Visual risk control assessment method for lender's social attributes based on image recognition

Abstract

The present application relates to a visual risk control assessment method for the social attributes of lenders based on image recognition. The method comprises: obtaining visual data of the lender during remote interaction with the lender; wherein the visual data comprises at least one of image data and video data; inputting the visual data into a pre-trained assessment model to extract the borrower's character portrait features and surrounding environment features from the visual data through the assessment model; and assessing the lender's social attributes based on the character portrait features and surrounding environment features through the assessment model; social attributes include social background and living conditions. The solution provided by the present application can solve the problems of insufficient data, privacy leakage, low assessment efficiency, and low assessment accuracy in related technologies.

Description

Technical Field

[0001] The present application relates to the field of remote risk control technology, and in particular to a visual risk control assessment method for lenders' social attributes based on image recognition. Background Art

[0002] In the process of loan service, it is necessary to evaluate the social attributes of borrowers, including their age, gender, occupation, family situation, etc., for use in loan approval, credit scoring, risk assessment and other fields to reduce the risk of loan business. Especially when the traditional credit scoring system cannot cover or fully reflect the credit status of certain groups of people, evaluating the social attributes of borrowers can be used as an auxiliary tool to provide more useful information as a basis for credit assessment.

[0003] In related technologies, loan processing usually requires the borrower to be present in person, and the staff will evaluate the borrower's social attributes online. However, the manual evaluation method is not only inefficient and involves privacy leakage, but also has the influence of human subjective judgment, resulting in low evaluation accuracy. In addition, the analysis method of borrowers in related technologies is mainly based on structured data (such as name, age, etc.), lacking the analysis and judgment of visual risk control of unstructured data (such as images, videos, etc.), and insufficient data will also affect the accuracy of the evaluation. Summary of the invention

[0004] In order to solve or partially solve the problems existing in the related art, the present application provides a visual risk control assessment method for the social attributes of lenders based on image recognition, which can solve the problems of insufficient data, privacy leakage, low assessment efficiency, and low assessment accuracy in the related art.

[0005] Ensure effective analysis while protecting user privacy, and improve evaluation efficiency and accuracy.

[0006] In a first aspect, the present application provides a visual risk control assessment method for social attributes of lenders based on image recognition, the method comprising: During the remote interaction with the lender, obtaining visual data of the lender; wherein the visual data includes at least one of image data and video data; Inputting the visual data into a pre-trained evaluation model to extract the borrower's character portrait features and surrounding environment features from the visual data through the evaluation model; The social attributes of the lender are evaluated through the evaluation model according to the character portrait features and the surrounding environment features; the social attributes include social background and living conditions.

[0007] In one embodiment, the step of evaluating the social attributes of the lender according to the character portrait features and the surrounding environment features by the evaluation model includes: The social background of the borrower is evaluated according to the characteristics of the character portrait by the evaluation model, and the living conditions of the borrower are evaluated according to the characteristics of the surrounding environment by the evaluation model; Outputting the social background and the living conditions as the social attributes of the lender through the evaluation model, and outputting the social attribute score of the lender; The method further comprises: Based on the social attribute score, assessing the risk level of the lender; The social attributes and risk level of the lender are displayed visually.

[0008] In one embodiment, the character portrait features include at least two of hair style, tattoo area, clothing style, and jewelry style; and the step of evaluating the social background of the lender based on the character portrait features by the evaluation model includes: Determine whether each of the character portrait features meets the corresponding preset portrait conditions through the evaluation model; If at least two of the character portrait characteristics meet the corresponding preset portrait conditions, the lender is assessed as a social person with potential malicious lending; The step of judging whether each of the character portrait features meets the corresponding preset portrait conditions by the evaluation model includes: Determining whether the hair style matches a preset hair style, determining whether the tattoo area exceeds a preset tattoo area, determining whether the dress style matches a preset dress style, and determining whether the jewelry style matches a preset jewelry style by using the evaluation model; The preset hair style, the preset tattoo area, the preset clothing style and the preset accessory style are all set based on the typical image characteristics of the social personnel with potential malicious loans.

[0009] In one embodiment, the surrounding environment characteristics include at least one of a living environment and a working environment; and the evaluating the living conditions of the borrower according to the surrounding environment characteristics by the evaluation model includes: Determine whether each of the surrounding environment characteristics meets the corresponding preset environmental conditions through the evaluation model; If at least one of the surrounding environmental characteristics meets the corresponding preset environmental conditions, the borrower is assessed as a social person with low life stability; The step of judging whether each of the surrounding environment characteristics meets the corresponding preset environmental conditions by the evaluation model includes: The evaluation model is used to determine whether the living environment is lower than a first preset standard, and to determine whether the working environment is lower than a second preset standard.

[0010] In one embodiment, the step of evaluating the risk level of the lender based on the social attribute score includes: Determining whether the social attribute score is greater than or equal to a preset risk threshold; If the social attribute score is greater than or equal to the preset risk threshold, the risk level of the borrower is assessed to be high risk; the high risk is used to indicate that the borrower has a high risk of loan default.

[0011] In one embodiment, the evaluation model is trained in the following manner: Acquire multiple sample images; wherein each of the sample images contains a corresponding lender and a true value label, and the true value label includes at least one of a first label and a second label, the first label is used to mark whether the lender in the sample image is a social person with potential malicious loans, and the second label is used to mark whether the lender in the sample image is a social person with low life stability; Obtain a deep learning model as an evaluation model to be trained; The evaluation model is trained using a plurality of the sample images to obtain recognition results of the evaluation model for each of the sample images; the recognition results are used to characterize whether the lender in the sample image is a social person with potential malicious lending and / or a social person with low life stability; Matching the recognition results belonging to the same sample image with the true value label to obtain matching results for each sample image; Calculating an evaluation index of the evaluation model according to the matching results of each of the sample images; the evaluation index includes at least one of accuracy and recall; Determine whether the evaluation index is greater than or equal to a preset index; If the evaluation index is greater than or equal to the preset index, it is determined that the training of the evaluation model is completed.

[0012] In one embodiment, before inputting the visual data into a pre-trained evaluation model, the method further comprises: Perform a preprocessing operation on the visual data; wherein the preprocessing operation includes at least one of image denoising and image enhancement.

[0013] The second aspect of the present application provides a visual risk control assessment system for social attributes of lenders based on image recognition, the system comprising: A visual data acquisition module, used to acquire visual data of the lender during the remote interaction with the lender; wherein the visual data includes at least one of image data and video data; A feature extraction module, used for inputting the visual data into a pre-trained evaluation model, so as to extract the borrower's character portrait features and surrounding environment features from the visual data through the evaluation model; The social attribute evaluation module is used to evaluate the social attributes of the lender according to the character portrait features and the surrounding environment features through the evaluation model; the social attributes include social background and living conditions.

[0014] A third aspect of the present application provides an electronic device, including: Processor; and The memory stores executable codes thereon, and when the executable codes are executed by the processor, the processor is caused to execute the method as described above.

[0015] A fourth aspect of the present application provides a computer-readable storage medium having executable code stored thereon. When the executable code is executed by a processor of an electronic device, the processor is caused to execute the method as described above.

[0016] The technical solution provided by this application may have the following beneficial effects: The solution provided by the present application obtains the visual data of the lender during the remote interaction with the lender; wherein the visual data includes at least one of image data and video data; the visual data is input into a pre-trained evaluation model to extract the borrower's character portrait features and surrounding environment features from the visual data through the evaluation model; the social attributes of the lender are evaluated by the evaluation model based on the character portrait features and surrounding environment features; the social attributes include social background and living conditions. The present application automates the evaluation of the social attributes of the lender through the evaluation model without manual intervention, thereby greatly improving the evaluation efficiency, and the privacy of the lender will not be disclosed to the staff, thereby ensuring that the privacy of the lender can be effectively analyzed while protecting it, and also reducing the impact of human subjective judgment and reducing the risk caused by human bias, thereby improving the accuracy of the evaluation. Furthermore, the present application introduces image recognition technology to realize the analysis and judgment of visual risk control of unstructured data (such as visual data such as image data and video data), which can supplement the missing dimensions in traditional credit evaluation (such as the social background and living conditions of the lender), thereby realizing multi-angle evaluation and further improving the accuracy of the evaluation.

[0017] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The above and other objects, features and advantages of the present application will become more apparent by describing in more detail the exemplary embodiments of the present application in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same components in the exemplary embodiments of the present application.

[0019] Figure 1 It is a flow chart of a visual risk control assessment method for social attributes of lenders based on image recognition according to an embodiment of the present application; Figure 2 It is another flow chart of the visual risk control assessment method of the lender's social attributes based on image recognition shown in the embodiment of the present application; Figure 3 It is a structural schematic diagram of a lender social attribute visual risk control assessment system based on image recognition shown in an embodiment of the present application; Figure 4 It is a schematic diagram of the structure of an electronic device shown in an embodiment of the present application. DETAILED DESCRIPTION

[0020] The embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to make the present application more thorough and complete, and to fully convey the scope of the present application to those skilled in the art.

[0021] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0022] It should be understood that although the terms "first", "second", "third", etc. may be used in this application to describe various information, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of this application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0023] In related technologies, loan processing usually requires the borrower to be present in person, and the staff will evaluate the borrower's social attributes online. However, the manual evaluation method is not only inefficient and involves privacy leakage, but also has the influence of human subjective judgment, resulting in low evaluation accuracy. In addition, the analysis method of borrowers in related technologies is mainly based on structured data (such as name, age, etc.), lacking the analysis and judgment of visual risk control of unstructured data (such as images, videos, etc.), and insufficient data will also affect the accuracy of the evaluation.

[0024] In response to the above problems, the embodiment of the present application provides a visual risk control assessment method for the social attributes of lenders based on image recognition. The assessment model is used to automatically complete the assessment of the social attributes of lenders without manual intervention, thereby greatly improving the assessment efficiency, and the privacy of lenders will not be disclosed to staff, thereby ensuring that the privacy of lenders can be effectively analyzed while protecting them. In addition, it can also reduce the impact of human subjective judgment and reduce the risks caused by human bias, thereby improving the accuracy of the assessment. Furthermore, the present application introduces image recognition technology to realize the analysis and judgment of visual risk control of unstructured data (such as visual data such as image data and video data), which can supplement the missing dimensions in traditional credit assessment (such as the social background and living conditions of the lender), thereby realizing multi-angle assessment and further improving the accuracy of the assessment.

[0025] The technical solution of the embodiments of the present application is described in detail below with reference to the accompanying drawings.

[0026] Figure 1 It is a flow chart of a visual risk control assessment method for lender social attributes based on image recognition shown in an embodiment of the present application.

[0027] See also Figure 1 The visual risk control assessment method of lender social attributes based on image recognition in this application includes: S110, obtaining visual data of the lender during remote interaction with the lender; wherein the visual data includes at least one of image data and video data.

[0028] In the embodiment of the present application, it can be applied to the visual risk control assessment system of the social attributes of lenders based on image recognition (hereinafter referred to as the "assessment system"). When a user needs a video interview or loan application, the user acts as a lender. The lender can initiate a remote interaction request to the assessment system through a terminal device (such as a mobile phone, computer, etc.). The assessment system responds to the remote interaction request and remotely interacts with the lender. The assessment system can perform visual risk control on the lender in the current remote interaction scenario.

[0029] In one example, after receiving a remote interaction request, the evaluation system may send a collection instruction to the terminal device currently used by the lender. The collection instruction is used to instruct the camera of the terminal device to collect the lender's visual data in real time. Therefore, during the remote interaction process with the lender, the evaluation system can obtain the lender's visual data in real time.

[0030] In another example, the remote interaction request may carry visual data uploaded by the lender, so during the remote interaction process with the lender, the evaluation system can obtain the visual data of the lender from the remote interaction request.

[0031] The visual data may include at least one of static image data and dynamic video data, and both the image data and the video data at least include a front angle of the lender. For example, the image data may be a front photo of the lender.

[0032] S120, inputting the visual data into a pre-trained evaluation model, so as to extract the borrower's character portrait features and surrounding environment features from the visual data through the evaluation model.

[0033] The embodiment of the present application can pre-train the evaluation model, and the evaluation model can be a deep learning model, for example, the evaluation model can be any deep learning model of CLIP (Contrastive Language-Image Pre-training), ResNet (Residual Network), CNN (Convolutional Neural Networks). The embodiment of the present application can deploy the trained evaluation model to the evaluation system for online evaluation of the evaluation system.

[0034] In actual applications, the evaluation system can call a pre-trained evaluation model, and then input the borrower's visual data into the evaluation model, so as to extract the borrower's key features from the visual data through the evaluation model. The key features may include character portrait features and surrounding environment features, wherein the character portrait features may include at least two of the hair style, tattoo area, clothing style, and jewelry style, and the surrounding environment features may include at least one of the living environment and the working environment. In addition, character portrait features may also include face shape, eyes, nose, mouth, wrinkles, etc.

[0035] S130, evaluating the social attributes of the borrower according to the character portrait characteristics and the surrounding environment characteristics through the evaluation model; the social attributes include social background and living conditions.

[0036] In one example, the character portrait feature can reflect the image feature of the borrower, so the evaluation model can accurately evaluate the social background of the borrower based on the image feature of the borrower reflected by the character portrait feature. In addition, the evaluation model can also accurately evaluate the occupation of the borrower based on the image feature of the borrower reflected by the character portrait feature. In addition, the character portrait feature can also reflect the facial features of the borrower, so the evaluation model can accurately evaluate the age and gender of the borrower based on the facial features of the borrower reflected by the character portrait feature.

[0037] In another example, the surrounding environment characteristics can reflect the borrower's socioeconomic status, so the assessment model can accurately assess the borrower's living conditions based on the borrower's socioeconomic status reflected by the surrounding environment characteristics. In addition, the assessment model can also accurately assess the borrower's family situation based on the borrower's socioeconomic status reflected by the surrounding environment characteristics.

[0038] The evaluation model can output the borrower's social background and living conditions as the borrower's social attributes, and can also output the borrower's age, gender, occupation, and family situation as the borrower's social attributes.

[0039] As can be seen from this example, the solution provided by this application obtains the visual data of the lender during the remote interaction with the lender; wherein the visual data includes at least one of image data and video data; the visual data is input into a pre-trained evaluation model to extract the borrower's character portrait features and surrounding environment features from the visual data through the evaluation model; the social attributes of the lender are evaluated by the evaluation model based on the character portrait features and surrounding environment features; the social attributes include social background and living conditions. This application automates the evaluation of the social attributes of the lender through the evaluation model without manual intervention, thereby greatly improving the evaluation efficiency, and the privacy of the lender will not be disclosed to the staff, thereby ensuring that the privacy of the lender can be effectively analyzed while protecting it, and also reducing the impact of human subjective judgment and reducing the risks caused by human bias, thereby improving the accuracy of the evaluation. Furthermore, this application introduces image recognition technology to realize the analysis and judgment of visual risk control of unstructured data (such as visual data such as image data and video data), which can supplement the missing dimensions in traditional credit evaluation (such as the social background and living conditions of the lender), thereby realizing multi-angle evaluation and further improving the accuracy of the evaluation.

[0040] Figure 2 It is another flow chart of the visual risk control assessment method of the lender's social attributes based on image recognition shown in this application.

[0041] See also Figure 2 The visual risk control assessment method of lender social attributes based on image recognition in this application includes: S210, obtaining visual data of the lender during remote interaction with the lender; wherein the visual data includes at least one of image data and video data.

[0042] When a user needs a video interview or loan application, the user acts as a lender. The lender can initiate a remote interaction request to the assessment system through a terminal device. The assessment system responds to the remote interaction request and interacts remotely with the lender. The assessment system can perform visual risk control on the lender in the current remote interaction scenario.

[0043] In one example, after receiving a remote interaction request, the evaluation system may send a collection instruction to the terminal device currently used by the lender. The collection instruction is used to instruct the camera of the terminal device to collect the lender's visual data in real time. Therefore, during the remote interaction process with the lender, the evaluation system can obtain the lender's visual data in real time.

[0044] In another example, the remote interaction request may carry visual data uploaded by the lender, so during the remote interaction process with the lender, the evaluation system can obtain the visual data of the lender from the remote interaction request.

[0045] The visual data may include at least one of static image data and dynamic video data, and both the image data and the video data at least include a front angle of the lender. For example, the image data may be a front photo of the lender.

[0046] S220, inputting the visual data into a pre-trained evaluation model to extract the borrower's character portrait features and surrounding environment features from the visual data through the evaluation model.

[0047] The embodiment of the present application can pre-train the evaluation model, and the evaluation model can be any deep learning model among CLIP, ResNet, and CNN. The embodiment of the present application can deploy the trained evaluation model into the evaluation system for online evaluation by the evaluation system.

[0048] In actual applications, the evaluation system can call a pre-trained evaluation model, and then input the borrower's visual data into the evaluation model, so as to extract the borrower's key features from the visual data through the evaluation model. The key features may include character portrait features and surrounding environment features, wherein the character portrait features may include at least two of the hair style, tattoo area, clothing style, and jewelry style, and the surrounding environment features may include at least one of the living environment and the working environment. In addition, character portrait features may also include face shape, eyes, nose, mouth, wrinkles, etc.

[0049] In one embodiment, before inputting the visual data into the pre-trained evaluation model, the method may further include: Perform a preprocessing operation on the visual data; wherein the preprocessing operation includes at least one of image denoising and image enhancement.

[0050] Before inputting the borrower's visual data into the evaluation model, the evaluation system may preprocess the visual data to improve the recognition accuracy of the subsequent evaluation model for the visual data. The preprocessing operation may include at least one of image denoising and image enhancement. Image denoising can reduce noise in the visual data and improve the quality of the visual data. Image enhancement can improve the visibility of the visual data, so that the subsequent evaluation model can better understand the information in the visual data.

[0051] S230, evaluating the social background of the borrower according to the character portrait characteristics through the evaluation model, and evaluating the living conditions of the borrower according to the surrounding environment characteristics through the evaluation model.

[0052] In one example, the character portrait features may include hair style, tattoo area, clothing style, and jewelry style, etc. These character portrait features can reflect the image features of the lender. Therefore, the evaluation model can accurately evaluate the social background of the lender based on the image features of the lender reflected by the character portrait features. The social background is used to characterize whether the lender is a social person with potential malicious loans. In addition, the evaluation model can also accurately evaluate the borrower's occupation based on the image features of the lender reflected by the character portrait features. In addition, the character portrait features may include face shape, eyes, nose, mouth, wrinkles, etc. These character portrait features can reflect the facial features of the lender. Therefore, the evaluation model can accurately evaluate the age and gender of the lender based on the facial features of the lender reflected by the character portrait features.

[0053] In another example, the surrounding environment characteristics may include living environment and working environment, etc. These surrounding environment characteristics may reflect the borrower's socioeconomic status. Therefore, the evaluation model may accurately evaluate the borrower's living conditions based on the borrower's socioeconomic status reflected by the surrounding environment characteristics. The living conditions are used to characterize whether the borrower is a social person with low life stability. In addition, the evaluation model may also accurately evaluate the borrower's family situation based on the borrower's socioeconomic status reflected by the surrounding environment characteristics.

[0054] In one embodiment, the character portrait features include at least two of hair style, tattoo area, clothing style, and jewelry style; and evaluating the social background of the lender based on the character portrait features through the evaluation model may include: The evaluation model is used to determine whether each character profile feature meets the corresponding preset character profile conditions; if at least two character profile features meet the corresponding preset character profile conditions, the lender is assessed to be a member of society with the potential for malicious lending.

[0055] In one embodiment, judging whether each character portrait feature meets the corresponding preset portrait condition through the evaluation model may include: The evaluation model is used to determine whether the hair style matches the preset hair style, whether the tattoo area exceeds the preset tattoo area, whether the dressing style matches the preset dressing style, and whether the jewelry style matches the preset jewelry style; among which, the preset hair style, preset tattoo area, preset dressing style and preset jewelry style are all set based on the typical image characteristics of social personnel with potential malicious loans.

[0056] The embodiments of the present application can pre-set corresponding preset hair styles, preset tattoo areas, preset clothing styles, and preset jewelry styles based on the typical image features of social personnel with potential malicious loans, so that the subsequent evaluation model can determine whether the borrower's hair style matches the preset hair style, whether the borrower's tattoo area exceeds the preset tattoo area, whether the borrower's clothing style matches the preset clothing style, and whether the borrower's jewelry style matches the preset jewelry style.

[0057] In one example, the preset hair style may be, for example, a bald head, a Shamatte hairstyle, etc. If the borrower's hair style is a bald head or a Shamatte hairstyle, it is identified that the borrower's hair style matches the preset hair style, and therefore it can be determined that the borrower's hair style meets the corresponding preset portrait condition.

[0058] In another example, the preset tattoo area may be, for example, X% of human skin, where X is an integer greater than or equal to zero. If the borrower's tattoo area is Y%, and Y%>X%, for example, the borrower has tattooed arms (i.e., most or all of the borrower's arms are covered with tattoos), it is identified that the borrower's tattoo area exceeds the preset tattoo area, and therefore it can be determined that the borrower's tattoo area meets the corresponding preset portrait condition. In addition, when X=0, as long as it is identified that the borrower has tattoos, it can be directly determined that the borrower's tattoo area meets the corresponding preset portrait condition.

[0059] In another example, the preset dressing style may be, for example, shirtless, floral shirt, etc. If the lender's dressing style is shirtless or floral shirt, it is identified that the lender's dressing style matches the preset dressing style, and therefore it can be determined that the lender's dressing style meets the corresponding preset portrait condition.

[0060] In another example, the preset jewelry style may be, for example, a large gold chain. If the borrower's jewelry style is a large gold chain, it is identified that the borrower's jewelry style matches the preset jewelry style. Therefore, it can be determined that the borrower's jewelry style meets the corresponding preset portrait conditions.

[0061] If at least two of the above-mentioned character portrait features meet the corresponding preset character portrait conditions, the lender can be assessed as a member of society with the potential for malicious lending.

[0062] In addition, the assessment model can also assess the borrower's occupation based on the borrower's dress style. For example, if it recognizes that the borrower is wearing the uniform of a specific occupation (such as medical staff, security personnel, sanitation workers, etc.), the borrower's occupation can be accurately assessed.

[0063] In addition, the assessment model can also assess the borrower's age and gender based on the borrower's face shape, eyes, nose, mouth, wrinkles, etc.

[0064] In one embodiment, the surrounding environment characteristics include at least one of a living environment and a working environment; and evaluating the borrower's living conditions according to the surrounding environment characteristics through the evaluation model may include: The evaluation model is used to determine whether various surrounding environmental characteristics meet the corresponding preset environmental conditions; if at least one surrounding environmental characteristic meets the corresponding preset environmental conditions, the borrower is assessed to be a social person with low life stability.

[0065] In one embodiment, judging whether each surrounding environment feature meets the corresponding preset environmental condition through the evaluation model may include: The evaluation model is used to determine whether the living environment is lower than a first preset standard, and whether the working environment is lower than a second preset standard.

[0066] The embodiment of the present application can pre-set the corresponding first preset standard and second preset standard based on the typical scenario characteristics of social people with low life stability, so that the subsequent evaluation model can determine whether the borrower's living environment is lower than the first preset standard, and can determine whether the borrower's working environment is lower than the second preset standard.

[0067] In one example, the first preset standard can be characterized by the type of house (such as commercial housing, public housing, affordable housing, etc.) or the degree of dilapidation. If the borrower's living environment is affordable housing (i.e., housing suitable for purchase by low-income families) or is dirty and messy, it is identified that the borrower's living environment is lower than the first preset standard. Therefore, it can be determined that the borrower's living environment meets the corresponding preset environmental conditions.

[0068] In another example, the second preset standard can be characterized by the grade of the office building (such as super-Grade A office buildings, Grade A office buildings, Grade B office buildings, Grade C office buildings, etc.) or the geographical location or decoration facilities. If the borrower's working environment is a Grade C office building (i.e., a basic office building, mainly for start-ups and individual business operators) or the geographical location is remote or the decoration facilities are poor, then the borrower's working environment is identified as being lower than the second preset standard, and it can be determined that the borrower's working environment meets the corresponding preset environmental conditions.

[0069] If at least one of the above-mentioned surrounding environmental characteristics meets the corresponding preset environmental conditions, the borrower can be assessed as a social person with low life stability.

[0070] In addition, the assessment model can also assess the borrower's family situation (such as family background, family members, etc.) based on the borrower's living environment.

[0071] S240, outputting the social background and living conditions as the social attributes of the lender through the evaluation model, and outputting the social attribute score of the lender.

[0072] The evaluation model can output the borrower's social background and living conditions as the borrower's social attributes, and can also output the borrower's age, gender, occupation, and family situation as the borrower's social attributes.

[0073] In addition, the evaluation model can output the social attribute score of the lender. Specifically, each character portrait feature and each surrounding environment feature has a corresponding score. For example, whenever any character portrait feature meets the corresponding preset portrait condition, the evaluation model can count the score corresponding to the character portrait feature as 10 points. For example, whenever any surrounding environment feature meets the corresponding preset environment condition, the evaluation model can count the score corresponding to the surrounding environment feature as 10 points. Then the evaluation model can accumulate the scores corresponding to each character portrait feature and each surrounding environment feature to obtain the social attribute score of the lender.

[0074] S250 assesses the risk level of borrowers based on social attribute scores.

[0075] After the evaluation system obtains the social attribute score of the lender output by the evaluation model, it can evaluate the risk level of the lender based on the social attribute score. The risk level can include at least one of low risk and high risk, where low risk is used to represent that the lender has a low risk of loan default, and high risk is used to represent that the lender has a high risk of loan default.

[0076] In one embodiment, assessing the risk level of a lender based on the social attribute score may include: Determine whether the social attribute score is greater than or equal to the preset risk threshold; if the social attribute score is greater than or equal to the preset risk threshold, the borrower's risk level is assessed as high risk; high risk is used to indicate that the borrower has a high risk of loan default.

[0077] The assessment system can compare the social attribute score with a preset risk threshold, wherein the preset risk threshold can be set according to actual conditions, for example, the preset risk threshold can be set to 30 points.

[0078] In one example, assuming that the preset risk threshold is set to 30 points and the borrower's social attribute score is 50 points, it can be determined that the borrower's social attribute score is greater than the preset risk threshold. Therefore, the assessment system can assess the borrower's risk level as high risk, which means that the borrower's loan default risk is high. Therefore, the assessment system can reject the borrower's loan application in order to reduce the risk of loan business.

[0079] In another example, assuming that the preset risk threshold is set to 30 points and the borrower's social attribute score is 10 points, it can be determined that the borrower's social attribute score is less than the preset risk threshold, so the assessment system can assess the borrower's risk level as low risk, which means that the borrower's loan default risk is low, so the assessment system can approve the borrower's loan application.

[0080] In addition, in addition to assessing the risk level of the borrower based on the social attribute score, the embodiments of the present application can also combine multimodal data (such as the borrower's credit data, voice data, and behavioral data, etc.) for comprehensive assessment to improve the assessment accuracy.

[0081] In one application, if the credit scoring system includes the borrower's credit data, the assessment system may request the borrower's credit data from the credit scoring system so as to comprehensively assess the borrower's risk level based on the borrower's social attribute score and credit data.

[0082] In another application, during the remote interaction with the lender, the evaluation system can obtain the lender's voice data so as to comprehensively evaluate the lender's risk level based on the lender's social attribute score and voice data. For example, if the voice data reflects that the lender has unnatural speech, such as the lender's stumbling speech, or the lender communicating with the person next to him about irrelevant content, or the lender's speech is casual and crude, when the lender has unnatural speech, the evaluation system can infer that the lender has fraudulent intentions, and therefore can evaluate the lender's risk level as high risk.

[0083] In another application, during the remote interaction with the lender, the evaluation system can obtain the lender's behavioral data so as to comprehensively evaluate the lender's risk level based on the lender's social attribute score and behavioral data. For example, if the behavioral data reflects that the lender has unnatural movements, such as the lender frequently scratching his head, or the lender's eyes flickering, or the lender whispering with the person next to him, when the lender has unnatural movements, the evaluation system can infer that the lender has fraudulent intentions, and therefore can evaluate the lender's risk level as high risk.

[0084] S260, visually displays the social attributes and risk levels of lenders.

[0085] The evaluation system can display the evaluation results for the lender in a visual manner, and the evaluation results may include the lender's social attributes and risk level.

[0086] In addition, the assessment system can provide the lender with an explainable explanation in conjunction with the calculation process. For example, if the lender's risk level is determined to be high due to hair style (e.g., bald head), tattoo area (e.g., tattooed arms), living environment (e.g., dirty and messy), etc., the assessment system can provide the lender with these explainable explanations.

[0087] It should be noted that the embodiments of the present application may include but are not limited to the following application scenarios: Credit assessment: Supplement social attribute data for users who lack credit records and improve risk assessment.

[0088] Risk control: Accurately identify potential high-risk users through analysis of character portraits and surrounding environments.

[0089] In one embodiment, the evaluation model can be trained in the following manner: Acquire multiple sample images; wherein each sample image contains a corresponding lender and a true value label, the true value label includes at least one of a first label and a second label, the first label is used to mark whether the lender in the sample image is a social person with potential malicious loans, and the second label is used to mark whether the lender in the sample image is a social person with low life stability; obtain a deep learning model as an evaluation model to be trained; use multiple sample images to train the evaluation model to obtain the recognition result of the evaluation model for each sample image; the recognition result is used to characterize whether the lender in the sample image is a social person with potential malicious loans and / or a social person with low life stability; match the recognition results belonging to the same sample image with the true value label to obtain the matching results of each sample image; calculate the evaluation index of the evaluation model according to the matching results of each sample image; the evaluation index includes at least one of accuracy and recall rate; judge whether the evaluation index is greater than or equal to the preset index; if the evaluation index is greater than or equal to the preset index, determine that the evaluation model training is completed.

[0090] The training process of the evaluation model can include steps such as data preparation, model selection, model training, model testing, and model deployment.

[0091] In the data preparation step, the embodiment of the present application can collect frontal photos of different lenders, and then attach corresponding truth labels to each frontal photo, so that these frontal photos can be used as a training set (i.e., multiple sample pictures). Each sample picture contains only one lender, and the lenders in different sample pictures can be different. The truth label can include at least one of a first label and a second label, the first label can be used to mark whether the lender in the sample picture is a social person with potential malicious loans, and the second label can be used to mark whether the lender in the sample picture is a social person with low life stability.

[0092] In the model selection step, the embodiment of the present application can select a suitable deep learning model (such as CLIP, ResNet, CNN, etc.) as the evaluation model to be trained.

[0093] In the model training step, the embodiment of the present application can use a prepared training set (i.e., multiple sample images) to train the evaluation model, and by continuously adjusting the parameters of the evaluation model, the recognition accuracy of the evaluation model can be improved.

[0094] In the model testing step, the embodiment of the present application can obtain the recognition results of the evaluation model in the previous step for each sample image. The recognition results can be used to characterize whether the lender in the sample image is a social person with potential malicious loans, and / or, the recognition results can be used to characterize whether the lender in the sample image is a social person with low life stability. These recognition results can be used as a test set so that the evaluation model can be tested using the test set.

[0095] In a specific implementation, the embodiment of the present application can match the recognition results belonging to the same sample image with the true value label to obtain the matching result of each sample image. The matching result can be used to characterize whether the recognition result of the evaluation model is consistent with the manually annotated true value label. Therefore, the evaluation index of the evaluation model can be calculated according to the matching result of each sample image. Among them, the evaluation index may include at least one of accuracy and recall. The accuracy indicates the percentage of the correct results recognized by the evaluation model in the total samples, and the recall indicates the percentage of the correct results recognized in all samples that were originally positive. The embodiment of the present application can determine whether the evaluation index of the evaluation model is greater than or equal to the preset index, wherein the preset index may include a preset accuracy rate and a preset recall rate, so it can be determined whether the accuracy of the evaluation model is greater than or equal to the preset accuracy rate, and whether the recall rate of the evaluation model is greater than or equal to the preset recall rate. If the accuracy of the evaluation model is ≥ the preset accuracy rate, and the recall rate of the evaluation model is ≥ the preset recall rate, it means that the performance of the evaluation model is good, so it can be determined that the evaluation model training is completed.

[0096] In the model deployment step, the embodiment of the present application can deploy the trained evaluation model into the evaluation system for use in online evaluation of the evaluation system.

[0097] In addition, the embodiment of the present application can use reinforcement learning technology to optimize the evaluation model. Specifically, when the matching result represents that the recognition result of the evaluation model is inconsistent with the manually annotated true value label, the embodiment of the present application can obtain the tester's feedback on the matching result, so as to continuously optimize the evaluation model according to the feedback, thereby improving the evaluation accuracy.

[0098] It can be seen from this example that the solution provided by this application can be both real-time, convenient and adaptable, and provides an efficient and intelligent solution for visual risk control in remote interaction scenarios. The solution provided by this application may include but is not limited to the following beneficial effects: 1. Rich data dimensions: By introducing image recognition technology, we can supplement the missing dimensions in traditional credit assessment (such as the borrower's social background and living conditions), achieve multi-angle assessment, and improve assessment accuracy.

[0099] 2. Automation and efficiency: The social attribute assessment of the borrower is automatically completed through the assessment model without human intervention, which greatly improves the assessment efficiency and is suitable for processing massive user data.

[0100] 3. Strong scalability: New image features can be added to improve evaluation accuracy according to actual needs.

[0101] 4. Covering people with credit gaps: Among users with insufficient or no credit history, image recognition provides a new basis for assessment, which is particularly suitable for people in emerging markets or those without bank accounts.

[0102] 5. Accurate risk prediction: Image data combined with traditional data can improve the recognition accuracy of the assessment model and help better distinguish between low-risk and high-risk borrowers.

[0103] 6. Fairness and consistency: Model-based evaluation can reduce the impact of human subjective judgment, reduce the risk of human bias, and improve evaluation accuracy. In addition, potential algorithmic discrimination can be further eliminated by optimizing model design.

[0104] Corresponding to the aforementioned application function implementation method embodiment, the present application also provides a lender social attribute visual risk control assessment system based on image recognition, an electronic device, a computer-readable storage medium and corresponding embodiments.

[0105] Figure 3 It is a structural diagram of a visual risk control assessment system for lender social attributes based on image recognition shown in an embodiment of the present application.

[0106] See also Figure 3 The present application provides a visual risk control assessment system for social attributes of lenders based on image recognition, which may include: The visual data acquisition module 310 is used to acquire the visual data of the lender during the remote interaction with the lender; wherein the visual data includes at least one of image data and video data; A feature extraction module 320 is used to input the visual data into a pre-trained evaluation model so as to extract the borrower's character portrait features and surrounding environment features from the visual data through the evaluation model; The social attribute evaluation module 330 is used to evaluate the social attributes of the lender according to the character portrait features and the surrounding environment features through the evaluation model; the social attributes include social background and living conditions.

[0107] In one embodiment, the social attribute assessment module 330 may include: The evaluation submodule is used to evaluate the social background of the borrower based on the character portrait characteristics through the evaluation model, and to evaluate the living conditions of the borrower based on the surrounding environment characteristics through the evaluation model; The output submodule is used to output the social background and living conditions as the social attributes of the lender through the evaluation model, and output the social attribute score of the lender.

[0108] In one embodiment, the system may further include: A risk level assessment module is used to assess the risk level of the borrower based on the social attribute score; The result display module is used to visually display the social attributes and risk levels of borrowers.

[0109] In one embodiment, the character portrait features include at least two of hair style, tattoo area, clothing style, and jewelry style; and the evaluation submodule may include: The first condition judgment unit is used to judge whether each character portrait feature meets the corresponding preset portrait condition through the evaluation model; The first evaluation unit is used to evaluate the lender as a social person with potential malicious lending if at least two character portrait features meet corresponding preset portrait conditions.

[0110] In one implementation, the first condition determination unit may include: a hair style judgment subunit, for judging whether the hair style matches a preset hair style by evaluating the model, and The tattoo area determination subunit is used to determine whether the tattoo area exceeds the preset tattoo area, and a dressing style judging subunit, for judging whether the dressing style matches the preset dressing style, and The jewelry style judgment subunit is used to judge whether the jewelry style is matched with the preset jewelry style; Among them, the preset hair style, preset tattoo area, preset clothing style and preset jewelry style are all set based on the typical image characteristics of social personnel with potential malicious loans.

[0111] In one embodiment, the surrounding environment characteristics include at least one of a living environment and a working environment; the evaluation submodule may further include: The second condition judgment unit is used to judge whether each surrounding environment feature meets the corresponding preset environmental conditions through the evaluation model; The second evaluation unit is used to evaluate the borrower as a social person with low life stability if at least one surrounding environment feature meets the corresponding preset environmental condition.

[0112] In one implementation, the second condition determination unit may include: a living environment judgment subunit, configured to judge whether the living environment is lower than a first preset standard through an evaluation model, and The working environment determination subunit is used to determine whether the working environment is lower than a second preset standard.

[0113] In one embodiment, the risk level assessment module may include: A threshold judgment submodule is used to judge whether the social attribute score is greater than or equal to a preset risk threshold; The high-risk assessment submodule is used to assess the risk level of the borrower as high risk if the social attribute score is greater than or equal to the preset risk threshold; high risk is used to indicate that the borrower has a high risk of loan default.

[0114] In one embodiment, the evaluation model can be trained by the following modules: A sample image acquisition module is used to acquire multiple sample images; wherein each sample image contains a corresponding lender and a true value label, and the true value label includes at least one of a first label and a second label, the first label is used to mark whether the lender in the sample image is a social person with potential malicious loans, and the second label is used to mark whether the lender in the sample image is a social person with low life stability; A model acquisition module, used to acquire a deep learning model as an evaluation model to be trained; The module training module is used to train the evaluation model using multiple sample images to obtain the recognition results of the evaluation model for each sample image; the recognition results are used to characterize whether the borrower in the sample image is a social person with potential malicious loans and / or a social person with low life stability; A matching module is used to match the recognition results belonging to the same sample image with the true value label to obtain the matching results of each sample image; An evaluation index calculation module, used to calculate the evaluation index of the evaluation model according to the matching results of each sample image; the evaluation index includes at least one of accuracy and recall rate; An indicator judgment module is used to judge whether the evaluation indicator is greater than or equal to the preset indicator; The training completion determination module is used to determine that the evaluation model training is completed if the evaluation index is greater than or equal to the preset index.

[0115] In one embodiment, before inputting the visual data into the pre-trained evaluation model, the system may further include: The preprocessing module is used to perform preprocessing operations on visual data; wherein the preprocessing operation includes at least one of image denoising and image enhancement.

[0116] As can be seen from this example, the solution provided by this application obtains the visual data of the lender during the remote interaction with the lender; wherein the visual data includes at least one of image data and video data; the visual data is input into a pre-trained evaluation model to extract the borrower's character portrait features and surrounding environment features from the visual data through the evaluation model; the social attributes of the lender are evaluated by the evaluation model based on the character portrait features and surrounding environment features; the social attributes include social background and living conditions. This application automates the evaluation of the social attributes of the lender through the evaluation model without manual intervention, thereby greatly improving the evaluation efficiency, and the privacy of the lender will not be disclosed to the staff, thereby ensuring that the privacy of the lender can be effectively analyzed while protecting it, and also reducing the impact of human subjective judgment and reducing the risks caused by human bias, thereby improving the accuracy of the evaluation. Furthermore, this application introduces image recognition technology to realize the analysis and judgment of visual risk control of unstructured data (such as visual data such as image data and video data), which can supplement the missing dimensions in traditional credit evaluation (such as the social background and living conditions of the lender), thereby realizing multi-angle evaluation and further improving the accuracy of the evaluation.

[0117] Regarding the system in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated again here.

[0118] Figure 4 It is a schematic diagram of the structure of an electronic device shown in an embodiment of the present application.

[0119] See also Figure 4 , the electronic device 400 includes a memory 410 and a processor 420.

[0120] The processor 420 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.

[0121] The memory 410 may include various types of storage units, such as system memory, read-only memory (ROM), and permanent storage devices. Among them, ROM can store static data or instructions required by the processor 420 or other modules of the computer. The permanent storage device may be a readable and writable storage device. The permanent storage device may be a non-volatile storage device that does not lose the stored instructions and data even after the computer is powered off. In some embodiments, the permanent storage device uses a large-capacity storage device (such as a magnetic or optical disk, flash memory) as a permanent storage device. In some other embodiments, the permanent storage device may be a removable storage device (such as a floppy disk, optical drive). The system memory may be a readable and writable storage device or a volatile readable and writable storage device, such as a dynamic random access memory. The system memory may store some or all instructions and data required by the processor at runtime. In addition, the memory 410 may include any combination of computer-readable storage media, including various types of semiconductor storage chips (such as DRAM, SRAM, SDRAM, flash memory, programmable read-only memory), and disks and / or optical disks may also be used. In some embodiments, the memory 410 may include a readable and / or writable removable storage device, such as a laser disc (CD), a read-only digital versatile disc (such as a DVD-ROM, a double-layer DVD-ROM), a read-only Blu-ray disc, an ultra-density optical disc, a flash memory card (such as an SD card, a mini SD card, a Micro-SD card, etc.), a magnetic floppy disk, etc. Computer-readable storage media do not include carrier waves and transient electronic signals transmitted wirelessly or wired.

[0122] The memory 410 stores executable codes, and when the executable codes are processed by the processor 420 , the processor 420 can execute part or all of the methods described above.

[0123] In addition, the method according to the present application may also be implemented as a computer program or a computer program product, which includes computer program code instructions for executing some or all of the steps in the above method of the present application.

[0124] Alternatively, the present application can also be implemented as a computer-readable storage medium (or non-transitory machine-readable storage medium or machine-readable storage medium) on which executable code (or computer program or computer instruction code) is stored. When the executable code (or computer program or computer instruction code) is executed by a processor of an electronic device (or server, etc.), the processor executes part or all of the steps of the above-mentioned method according to the present application.

[0125] The embodiments of the present application have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A visual risk control assessment method for social attributes of lenders based on image recognition, characterized in that: The method comprises: During the remote interaction with the lender, obtaining visual data of the lender; wherein the visual data includes at least one of image data and video data; Inputting the visual data into a pre-trained evaluation model to extract the borrower's character portrait features and surrounding environment features from the visual data through the evaluation model; The social attributes of the lender are evaluated through the evaluation model according to the character portrait features and the surrounding environment features; the social attributes include social background and living conditions.

2. The method according to claim 1, characterized in that: The step of evaluating the social attributes of the lender according to the character portrait features and the surrounding environment features through the evaluation model includes: Using the evaluation model to evaluate the social background of the borrower based on the characteristics of the character portrait, and using the evaluation model to evaluate the living conditions of the borrower based on the characteristics of the surrounding environment; Outputting the social background and the living conditions as the social attributes of the lender through the evaluation model, and outputting the social attribute score of the lender; The method further comprises: Based on the social attribute score, assessing the risk level of the lender; The social attributes and risk level of the lender are displayed visually.

3. The method according to claim 2, characterized in that The character portrait features include at least two of hair style, tattoo area, clothing style, and jewelry style; and the social background of the lender is evaluated based on the character portrait features by the evaluation model, including: Determine whether each of the character portrait features meets the corresponding preset portrait conditions through the evaluation model; If at least two of the character portrait characteristics meet the corresponding preset portrait conditions, the lender is assessed as a social person with potential malicious lending; The step of judging whether each of the character portrait features meets the corresponding preset portrait conditions by the evaluation model includes: Determining whether the hair style matches a preset hair style, determining whether the tattoo area exceeds a preset tattoo area, determining whether the dress style matches a preset dress style, and determining whether the jewelry style matches a preset jewelry style by using the evaluation model; The preset hair style, the preset tattoo area, the preset clothing style and the preset accessory style are all set based on the typical image characteristics of the social personnel with potential malicious loans.

4. The method according to claim 2, characterized in that: The surrounding environment characteristics include at least one of a living environment and a working environment; the evaluation model is used to evaluate the living conditions of the borrower according to the surrounding environment characteristics, including: Determine whether each of the surrounding environment characteristics meets the corresponding preset environmental conditions through the evaluation model; If at least one of the surrounding environmental characteristics meets the corresponding preset environmental conditions, the borrower is assessed as a social person with low life stability; The step of judging whether each of the surrounding environment characteristics meets the corresponding preset environmental conditions by the evaluation model includes: The evaluation model is used to determine whether the living environment is lower than a first preset standard, and to determine whether the working environment is lower than a second preset standard.

5. The method according to claim 2, characterized in that: The step of evaluating the risk level of the lender based on the social attribute score includes: Determining whether the social attribute score is greater than or equal to a preset risk threshold; If the social attribute score is greater than or equal to the preset risk threshold, the risk level of the borrower is assessed to be high risk; the high risk is used to indicate that the borrower has a high risk of loan default.

6. The method according to claim 1, characterized in that The evaluation model is trained in the following way: Acquire multiple sample images; wherein each of the sample images contains a corresponding lender and a corresponding true value label, wherein the true value label includes at least one of a first label and a second label, wherein the first label is used to mark whether the lender in the sample image is a social person with potential malicious lending, and the second label is used to mark whether the lender in the sample image is a social person with low life stability; Obtain a deep learning model as an evaluation model to be trained; The evaluation model is trained using a plurality of the sample images to obtain recognition results of the evaluation model for each of the sample images; the recognition results are used to characterize whether the lender in the sample image is a social person with potential malicious lending and / or a social person with low life stability; Matching the recognition results belonging to the same sample image with the true value label to obtain matching results for each sample image; Calculating an evaluation index of the evaluation model according to the matching results of each of the sample images; the evaluation index includes at least one of accuracy and recall; Determine whether the evaluation index is greater than or equal to a preset index; If the evaluation index is greater than or equal to the preset index, it is determined that the training of the evaluation model is completed.

7. The method according to claim 1, characterized in that Before inputting the visual data into the pre-trained evaluation model, the method further includes: Perform a preprocessing operation on the visual data; wherein the preprocessing operation includes at least one of image denoising and image enhancement.

8. A visual risk control assessment system for social attributes of lenders based on image recognition, characterized in that: The system comprises: A visual data acquisition module, used to acquire visual data of the lender during the remote interaction with the lender; wherein the visual data includes at least one of image data and video data; A feature extraction module, used for inputting the visual data into a pre-trained evaluation model, so as to extract the borrower's character portrait features and surrounding environment features from the visual data through the evaluation model; The social attribute evaluation module is used to evaluate the social attributes of the lender according to the character portrait features and the surrounding environment features through the evaluation model; the social attributes include social background and living conditions.

9. An electronic device, characterized in that: include: processor; as well as A memory having executable codes stored thereon, which, when executed by the processor, causes the processor to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having executable codes stored thereon, which, when executed by a processor of an electronic device, causes the processor to execute the method according to any one of claims 1 to 7.

Images