Credit risk value assessment method and device

Abstract

The present invention provides a credit risk value assessment method and device, which can be used in the financial field, the credit risk value assessment field or other technical fields. The method comprises: obtaining the demeanor information of the target customer when handling the credit business based on the image data of the target customer collected when the target customer handles the credit business; determining the credit risk value of the target customer based on the demeanor information of the target customer when handling the credit business and the text information of the target customer's credit materials. The device is used to execute the above method. Compared with manual information review, the credit risk value assessment method and device provided by the embodiment of the present invention can reduce audit errors and avoid subjective errors of auditors, thereby improving the accuracy of credit risk assessment; and compared with traditional credit review, it can further improve the credit risk assessment effect.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a credit risk value assessment method and device. Background Art

[0002] Currently, bank customers typically apply for credit at bank counters. Traditional credit review primarily involves manual document review. However, manual review is prone to errors, and the analysis and assessment of credit documents are subject to subjective factors, resulting in inaccurate bank assessments of customer credit risk. Summary of the Invention

[0003] In response to the problems in the prior art, embodiments of the present invention provide a credit risk value assessment method and apparatus, which can at least partially solve the problems in the prior art.

[0004] In one aspect, the present invention provides a credit risk value assessment method, comprising:

[0005] obtaining, based on the image data of the target customer collected when the target customer applies for a credit service, information on the demeanor of the target customer when applying for the credit service;

[0006] The credit risk value of the target customer is determined based on the demeanor information of the target customer when handling the credit business and the text information of the target customer's credit materials.

[0007] Optionally, obtaining the target customer's demeanor information when handling the credit business based on the target customer's image data collected when the target customer handles the credit business includes:

[0008] Based on the image data of the target customer collected when the target customer applies for credit business, expression recognition, eye tracking and / or action recognition are performed on the target customer to obtain the target customer's expression information.

[0009] Optionally, determining the credit risk value of the target customer based on the demeanor of the target customer when handling the credit business and text information of the target customer's credit materials includes:

[0010] Adjusting the classification threshold of a pre-trained text risk point extraction model based on the target customer's demeanor when handling the credit business;

[0011] Inputting the text information of the target customer's credit material into the text risk point extraction model after adjusting the classification threshold, so that the text risk point extraction model outputs the risk points in the text information;

[0012] The credit risk value of the target customer is determined based on the risk points output by the text risk point extraction model.

[0013] Optionally, the expression information includes an abnormal expression, and the abnormal expression includes at least one of the following: fear, panic, wandering eyes, shaking legs, wiping sweat, and changing the position of hands;

[0014] The step of adjusting the classification threshold of the pre-trained text risk point extraction model based on the demeanor information of the target customer when handling the credit business includes:

[0015] According to the number of times the target customer shows abnormal expressions when handling the credit business, the classification threshold of the pre-trained text risk point extraction model is adjusted.

[0016] Optionally, before adjusting the classification threshold of the pre-trained text risk point extraction model based on the demeanor information of the target customer when handling the credit business, the method further includes:

[0017] The text information of each credit material in the first training set is used as input, and the risk type corresponding to each word in the text information is used as a label to train the first algorithm model to obtain the text risk point extraction model.

[0018] Optionally, the step of taking the text information of each credit document in the first training set as input and the risk type corresponding to each word in the text information as a label to train the first algorithm model to obtain the text risk point extraction model includes:

[0019] Performing vector encoding on each word in the text information of each credit document in the first training set to obtain a word vector corresponding to each word;

[0020] For each word in the text information, combine the word vectors corresponding to the first N words and the last N words adjacent to the word to generate a combined vector;

[0021] The combined vector is used as input and the risk type corresponding to the word is used as a label to train the first algorithm model to obtain the text risk point extraction model.

[0022] Optionally, determining the credit risk value of the target customer based on the risk points output by the text risk point extraction model includes:

[0023] Determining the number of risk points in the target customer's credit material based on the risk points output by the text risk point extraction model;

[0024] The number of risk points in the target customer's credit materials, the target customer's demeanor when handling the credit business, and the target customer's attribute information are input into a pre-trained risk assessment model so that the risk assessment model outputs the target customer's credit risk value.

[0025] Optionally, the method further includes:

[0026] The second algorithm model is trained using the demeanor information of each customer in the second training set when handling the credit business, the customer's attribute information, and the number of risk points of the target risk type in the customer's credit materials as input, and the customer's credit risk value as a label to obtain the risk assessment model, wherein the customer's credit risk value is determined based on the customer's credit information.

[0027] In another aspect, the present invention provides a credit risk value assessment device, comprising:

[0028] an acquisition module, configured to acquire, based on image data of a target customer collected when the target customer applies for a credit service, information on the demeanor of the target customer when the target customer applies for the credit service;

[0029] The determination module is used to determine the credit risk value of the target customer based on the demeanor information of the target customer when handling the credit business and the text information of the target customer's credit materials.

[0030] Optionally, the acquisition module is specifically configured to:

[0031] Based on the image data of the target customer collected when the target customer applies for credit business, expression recognition, eye tracking and / or action recognition are performed on the target customer to obtain the target customer's expression information.

[0032] Optionally, the determining module includes:

[0033] an adjusting unit, configured to adjust a classification threshold of a pre-trained text risk point extraction model according to the demeanor information of the target customer when handling the credit business;

[0034] An input unit, configured to input the text information of the target customer's credit material into the text risk point extraction model after adjusting the classification threshold, so that the text risk point extraction model outputs the risk points in the text information;

[0035] A determination unit is used to determine the credit risk value of the target customer based on the risk points output by the text risk point extraction model.

[0036] Optionally, the expression information includes an abnormal expression, and the abnormal expression includes at least one of the following: fear, panic, wandering eyes, shaking legs, wiping sweat, and changing the position of hands;

[0037] The adjustment unit is specifically used for:

[0038] According to the number of times the target customer shows abnormal expressions when handling the credit business, the classification threshold of the pre-trained text risk point extraction model is adjusted.

[0039] Optionally, the determining module further includes:

[0040] The first training unit is used to take the text information of each credit material in the first training set as input and the risk type corresponding to each word in the text information as a label to train the first algorithm model to obtain the text risk point extraction model.

[0041] Optionally, the first training unit is used to:

[0042] Performing vector encoding on each word in the text information of each credit document in the first training set to obtain a word vector corresponding to each word;

[0043] For each word in the text information, combine the word vectors corresponding to the first N words and the last N words adjacent to the word to generate a combined vector;

[0044] The combined vector is used as input and the risk type corresponding to the word is used as a label to train the first algorithm model to obtain the text risk point extraction model.

[0045] Optionally, the determining unit is specifically configured to:

[0046] Determining the number of risk points in the target customer's credit material based on the risk points output by the text risk point extraction model;

[0047] The number of risk points in the target customer's credit materials, the target customer's demeanor when handling the credit business, and the target customer's attribute information are input into a pre-trained risk assessment model so that the risk assessment model outputs the target customer's credit risk value.

[0048] Optionally, the determining module further includes:

[0049] The second training unit is used to take the demeanor information of each customer in the second training set when handling the credit business, the customer's attribute information, and the number of risk points of the target risk type in the customer's credit materials as input, and the customer's credit risk value as a label, to train the second algorithm model to obtain the risk assessment model, wherein the customer's credit risk value is determined based on the customer's credit information.

[0050] On the other hand, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the credit risk value assessment method described in any one of the above embodiments are implemented.

[0051] In another aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the credit risk value assessment method described in any of the above embodiments.

[0052] The credit risk assessment method and device provided in embodiments of the present invention utilize computer technology to automatically identify textual information within a target customer's credit documents and, by combining this textual information with the target customer's demeanor during the credit process, automatically determine the target customer's credit risk. This reduces audit errors and mitigates subjective errors by auditors, improving the accuracy of credit risk assessments compared to manual information review. Furthermore, compared to traditional credit reviews, this method utilizes demeanor information from customers during the initial review process, further enhancing the effectiveness of credit risk assessments. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0054] Figure 1 It is a flowchart of a credit risk value assessment method provided by one embodiment of the present invention.

[0055] Figure 2 This is a partial flow chart of a credit risk value assessment method provided by one embodiment of the present invention.

[0056] Figure 3 This is a partial flow chart of a credit risk value assessment method provided by one embodiment of the present invention.

[0057] Figure 4 This is a partial flow chart of a credit risk value assessment method provided by one embodiment of the present invention.

[0058] Figure 5 This is a partial flow chart of a credit risk value assessment method provided by one embodiment of the present invention.

[0059] Figure 6 It is a structural diagram of a credit risk value assessment device provided by an embodiment of the present invention.

[0060] Figure 7 It is a schematic diagram of the physical structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0061] To make the purpose, technical solutions and advantages of the embodiments of the present invention more clear, the embodiments of the present invention are further described in detail below with reference to the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention. It should be noted that, unless there is a conflict, the embodiments and features in the embodiments of this application can be combined with each other in any manner.

[0062] The credit risk value assessment method provided in the embodiment of the present invention is implemented by a subject including but not limited to a computer. The acquisition, storage, use, and processing of data in the technical solution of this application are in compliance with the relevant provisions of national laws and regulations.

[0063] Figure 1 FIG. 1 is a flow chart of a credit risk value assessment method provided by an embodiment of the present invention. Figure 1 As shown, the credit risk value assessment method provided by the embodiment of the present invention includes:

[0064] S101. Acquire, based on image data of a target customer collected when the target customer applies for a credit service, information on the target customer's demeanor when applying for the credit service;

[0065] In this step, when the target customer handles credit business at the branch counter (including submitting credit materials and / or waiting for preliminary review, etc.), the branch's video acquisition equipment can capture the target customer's facial expressions and movements video stream, and the facial expression analysis system can convert the target customer's video stream information into a series of recognizable image data; the specific video acquisition equipment can be the branch's surveillance camera, or a high-definition camera, etc., and the facial expression analysis system analyzes and identifies the target customer's image data to obtain the target customer's facial expression information.

[0066] After the video capture equipment at the outlet captures the video stream of the target customer's facial expressions and movements, the customer's video stream can be stored in the database first. In this way, the facial expression analysis system does not need to perform data analysis in real time during peak transaction periods. The data analysis can be delayed, and the analysis time can even be postponed to outside the outlet's business hours, reducing the real-time analysis requirements of the facial expression analysis system and allowing more complex and time-consuming data analysis models to be used in actual use.

[0067] S102: Determine the credit risk value of the target customer based on the demeanor of the target customer when handling the credit business and the text information of the target customer's credit materials.

[0068] In this step, the target customer's credit materials may include household registration booklet, marriage certificate, employment certificate, real estate certificate, lease contract, credit report, business license and / or insurance policy for policy loan, etc. After the credit personnel receive the credit materials submitted by the target customer, they can obtain the image information of the credit materials through the image acquisition equipment set up at the branch, and pre-store the image information of the credit materials in the database. The specific image acquisition equipment can be a high-definition camera or other equipment.

[0069] Credit documents can be converted from image information to text using a deep text analysis system. This system can consist of text analysis processes such as text recognition, image classification, and deep text analysis of risk points. The deep text analysis system processes the image information of credit documents as follows: text information extraction, converting the credit documents from image information to text. Specifically, this process can be divided into two steps. First, different types of credit documents are categorized using an image classification model. Mature image classification frameworks such as AlexNet and VGGNet can be used, and the image classification model for this step is obtained through data annotation and model training. Second, the text information of each type of credit document is extracted using a dedicated optical character recognition (OCR) model. These OCR models are currently available and have specialized text recognition optimization and integration methods for different types of credit documents. For example, a common ID card text recognition model uses pre-recognition of various text regions in the document to improve recognition performance through various optimization and integration methods. For example, if the date region is recognized as "I," it will be adjusted to the number "1." Using a deep text analysis system to extract credit risk points assists credit reviewers in reviewing documents, reducing their workload.

[0070] When the deep text analysis system performs deep text analysis of risk points, it can adjust the strictness of the text information review of the target customer's credit materials based on the target customer's demeanor when handling the credit business, and review the text information of the target customer's credit materials according to the adjusted review strictness to obtain the target customer's credit risk value.

[0071] The credit risk assessment method provided by an embodiment of the present invention utilizes computer technology to automatically identify textual information within a target customer's credit documents and, by combining this textual information with the target customer's demeanor during the credit process, automatically determines the target customer's credit risk value. This reduces audit errors compared to manual information review, avoids subjective errors by auditors, and improves the accuracy of credit risk assessment. Furthermore, compared to traditional credit reviews, it utilizes the demeanor of the customer during the initial review process, further enhancing the effectiveness of credit risk assessment.

[0072] Optionally, obtaining the expression information of the target customer when handling the credit business based on the image data of the target customer collected when the target customer handles the credit business includes: performing facial expression recognition, eye tracking and / or motion recognition on the target customer based on the image data of the target customer collected when the target customer handles the credit business to obtain the expression information of the target customer.

[0073] In this embodiment, the image data of the target customer collected when the target customer applies for a credit transaction can be obtained by sampling the target customer's video stream according to certain rules by the expression analysis system, such as collecting a static image every 300ms, etc., and sampling the continuous video stream into an image sequence, which serves as the basic data processing for the expression analysis system to perform expression recognition, eye tracking, and action recognition.

[0074] The steps of performing facial expression recognition, eye tracking, and / or motion recognition on the target customer to obtain the target customer's facial expression information can also be performed by the facial expression analysis system. The facial expression analysis system can be composed of three recognition models: facial expression recognition, eye tracking, and motion recognition. These recognition models can all be universal. A large number of pre-classified facial expression and motion photos can be used in advance to train the three recognition models using a neural network classification training method. These models can then be used to identify facial expressions and motions in every frame or every few frames of the target customer's video stream.

[0075] Specifically, the facial expression recognition model can categorize a target customer's facial expressions into one or more of anger, happiness, sadness, surprise, disgust, fear, and so on, continuously capturing their facial expressions. The eye tracking model captures the location of a target customer's eye gaze or their movement relative to their head, determining whether they are looking erratically or distracted. The action recognition model can capture the target customer's movements, including but not limited to leg shaking, sweat wiping, and hand repositioning. These recognition models can be used to perform data analysis within the branch, reducing the time consumed by data transmission and improving data analysis efficiency.

[0076] The credit risk assessment method provided in this embodiment fully utilizes the facial expressions and movements of target customers when handling credit business, and adds expression recognition, eye tracking, and movement recognition functions on the basis of the original credit review, thereby improving the credit risk assessment effect.

[0077] like Figure 2 As shown, optionally, determining the credit risk value of the target customer based on the demeanor information of the target customer when handling the credit business and the text information of the target customer's credit materials includes:

[0078] S1021. Adjusting the classification threshold of a pre-trained text risk point extraction model based on the target customer's demeanor when handling the credit business;

[0079] In this step, the pre-trained text risk point extraction model is used to automatically analyze the text information of the credit materials and extract the risk points in the text information of the credit materials; when the text risk point extraction model is a binary classification model, the larger the classification threshold, the more stringent the text risk point extraction model is in evaluating the risk points.

[0080] Adjusting the classification threshold of the pre-trained text risk point extraction model based on the facial expression information of the target customer when handling the credit business may include: when the target customer shows abnormal expressions such as fear, panic, wandering eyes, shaking legs, wiping sweat, changing the position of hands, etc. when handling the credit business, increasing the classification threshold of the pre-trained text risk point extraction model, so that the pre-trained text risk point extraction model can more strictly evaluate risk points.

[0081] S1022: Inputting the text information of the target customer's credit material into the text risk point extraction model after adjusting the classification threshold, so that the text risk point extraction model outputs the risk points in the text information;

[0082] In this step, the risk point extraction model is trained with the text information of the customer's credit materials as input and the labels of the risk points in the text information as output; therefore, when using the pre-trained text risk point extraction model, the text information of the target customer's credit materials is input into the text risk point extraction model after adjusting the classification threshold, and the text risk point extraction model can output the risk points in the text information.

[0083] S1023. Determine the credit risk value of the target customer based on the risk points output by the text risk point extraction model.

[0084] In this step, the credit risk value of the target customer may be determined according to the number of risk points and / or the risk type of the risk points.

[0085] Optionally, in the above embodiment, the facial expression information may include abnormal facial expressions, and the abnormal facial expressions include at least one of the following: fear, panic, wandering eyes, shaking legs, wiping sweat, and changing the position of hands; adjusting the classification threshold of the pre-trained text risk point extraction model based on the facial expression information of the target customer when handling the credit business may include: adjusting the classification threshold of the pre-trained text risk point extraction model based on the number of times the target customer has abnormal facial expressions when handling the credit business.

[0086] In this embodiment, if the customer displays the aforementioned abnormal demeanor, it indicates that the target customer's credit risk may be high. Therefore, after the demeanor analysis system analyzes the target customer's demeanor information at different times while handling the credit transaction, it can count the number of times the target customer displays the abnormal demeanor (fear, panic, wandering eyes, shaking legs, wiping sweat, changing hand positions, etc.) while handling the credit transaction. Based on the number of times the target customer displays the abnormal demeanor while handling the credit transaction, the classification threshold of the pre-trained text risk point extraction model is adjusted.

[0087] Adjusting the classification threshold of the pre-trained text risk point extraction model based on the number of times the target customer shows an abnormal expression when handling the credit business may specifically include: if the target customer does not show the abnormal expression when handling the credit business, then the classification threshold of the pre-trained text risk point extraction model is not adjusted; if the number of times the target customer shows the abnormal expression when handling the credit business is within a first numerical range, then the classification threshold of the pre-trained text risk point extraction model is increased according to a first amplification factor; if the number of times the target customer shows the abnormal expression when handling the credit business is within a second numerical range, then the classification threshold of the pre-trained text risk point extraction model is increased according to a second amplification factor.

[0088] For example, when the target customer shows abnormal demeanor 1-2 times (first numerical range), the target customer is determined to be a medium-risk customer, and the classification threshold is amplified by 1.05 as the first amplification factor (the classification threshold is multiplied by 1.05); when the target customer shows abnormal demeanor greater than or equal to 3 times (second numerical range), the target customer is determined to be a high-risk customer, and the classification threshold is amplified by 1.10 as the second amplification factor (the classification threshold is multiplied by 1.10).

[0089] like Figure 3 As shown, optionally, before adjusting the classification threshold of the pre-trained text risk point extraction model based on the demeanor information of the target customer when handling the credit business, the method further includes:

[0090] S1020. Take the text information of each credit document in the first training set as input, take the risk type corresponding to each word in the text information as a label, train the first algorithm model, and obtain the text risk point extraction model.

[0091] In this embodiment, before model training, the text information of each credit document in the first training set can be preprocessed to convert the raw document data into a format suitable for model input. This preprocessing method can employ a sequence labeling process to label each word in the text, specifically, to identify each word's risk type. For example, words unrelated to risk points can be labeled O, the first word of each risk point text in the credit document text (a risk point text can be a sentence) can be labeled B, the last word of the risk point text can be labeled E, and words between B and E can be labeled I. The text information of credit documents may include a variety of risk points, including but not limited to suspected false descriptions and ambiguous content. Different risk points can be labeled with different labels. For example, the first, middle, and last words of a suspected false description risk point (represented by the pinyin initials XJ for "false") can be labeled B_XJ, I_XJ, and E_XJ, respectively. Ambiguous risk points (represented by the pinyin initials BC for "unclear") can be labeled B_BC, I_BC, and E_BC, respectively.

[0092] After the above preprocessing of the text information of the credit documents, the text information of each credit document in the first training set is used as input, and the risk type corresponding to each word in the text information is used as a label to train the first algorithm model to obtain the text risk point extraction model. The first algorithm model can be a bidirectional LSTM, CRF, HMM, or SLE.

[0093] When there are multiple types of risk points in the text, for each type of risk point, a text risk point extraction model for that type of risk point can be specially trained, and the trained text risk point model is only used to extract that type of risk point.

[0094] like Figure 4 As shown, optionally, the step of taking the text information of each credit document in the first training set as input and the risk type corresponding to each word in the text information as a label to train the first algorithm model to obtain the text risk point extraction model may include:

[0095] S10201. Perform vector encoding on each word in the text information of each credit document in the first training set to obtain a word vector corresponding to each word;

[0096] In this step, the encoding technology can be selected from but is not limited to word2vec, BERT, Bi-LSTM, or other neural network algorithms. When using a neural network algorithm, it is necessary to jointly train it with the neural network algorithm used in the risk point extraction model. Pre-trained models such as word2vec and BERT algorithms have been pre-trained on a large data set, so that the word vectors obtained after vector encoding using word2vec and BERT algorithms can be used directly without the need for joint training.

[0097] S10202. For each word in the text information, combine the word vectors corresponding to the first N words and the last N words adjacent to the word to generate a combined vector;

[0098] In this step, the N words before the current word and the N words after the current word, a total of k words, are used. k is a hyperparameter of the model and will be optimized based on the actual training results.

[0099] The vector concatenation method can be used to combine the word vectors of the k words mentioned above (in the order in the text). The k vectors of length x will be directly combined into a combined vector of length k*x as the model input. For example, for the sentence "ABCDEFGHIJKL", when the model determines whether E is part of the text of the suspected false description risk point (B_XJ, I_XJ, E_XJ), it will be input BCDFGH. The model will determine the label of the word E based on the words before and after it.

[0100] S10203: Take the combined vector as input and the risk type corresponding to the word as a label, train the first algorithm model, and obtain the text risk point extraction model.

[0101] In this step, the text risk point extraction model takes a combined vector consisting of k word vectors as input and uses the risk type pre-annotated by the core word of this combined vector as the label. In the suspected false description risk point extraction model, the risk type label of an ambiguous risk point is treated as 0. Similarly, in the ambiguous risk point extraction model, the risk type label of a suspected false description risk point is treated as 0. Each risk point extraction model is trained specifically for that risk point type.

[0102] The data constructed in the above data format is used as the training data of the model. Cross entropy is used as the effect evaluation in model training, and a better risk point extraction model will eventually be obtained.

[0103] When using the trained risk point extraction model to extract risk points from the text information of credit materials, the text information is first vector-encoded in accordance with the above training, and then each type of risk point extraction model is called separately to extract text risk points.

[0104] like Figure 5 As shown, optionally, determining the credit risk value of the target customer based on the risk points output by the text risk point extraction model includes:

[0105] S10231. Determine the number of risk points in the target customer's credit material based on the risk points output by the text risk point extraction model;

[0106] In this step, after the text risk point extraction model outputs the risk points in the credit material, the number of risk points in the credit material can be counted.

[0107] S10232. Input the number of risk points in the target customer's credit materials, the target customer's demeanor when handling the credit business, and the target customer's attribute information into a pre-trained risk assessment model, so that the risk assessment model outputs the target customer's credit risk value.

[0108] In this step, the target customer's attribute information may include gender, age, occupation, income, and other information. The risk assessment model is trained using the number of risk points in the customer's credit profile, the customer's demeanor when applying for the credit service, and the customer's attribute information as input, and the customer's credit risk value as output. After the number of risk points in the target customer's credit profile, the target customer's demeanor when applying for the credit service, and the target customer's attribute information are input into the pre-trained risk assessment model, the risk assessment model can output the target customer's credit risk value.

[0109] Optionally, before the above-mentioned step S10232, the method may further include: taking the demeanor information of each customer in the second training set when handling the credit business, the attribute information of the customer, and the number of risk points of the target risk type in the customer's credit materials as input, and taking the customer's credit risk value as a label, training the second algorithm model to obtain the risk assessment model, wherein the customer's credit risk value can be determined based on the customer's credit information, and the higher the customer's credit value, the lower the corresponding credit risk value.

[0110] In this embodiment, the customer's demeanor analysis results (e.g., wandering eyes, bewilderedness, fear, panic), customer attributes (e.g., gender, age, occupation, income, etc., which can be extracted by a text extraction model), and the risk point extraction model results (number of risk points) can be used as features. The credit risk value determined by the historical credit compliance and default records can be used as labels. Model training is performed based on candidate algorithms: SVM, KNN, Random Forest, GBDT, xgboost, and LSTM. Root mean square error (RMSE) is used as the model tuning metric. Ultimately, to obtain a more robust prediction, the average of the credit risk values predicted by the three models with the best prediction results is selected as the final credit risk value.

[0111] Figure 6 FIG. 1 is a schematic diagram of a credit risk value assessment device according to an embodiment of the present invention. Figure 6 As shown, the credit risk value assessment device provided by the embodiment of the present invention includes:

[0112] An acquisition module 21 is configured to acquire the demeanor information of the target customer when handling the credit business based on the image data of the target customer collected when the target customer handles the credit business;

[0113] The determination module 22 is configured to determine the credit risk value of the target customer based on the demeanor of the target customer when handling the credit business and the text information of the target customer's credit materials.

[0114] The credit risk assessment device provided in an embodiment of the present invention utilizes computer technology to automatically identify textual information within a target customer's credit documents and, by combining this textual information with the target customer's demeanor during the credit process, automatically determines the target customer's credit risk value. This reduces audit errors compared to manual information review, avoids subjective errors by auditors, and improves the accuracy of credit risk assessment. Furthermore, compared to traditional credit reviews, it utilizes demeanor information from customers during the process of submitting credit documents and awaiting preliminary review, further enhancing the effectiveness of credit risk assessments.

[0115] Optionally, the acquisition module is specifically configured to:

[0116] Based on the image data of the target customer collected when the target customer applies for credit business, expression recognition, eye tracking and / or action recognition are performed on the target customer to obtain the target customer's expression information.

[0117] Optionally, the determining module includes:

[0118] an adjusting unit, configured to adjust a classification threshold of a pre-trained text risk point extraction model according to the demeanor information of the target customer when handling the credit business;

[0119] An input unit, configured to input the text information of the target customer's credit material into the text risk point extraction model after adjusting the classification threshold, so that the text risk point extraction model outputs the risk points in the text information;

[0120] A determination unit is used to determine the credit risk value of the target customer based on the risk points output by the text risk point extraction model.

[0121] Optionally, the expression information includes an abnormal expression, and the abnormal expression includes at least one of the following: fear, panic, wandering eyes, shaking legs, wiping sweat, and changing the position of hands;

[0122] The adjustment unit is specifically used for:

[0123] According to the number of times the target customer shows abnormal expressions when handling the credit business, the classification threshold of the pre-trained text risk point extraction model is adjusted.

[0124] Optionally, the determining module further includes:

[0125] The first training unit is used to take the text information of each credit material in the first training set as input and the risk type corresponding to each word in the text information as a label to train the first algorithm model to obtain the text risk point extraction model.

[0126] Optionally, the first training unit is used to:

[0127] Performing vector encoding on each word in the text information of each credit document in the first training set to obtain a word vector corresponding to each word;

[0128] For each word in the text information, combine the word vectors corresponding to the first N words and the last N words adjacent to the word to generate a combined vector;

[0129] The combined vector is used as input and the risk type corresponding to the word is used as a label to train the first algorithm model to obtain the text risk point extraction model.

[0130] Optionally, the determining unit is specifically configured to:

[0131] Determining the number of risk points in the target customer's credit material based on the risk points output by the text risk point extraction model;

[0132] The number of risk points in the target customer's credit materials, the target customer's demeanor when handling the credit business, and the target customer's attribute information are input into a pre-trained risk assessment model so that the risk assessment model outputs the target customer's credit risk value.

[0133] Optionally, the determining module further includes:

[0134] The second training unit is used to take the demeanor information of each customer in the second training set when handling the credit business, the customer's attribute information, and the number of risk points of the target risk type in the customer's credit materials as input, and the customer's credit risk value as a label, to train the second algorithm model to obtain the risk assessment model, wherein the customer's credit risk value is determined based on the customer's credit information.

[0135] The embodiment of the device provided by the embodiment of the present invention can be specifically used to execute the processing flow of the above method embodiment. Its functions are not described in detail here, and reference can be made to the detailed description of the above method embodiment.

[0136] It should be noted that the credit risk value assessment method and device provided in the embodiments of the present invention can be used in the financial field or the information security field, and can also be used in any technical field other than the financial field. The embodiments of the present invention do not limit the application field of the credit risk value assessment method and device.

[0137] Figure 7 A schematic diagram of the physical structure of an electronic device provided by an embodiment of the present invention is shown in FIG. Figure 7 As shown, the electronic device may include: a processor 701, a communications interface 702, a memory 703, and a communication bus 704, wherein the processor 701, the communications interface 702, and the memory 703 communicate with each other via the communication bus 704. The processor 701 may call the logic instructions in the memory 703 to execute the method described in any of the above embodiments, for example, including: obtaining the target customer's demeanor information when applying for a credit transaction based on image data of the target customer collected when the target customer applies for a credit transaction; and determining the target customer's credit risk value based on the target customer's demeanor information when applying for a credit transaction and the text information of the target customer's credit materials.

[0138] In addition, the logic instructions in the above-mentioned memory 703 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0139] This embodiment discloses a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the methods provided by the above-mentioned method embodiments.

[0140] This embodiment provides a computer-readable storage medium, which stores a computer program. The computer program enables the computer to execute the methods provided by the above method embodiments.

[0141] The user information in the embodiments of this application is obtained through legal and compliant channels, and the acquisition, storage, use, and processing of the user information are authorized and agreed upon by the customer.

[0142] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, 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 magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0143] The present invention is described with reference to flowcharts and / or block diagrams of methods, 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 flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, 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 data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. 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.

[0144] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0145] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0146] Throughout this specification, reference to terms such as "one embodiment," "a specific embodiment," "some embodiments," "for example," "example," "specific example," or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0147] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A credit risk value assessment method, characterized in that: include: Obtaining, based on image data of the target customer collected when the target customer applies for a credit service, information about the target customer's demeanor when applying for the credit service; wherein the demeanor information includes abnormal demeanor, and the abnormal demeanor includes at least one of the following: fear, panic, wandering eyes, shaking legs, wiping sweat, and changing the position of hands; adjusting the classification threshold of a pre-trained text risk point extraction model based on the number of times the target customer displays abnormal expressions when handling the credit business; Inputting the text information of the target customer's credit material into the text risk point extraction model after adjusting the classification threshold, so that the text risk point extraction model outputs the risk points in the text information; Determining the number of risk points in the target customer's credit material based on the risk points output by the text risk point extraction model; Inputting the number of risk points in the target customer's credit materials, the target customer's demeanor when handling the credit business, and the target customer's attribute information into a pre-trained risk assessment model, so that the risk assessment model outputs the target customer's credit risk value; The method of adjusting the classification threshold of the pre-trained text risk point extraction model according to the number of times the target customer displays abnormal expressions when handling the credit business includes: If the target customer does not show the abnormal demeanor when applying for the credit business, then the classification threshold of the pre-trained text risk point extraction model is not adjusted; If the number of times the target customer displays the abnormal demeanor when handling the credit business is within a first numerical range, increasing the classification threshold of the pre-trained text risk point extraction model according to a first amplification factor; If the number of times the target customer displays the abnormal expression when handling the credit business is within a second numerical range, the classification threshold of the pre-trained text risk point extraction model is increased according to a second amplification factor.

2. The method according to claim 1, characterized in that The acquiring of the target customer's demeanor information when handling the credit business based on the target customer's image data collected when the target customer handles the credit business includes: Based on the image data of the target customer collected when the target customer applies for credit business, expression recognition, eye tracking and / or action recognition are performed on the target customer to obtain the target customer's expression information.

3. The method according to claim 1, characterized in that Before adjusting the classification threshold of the pre-trained text risk point extraction model based on the demeanor information of the target customer when handling the credit business, the method further includes: The text information of each credit material in the first training set is used as input, and the risk type corresponding to each word in the text information is used as a label to train the first algorithm model to obtain the text risk point extraction model.

4. The method according to claim 3, characterized in that The method of taking the text information of each credit document in the first training set as input and the risk type corresponding to each word in the text information as a label to train the first algorithm model to obtain the text risk point extraction model includes: Performing vector encoding on each word in the text information of each credit document in the first training set to obtain a word vector corresponding to each word; For each word in the text information, combine the word vectors corresponding to the first N words and the last N words adjacent to the word to generate a combined vector; The combined vector is used as input and the risk type corresponding to the word is used as a label to train the first algorithm model to obtain the text risk point extraction model.

5. The method according to claim 1, wherein The method further comprises: The demeanor information of each customer in the second training set when handling the credit business, the attribute information of the customer, and the number of risk points of the target risk type in the customer's credit materials are used as input, and the credit risk value of the customer is used as a label to train the second algorithm model to obtain the risk assessment model, wherein the credit risk value of the customer is determined based on the credit information of the customer.

6. A credit risk value assessment device, characterized in that: include: an acquisition module, configured to acquire, based on image data of a target customer collected when the target customer applies for a credit service, facial expression information of the target customer when applying for the credit service; the facial expression information includes abnormal facial expressions, and the abnormal facial expressions include at least one of the following: fear, panic, wandering eyes, shaking legs, wiping sweat, and changing the position of hands; Identify modules, including: an adjusting unit, configured to adjust a classification threshold of a pre-trained text risk point extraction model according to the number of times the target customer displays abnormal expressions when handling the credit business; An input unit, configured to input the text information of the target customer's credit material into the text risk point extraction model after adjusting the classification threshold, so that the text risk point extraction model outputs the risk points in the text information; a determination unit, configured to determine the number of risk points in the target customer's credit material based on the risk points output by the text risk point extraction model; and input the number of risk points in the target customer's credit material, the target customer's demeanor when handling the credit business, and the target customer's attribute information into a pre-trained risk assessment model, so that the risk assessment model outputs a credit risk value for the target customer; Among them, the adjustment unit is specifically used to not adjust the classification threshold of the pre-trained text risk point extraction model if the target customer does not show the abnormal expression when handling the credit business; if the number of times the target customer shows the abnormal expression when handling the credit business is within a first numerical range, then the classification threshold of the pre-trained text risk point extraction model is increased according to a first amplification factor; if the number of times the target customer shows the abnormal expression when handling the credit business is within a second numerical range, then the classification threshold of the pre-trained text risk point extraction model is increased according to a second amplification factor.

7. The device according to claim 6, characterized in that The acquisition module is specifically used for: Based on the image data of the target customer collected when the target customer applies for credit business, expression recognition, eye tracking and / or action recognition are performed on the target customer to obtain the target customer's expression information.

8. The device according to claim 6, characterized in that The determining module further includes: The first training unit is used to take the text information of each credit material in the first training set as input and the risk type corresponding to each word in the text information as a label to train the first algorithm model to obtain the text risk point extraction model.

9. The device according to claim 8, characterized in that The first training unit is used for: Performing vector encoding on each word in the text information of each credit document in the first training set to obtain a word vector corresponding to each word; For each word in the text information, combine the word vectors corresponding to the first N words and the last N words adjacent to the word to generate a combined vector; The combined vector is used as input and the risk type corresponding to the word is used as a label to train the first algorithm model to obtain the text risk point extraction model.

10. The device according to claim 6, characterized in that The determining module further includes: The second training unit is used to take the demeanor information of each customer in the second training set when handling the credit business, the customer's attribute information, and the number of risk points of the target risk type in the customer's credit materials as input, and the customer's credit risk value as a label, to train the second algorithm model to obtain the risk assessment model, wherein the customer's credit risk value is determined based on the customer's credit information.

11. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images