue defaults and fraudulent behaviors.

[0113] Although operations are depicted in a particular order, this should not be understood as requiring that the operations be performed in the particular order shown or in a sequential order.Multitasking and parallel processing may be advantageous under certain circumstances.

[0114] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0115] Corresponding to the above method embodiment, the embodiment of the present invention provides an online credit intelligent risk control device, whose structure is as follows: Figure 7 As shown, it may include: a data obtaining unit 10, an image feature vector obtaining unit 20, an audio feature vector obtaining unit 30, a credit feature vector obtaining unit 40 and a default probability obtaining unit 50.

[0116] The data acquisition unit 10 is used to obtain image data, audio data and financial credit data collected when the target customer handles credit business online.

[0117] The image feature vector obtaining unit 20 is used to extract features from the image data using a convolutional neural network to obtain an image feature vector.

[0118] The audio feature vector obtaining unit 30 is used to extract features from the audio data using a long short-term memory network to obtain an audio feature vector.

[0119] The credit feature vector obtaining unit 40 is used to perform feature fusion on the image feature vector, the audio feature vector and the financial credit data to obtain the credit feature vector of the target customer.

[0120] The default probability obtaining unit 50 is used to input the credit feature vector into a pre-trained credit intelligent risk control model to obtain the default probability of the target customer output by the credit intelligent risk control model, wherein the credit intelligent risk control model is constructed based on the random forest algorithm.

[0121] Optionally, the online credit intelligent risk control device may also include: a model training unit, which specifically includes: a sample data collection subunit, a sample vector acquisition subunit, a sample identifier adding subunit, a data set division subunit, a model training subunit, a performance verification subunit, a performance determination subunit, a model hyperparameter adjustment subunit and a model acquisition subunit.

[0122] The sample data collection subunit is used to collect image sample data, audio sample data and financial credit sample data collected when multiple sample customers handle credit business online.

[0123] The sample vector acquisition subunit is used for: for any sample customer: using a convolutional neural network to perform feature extraction on the sample customer's image sample data to obtain an image feature sample vector; using a long short-term memory network to perform feature extraction on the audio sample data to obtain an audio feature sample vector; performing feature fusion on the image feature sample vector, the audio feature sample vector and the financial credit sample data to obtain the credit feature sample vector of the sample customer.

[0124] The sample identification adding subunit is used to add a corresponding sample identification to the credit feature sample vector of each sample customer according to whether the sample customer is overdue or defaulted in the credit business.

[0125] The data set division subunit is used to divide the credit feature sample vectors of multiple sample customers with added sample identifiers into a training set and a test set.

[0126] The model training subunit is used to train the credit intelligent risk control model using the credit feature sample vectors in the training set.

[0127] The performance verification subunit is used to verify the performance of the trained credit intelligent risk control model using the credit feature sample vectors in the test set after each round of training cycle, and obtain the performance indicator data of the credit intelligent risk control model.

[0128] The performance determination subunit is used to determine whether the performance indicator data reaches the preset performance indicator threshold. If not, the model hyperparameter adjustment subunit is triggered. When the performance indicator data reaches the preset performance indicator threshold or the training round reaches the preset number of training iterations, the model acquisition subunit is triggered.

[0129] The model hyperparameter adjustment subunit is used to adjust the hyperparameters of the credit intelligent risk control model based on the gap between the performance indicator data and the preset performance indicator threshold, thereby triggering the model training subunit.

[0130] The model acquisition subunit is used to obtain the trained credit intelligent risk control model.

[0131] Optionally, the image feature vector obtaining unit 20 can be specifically used to use a ResNet convolutional neural network to classify the image background of the customer's selfie, and extract a structured image feature vector, wherein the image feature vector includes a probability distribution of the customer's selfie in different image background categories.

[0132] Optionally, image background categories include home, outdoor, business, car, and bank.

[0133] Optionally, the audio feature vector obtaining unit 30 can be specifically used to extract audio features of environmental background sounds using a preset audio processing technology; input the audio features into a long short-term memory network for audio background classification, and extract a structured audio feature vector, wherein the audio feature vector includes the probability distribution of the environmental background sounds in different audio background categories.

[0134] Optionally, audio background categories include Noise, Music, and Quiet.

[0135] Optionally, the financial credit data includes the target customer's credit data, internal bank source data, and designated external source data.

[0136] The present invention provides an online credit intelligent risk control device, which is used to: obtain image data, audio data and financial credit data collected by target customers when they handle credit business online; use convolutional neural networks to extract features from image data to obtain image feature vectors; use long short-term memory networks to extract features from audio data to obtain audio feature vectors; perform feature fusion on image feature vectors, audio feature vectors and financial credit data to obtain credit feature vectors of target customers; input the credit feature vectors into a pre-trained credit intelligent risk control model to obtain the default probability of the target customer output by the credit intelligent risk control model, wherein the credit intelligent risk control model is constructed based on a random forest algorithm. The present invention integrates multimodal data in credit risk assessment, extracts image and audio features respectively using convolutional neural networks and long short-term memory networks, and fuses them with financial credit data to generate a comprehensive credit feature vector, thereby providing a richer information input for the risk control model, so that the intelligent risk control model based on the random forest algorithm can more accurately assess the default probability of the customer, thereby significantly improving the comprehensiveness and accuracy of risk assessment, and providing more reliable support for credit decision-making.

[0137] Regarding the device in the above embodiment, the specific manner in which each unit performs the operation has been described in detail in the embodiment of the method, and will not be elaborated here.

[0138] The online credit intelligent risk control device includes a processor and a memory. The above-mentioned data acquisition unit 10, image feature vector acquisition unit 20, audio feature vector acquisition unit 30, credit feature vector acquisition unit 40 and default probability acquisition unit 50 are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to realize corresponding functions.

[0139] The processor contains a kernel, which calls the corresponding program unit from the memory. One or more kernels can be set, and the kernel parameters can be adjusted to integrate multimodal data in credit risk assessment, use convolutional neural networks and long short-term memory networks to extract image and audio features respectively, and fuse them with financial credit data to generate a comprehensive credit feature vector, which provides richer information input for the risk control model, so that the intelligent risk control model based on the random forest algorithm can more accurately assess the customer's default probability, thereby significantly improving the comprehensiveness and accuracy of risk assessment and providing more reliable support for credit decision-making.

[0140] An embodiment of the present invention provides a computer-readable storage medium having a program stored thereon, which implements the online credit intelligent risk control method when executed by a processor.

[0141] An embodiment of the present invention provides a processor, which is used to run a program, wherein the online credit intelligent risk control method is executed when the program is running.

[0142] like Figure 8 As shown, an embodiment of the present invention provides an electronic device 1000, which includes at least one processor 1001, at least one memory 1002 connected to the processor 1001, and a bus 1003; wherein the processor 1001 and the memory 1002 communicate with each other through the bus 1003; the processor 1001 is used to call the program instructions in the memory 1002 to execute the above-mentioned online credit intelligent risk control method. The electronic device in this article can be a server, a PC, a PAD, a mobile phone, etc.

[0143] The present invention also provides a computer program product, which, when executed on an electronic device, is suitable for executing the program steps of initializing the online credit intelligent risk control method.

[0144] The present invention is described with reference to flowcharts and / or block diagrams of methods, apparatuses, electronic devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable device to generate a machine, so that the instructions executed by the processor of the computer or other programmable device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0145] In a typical configuration, an electronic device includes one or more processors (CPU), a memory, and a bus. The electronic device may also include an input / output interface, a network interface, and the like.

[0146] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip. The memory is an example of a computer-readable medium.

[0147] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0148] 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 the present invention are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0149] It is understandable that before using the technical solutions disclosed in the embodiments of the present disclosure, the types, scope of use, usage scenarios, etc. of the personal information involved in the present disclosure should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0150] In the description of the present invention, it should be understood that the terms "up", "down", "front", "back", "left" and "right" etc. indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description. They do not indicate or imply that the positions or elements referred to must have specific directions, be constructed and operate in specific directions. Therefore, they should not be understood as limitations of the present invention.

[0151] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. It should also be noted that the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity 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, commodity or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0152] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0153] The above are only embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.

Claims

1. An online credit intelligent risk control method, characterized in that: include: Obtain image data, audio data, and financial credit data collected when target customers apply for credit services online; Using a convolutional neural network to extract features from the image data to obtain an image feature vector; Using a long short-term memory network to extract features from the audio data to obtain an audio feature vector; Performing feature fusion on the image feature vector, the audio feature vector and the financial credit data to obtain a credit feature vector of the target customer; The credit feature vector is input into a pre-trained credit intelligent risk control model to obtain the default probability of the target customer output by the credit intelligent risk control model, wherein the credit intelligent risk control model is constructed based on a random forest algorithm.

2. The method according to claim 1, characterized in that The training process of the credit intelligent risk control model includes: Collect image sample data, audio sample data, and financial credit sample data collected from multiple sample customers when they handle credit business online; For any of the sample customers: using the convolutional neural network to perform feature extraction on the image sample data of the sample customer to obtain an image feature sample vector, using the long short-term memory network to perform feature extraction on the audio sample data to obtain an audio feature sample vector, and performing feature fusion on the image feature sample vector, the audio feature sample vector and the financial credit sample data to obtain a credit feature sample vector of the sample customer; According to whether the sample customer is overdue or in default on the credit business, adding a corresponding sample identifier to the credit feature sample vector of each sample customer; Dividing the credit feature sample vectors of the plurality of sample customers to which the sample identifiers are added into a training set and a test set; The credit intelligent risk control model is trained using the credit feature sample vectors in the training set. After each round of training, the performance of the trained credit intelligent risk control model is verified using the credit feature sample vectors in the test set to obtain performance indicator data of the credit intelligent risk control model, and determine whether the performance indicator data reaches a preset performance indicator threshold. If not, based on the gap between the performance indicator data and the preset performance indicator threshold, the hyperparameters of the credit intelligent risk control model are adjusted, and the step of training the credit intelligent risk control model using the credit feature sample vectors in the training set is returned to be executed; When the performance indicator data reaches the preset performance indicator threshold, or the training round reaches the preset number of training iterations, the trained credit intelligent risk control model is obtained.

3. The method according to claim 1, characterized in that The image data includes a selfie of the target customer collected when the target customer handles a credit business online, and the use of a convolutional neural network to extract features from the image data to obtain an image feature vector includes: The image background of the customer's selfie is classified using a ResNet convolutional neural network to extract a structured image feature vector, wherein the image feature vector includes a probability distribution of the customer's selfie in different image background categories.

4. The method according to claim 3, characterized in that The image background categories include home, outdoor, company, car, and bank.

5. The method according to claim 1, characterized in that The audio data includes the ambient background sound collected when the target customer handles a credit business online, and the long short-term memory network is used to extract features from the audio data to obtain an audio feature vector, including: Extracting audio features of the ambient background sound using a preset audio processing technology; The audio features are input into a long short-term memory network for audio background classification, and a structured audio feature vector is extracted, wherein the audio feature vector includes the probability distribution of the environmental background sound in different audio background categories.

6. The method according to claim 5, characterized in that The audio background categories include noise, music, and silence.

7. The method according to any one of claims 1 to 6, characterized in that The financial credit data includes the target customer's credit data, bank internal source data, and designated external source data.

8. An online credit intelligent risk control device, characterized in that: include: a data acquisition unit, an image feature vector acquisition unit, an audio feature vector acquisition unit, a credit feature vector acquisition unit and a default probability acquisition unit, The data acquisition unit is used to obtain image data, audio data and financial credit data collected by the target customer when handling credit business online; The image feature vector obtaining unit is used to extract features from the image data using a convolutional neural network to obtain an image feature vector; The audio feature vector obtaining unit is used to extract features from the audio data using a long short-term memory network to obtain an audio feature vector; The credit feature vector obtaining unit is used to perform feature fusion on the image feature vector, the audio feature vector and the financial credit data to obtain the credit feature vector of the target customer; The default probability obtaining unit is used to input the credit feature vector into a pre-trained credit intelligent risk control model to obtain the default probability of the target customer output by the credit intelligent risk control model, wherein the credit intelligent risk control model is constructed based on a random forest algorithm.

9. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by the processor, the online credit intelligent risk control method as described in any one of claims 1 to 7 is implemented.

10. An electronic device, comprising at least one processor, and at least one memory and a bus connected to the processor; wherein: The processor and the memory communicate with each other via the bus; the processor is used to call program instructions in the memory to execute the online credit intelligent risk control method as described in any one of claims 1 to 7.

Images