Credit risk assessment method and device based on invoice data

Abstract

The invention discloses a credit risk assessment method and device based on invoice data. The method comprises the following steps: acquiring a plurality of invoice images of a target customer in a preset historical time period; performing text recognition and information extraction on each invoice image to obtain invoice key information in each invoice image; integrating the invoice key information of the plurality of invoice images to obtain a derivative variable; inputting the derivative variables into a risk assessment model, and processing the derivative variables by using the risk assessment model to obtain a risk assessment score for credit risk assessment of the target customer; and determining an evaluation result of credit risk evaluation on the target customer according to a risk threshold range in which the risk evaluation score is located. According to the method and the device, the technical problem that the evaluation result is inaccurate because the authenticity and the effectiveness cannot be guaranteed because the credit risk evaluation of the customer mainly depends on manual verification in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of financial risk assessment, and in particular to a credit risk assessment method and device based on invoice data. Background Art

[0002] In the current specific market environment, some small-scale corporate customers have the characteristics of frequent financing needs, weak risk resistance, irregular management, and imperfect financial systems. As a result, these customers' loans have the characteristics of small amounts, large numbers, little information, short processes, and difficult control. The potential risks of banks are large, and the non-performing rate of small and micro-enterprise loans of commercial banks has continued to rise. Traditional customer pre-loan assessment methods are mostly risk assessment tools based on traditional credit score cards. These tools mainly rely on the customer's financial statements, bank transaction records and other information for assessment. However, most customers have chaotic internal management, imperfect financial systems, low credibility of financial statements, and information asymmetry between banks and enterprises, which will lead to increased bank loan risks.

[0003] In general, the existing risk assessment methods for small and micro enterprises have the following defects: (1) Many existing credit risk assessment methods mainly rely on internal financial statements, bank transaction records and other information for assessment, and their authenticity and effectiveness cannot be guaranteed. There is no way to reflect the real-time operating conditions of the enterprise, which brings obstacles to the pre-loan, mid-loan and post-loan management of the enterprise; (2) The existing commercial invoice information extraction interface can easily lead to information leakage; (3) Enterprise information data may be tampered with and falsified, and most existing methods rely on manual verification.

[0004] Currently, no effective solution has been proposed to address the problem that the above-mentioned related technologies mainly rely on manual verification when conducting credit risk assessment on customers, and the authenticity and effectiveness cannot be guaranteed, resulting in inaccurate assessment results. Summary of the invention

[0005] The embodiments of the present invention provide a method and device for credit risk assessment based on invoice data, so as to at least solve the technical problem in the related art that credit risk assessment of customers mainly relies on manual verification, the authenticity and effectiveness cannot be guaranteed, and the assessment results are inaccurate.

[0006] According to one aspect of an embodiment of the present invention, a credit risk assessment method based on invoice data is provided, comprising: obtaining multiple invoice images of a target customer within a preset historical time period, wherein the target customer is an object for which a credit risk assessment is required; performing text recognition and information extraction on each of the invoice images to obtain key invoice information in each of the invoice images, wherein the key invoice information refers to invoice information whose influence on credit risk assessment is higher than an influence threshold; integrating the key invoice information of the multiple invoice images to obtain derived variables, wherein the derived variables are used to reflect the credit risk of the target customer within the preset historical time period. Financial health, the financial health is an evaluation parameter for credit risk assessment; the derived variables are input into a risk assessment model so as to process the derived variables using the risk assessment model to obtain a risk assessment score for credit risk assessment of the target customer, wherein the risk assessment model is trained using multiple sets of first training data by machine learning, each of the multiple sets of first training data comprising: sample derived variables and sample risk assessment scores corresponding to the sample derived variables; and the evaluation result of the credit risk assessment of the target customer is determined according to the risk threshold range of the risk assessment score.

[0007] Optionally, obtaining multiple invoice images of the target customer within a preset historical time period includes: obtaining multiple first sub-invoice images of the target customer within the preset historical time period based on a first acquisition channel, wherein the first acquisition channel refers to a channel for directly interacting with a target invoice providing platform to obtain the invoice image, and the target invoice providing platform refers to a platform that provides the invoice image of the target customer; obtaining multiple second sub-invoice images of the target customer within the preset historical time period based on a second acquisition channel, wherein the second acquisition channel refers to a channel for interacting with the target invoice providing platform through a third-party platform calling an API interface to obtain the invoice image, and the third-party platform refers to an invoice providing platform other than the target invoice providing platform; obtaining multiple third sub-invoice images of the target customer within the preset historical time period based on a third acquisition channel, wherein the third acquisition channel refers to a channel for directly logging into the target invoice providing platform by calling a preset program to obtain the invoice image; and determining that the first sub-invoice image, the second sub-invoice image and the third sub-invoice image are the invoice images of the target customer within the preset historical time period.

[0008] Optionally, performing text recognition and information extraction on each of the invoice images to obtain the key invoice information in each of the invoice images includes: using image processing technology to perform image preprocessing on multiple invoice images in sequence to obtain target invoice images corresponding to each of the invoice images; using an invoice text positioning model to determine the region type of each pixel region in each of the target invoice images, wherein the invoice text positioning model is trained by machine learning using multiple sets of second training data, each of the multiple sets of second training data includes: sample target invoice images, sample region types corresponding to the sample target invoice images; determining the pixel region whose region type is a text region as a target pixel region; inputting the target pixel region in each of the target invoice images into the invoice text recognition model in sequence, so as to perform text recognition on the target pixel region using the invoice text recognition model to obtain the key invoice information in each of the target invoice images, wherein the invoice text recognition model is trained by machine learning using multiple sets of third training data, each of the multiple sets of third training data includes: sample target pixel region, sample invoice key information corresponding to the sample target pixel region.

[0009] Optionally, after performing text recognition and information extraction on each of the invoice images to obtain the invoice key information in each of the invoice images, the credit risk assessment method based on invoice data also includes: integrating the invoice key information in multiple invoice images to obtain an invoice key information set; determining that the feature items in the invoice key information set whose frequency of appearance is higher than a frequency threshold are target feature items, wherein the feature items are field names in the invoice key information set; analyzing the target feature items using a priori algorithms to obtain feature association relationships between any two of the target feature items or any multiple of the target feature items; analyzing the feature association relationships using a K-means clustering algorithm to obtain an abnormal fluctuation index between the target feature items; and determining that the authenticity of the invoice key information has passed verification when the abnormal fluctuation index is lower than a fluctuation threshold.

[0010] Optionally, the risk assessment model includes a first risk assessment model and a second risk assessment model, and the derived variables are input into the risk assessment model so as to process the derived variables using the risk assessment model to obtain a risk assessment score for credit risk assessment of the target customer, including: performing risk analysis on the derived variables to obtain the correlation between the derived variables and the credit risk; determining the derived variables whose correlation is greater than a correlation threshold as the first derived variable, and determining the derived variables whose correlation is not greater than the correlation threshold as the second derived variable; using the first risk assessment model to determine a first risk assessment score obtained by performing credit risk assessment on the first derived variable, and using the second risk assessment model to determine a second risk assessment score obtained by performing credit risk assessment on the second derived variable; and weightedly fusing the first risk assessment score and the second risk assessment score to obtain the risk assessment score.

[0011] Optionally, performing risk analysis on the derived variables to obtain the correlation between the derived variables and the credit risk includes: obtaining variable characteristics of the derived variables, wherein the variable characteristics are used to analyze the correlation between the derived variables and the credit risk; and calculating the correlation between the derived variables and the credit risk using a first formula according to the variable characteristics, wherein the first formula is: i represents the label of the binning interval in the variable feature, N represents the total number of binning intervals in the variable feature, Bad s Indicates the number of low-risk customers whose risk level is not higher than the risk threshold in the i-th bin interval, Bad t Indicates the total number of low-risk customers whose risk level is not higher than the risk level threshold in all the binning intervals, Good s Good represents the number of high-risk customers whose risk level in the i-th bin interval is higher than the risk level threshold. t Indicates the total number of high-risk customers whose risk level is lower than the risk level threshold in all the binning intervals, WOE i It represents the difference between the risk status of the ith sub-box interval and the total risk status of all the sub-box intervals, and the risk status is obtained by analyzing the number of low-risk customers and the proportion of high-risk customer data.

[0012] Optionally, the risk threshold range includes: a first risk threshold range, a second risk threshold range and a third risk threshold range, and the assessment result of the credit risk assessment of the target customer is determined according to the risk threshold range in which the risk assessment score is located, including: when the risk assessment score is within the first risk threshold range, determining that the assessment result is low risk, wherein the first risk threshold range is a range in which the risk assessment score is less than the first risk threshold; when the risk assessment score is within the second risk threshold range, determining that the assessment result is medium risk, wherein the second risk threshold range is a range in which the risk assessment score is greater than or equal to the first risk threshold and less than the second risk threshold, and the second risk threshold is greater than the first risk threshold; when the risk assessment score is within the third risk threshold range, determining that the assessment result is high risk, wherein the third risk threshold range is a range in which the risk assessment score is greater than or equal to the second risk threshold.

[0013] According to another aspect of an embodiment of the present invention, there is also provided a credit risk assessment device based on invoice data, comprising: a first acquisition unit, used to acquire multiple invoice images of a target customer within a preset historical time period, wherein the target customer is an object for which a credit risk assessment is required; a second acquisition unit, used to perform text recognition and information extraction on each of the invoice images respectively, to obtain key invoice information in each of the invoice images, wherein the key invoice information refers to invoice information whose influence on the credit risk assessment is higher than an influence threshold; a third acquisition unit, used to integrate the key invoice information of the multiple invoice images, to obtain a derived variable, wherein the derived variable is used to reflect the influence of the target customer in the preset historical time period. The financial health within a time period, wherein the financial health is an evaluation parameter for credit risk assessment; a fourth acquisition unit, used to input the derived variable into a risk assessment model, so as to process the derived variable using the risk assessment model to obtain a risk assessment score for credit risk assessment of the target customer, wherein the risk assessment model is trained using multiple groups of first training data by machine learning, and each of the multiple groups of first training data includes: sample derived variables and sample risk assessment scores corresponding to the sample derived variables; a first determination unit, used to determine an evaluation result of the credit risk assessment of the target customer according to a risk threshold range within which the risk assessment score is located.

[0014] Optionally, the first acquisition unit includes: a first acquisition module, used to acquire multiple first sub-invoice images of the target customer within the preset historical time period based on a first acquisition channel, wherein the first acquisition channel refers to a channel for directly interacting with a target invoice providing platform to acquire the invoice image, and the target invoice providing platform refers to a platform that provides the invoice image of the target customer; a second acquisition module, used to acquire multiple second sub-invoice images of the target customer within the preset historical time period based on a second acquisition channel, wherein the second acquisition channel refers to a channel for interacting with the target invoice providing platform to acquire the invoice image through a third-party platform calling an API interface, and the third-party platform refers to an invoice providing platform other than the target invoice providing platform; a third acquisition module, used to acquire multiple third sub-invoice images of the target customer within the preset historical time period based on a third acquisition channel, wherein the third acquisition channel refers to a channel for directly logging into the target invoice providing platform by calling a preset program to acquire the invoice image; a first determination module, used to determine that the first sub-invoice image, the second sub-invoice image and the third sub-invoice image are the invoice images of the target customer within the preset historical time period.

[0015] Optionally, the second acquisition unit includes: a fourth acquisition module, which is used to perform image preprocessing on multiple invoice images in sequence using image processing technology to obtain target invoice images corresponding to each invoice image; a second determination module, which is used to determine the region type of each pixel region in each target invoice image using an invoice text positioning model, wherein the invoice text positioning model is trained using multiple sets of second training data in a machine learning manner, and each of the multiple sets of second training data includes: a sample target invoice image and a sample region type corresponding to the sample target invoice image; a third determination module, which is used to determine that the pixel region whose region type is a text region is a target pixel region; a fifth acquisition module, which is used to input the target pixel region in each target invoice image into an invoice text recognition model in sequence, so as to perform text recognition on the target pixel region using the invoice text recognition model to obtain the invoice key information in each target invoice image, wherein the invoice text recognition model is trained using multiple sets of third training data in a machine learning manner, and each of the third training data includes: a sample target pixel region and sample invoice key information corresponding to the sample target pixel region.

[0016] Optionally, the credit risk assessment device based on invoice data further includes: a fifth acquisition unit, configured to integrate the invoice key information in multiple invoice images to obtain an invoice key information set after performing text recognition and information extraction on each invoice image to obtain the invoice key information in each invoice image; a second determination unit, configured to determine a feature item with a frequency higher than a frequency threshold in the invoice key information set as a target feature item, where the feature item is a field name in the invoice key information set; a sixth acquisition unit, configured to analyze the target feature item by using a priori algorithm to obtain a feature association relationship between any two target feature items or any multiple target feature items; a seventh acquisition unit, configured to analyze the feature association relationship by using a K-means clustering algorithm to obtain an abnormal fluctuation index between the target feature items; a third determination unit, configured to determine that the authenticity of the invoice key information passes the verification when the abnormal fluctuation index is lower than a fluctuation threshold.

[0017] Optionally, the risk assessment model includes a first risk assessment model and a second risk assessment model. The fourth acquisition unit includes: a sixth acquisition module, configured to perform risk analysis on the derived variables to obtain the correlation between the derived variables and the credit risk; a fourth determination module, configured to determine the derived variables with a correlation greater than a correlation threshold as first derived variables, and determine the derived variables with a correlation not greater than the correlation threshold as second derived variables; a seventh acquisition module, configured to use the first risk assessment model to determine a first risk assessment score obtained by performing credit risk assessment on the first derived variables, and use the second risk assessment model to determine a second risk assessment score obtained by performing credit risk assessment on the second derived variables; an eighth acquisition module, configured to perform weighted fusion on the first risk assessment score and the second risk assessment score to obtain the risk assessment score.

[0018] Optionally, the sixth acquisition module includes: an acquisition sub-module, configured to acquire variable features of the derived variables, where the variable features are used to analyze the correlation between the derived variables and the credit risk; a calculation sub-module, configured to calculate the correlation between the derived variables and the credit risk according to the variable features by using a first formula, where the first formula is: i represents the label of the bin interval in the variable features, N represents the total number of the bin intervals in the variable features, Bad s represents the number of low-risk customers with a risk level not higher than a risk level threshold in the i-th bin interval, Bad t represents the total number of low-risk customers with a risk level not higher than the risk level threshold in all bin intervals, Good sGood represents the number of high-risk customers whose risk level in the i-th bin interval is higher than the risk level threshold. t Indicates the total number of high-risk customers whose risk level is lower than the risk level threshold in all the binning intervals, WOE i It represents the difference between the risk status of the ith sub-box interval and the total risk status of all the sub-box intervals, and the risk status is obtained by analyzing the number of low-risk customers and the proportion of high-risk customer data.

[0019] Optionally, the risk threshold range includes: a first risk threshold range, a second risk threshold range and a third risk threshold range, and the first determination unit includes: a fifth determination module, used to determine that the assessment result is low risk when the risk assessment score is within the first risk threshold range, wherein the first risk threshold range is a range in which the risk assessment score is less than the first risk threshold; a sixth determination module, used to determine that the assessment result is medium risk when the risk assessment score is within the second risk threshold range, wherein the second risk threshold range is a range in which the risk assessment score is greater than or equal to the first risk threshold and less than the second risk threshold, and the second risk threshold is greater than the first risk threshold; a seventh determination module, used to determine that the assessment result is high risk when the risk assessment score is within the third risk threshold range, wherein the third risk threshold range is a range in which the risk assessment score is greater than or equal to the second risk threshold.

[0020] According to another aspect of an embodiment of the present invention, there is further provided a credit risk assessment system based on invoice data, wherein the credit risk assessment system based on invoice data uses any one of the above-mentioned credit risk assessment methods based on invoice data.

[0021] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is further provided, wherein the computer-readable storage medium includes a stored program, wherein the program executes any one of the above-mentioned credit risk assessment methods based on invoice data.

[0022] According to another aspect of an embodiment of the present invention, a processor is further provided, the processor being used to run a program, wherein the program executes any one of the above-mentioned credit risk assessment methods based on invoice data when running.

[0023] According to another aspect of an embodiment of the present invention, a computer program product is provided, comprising computer instructions, wherein when the computer instructions are executed by a processor, any one of the above-mentioned credit risk assessment methods based on invoice data is executed.

[0024] In an embodiment of the present invention, multiple invoice images of a target customer within a preset historical time period can be obtained, wherein the target customer is an object that needs to be evaluated for credit risk; text recognition and information extraction are performed on each invoice image respectively to obtain key invoice information in each invoice image, wherein the key invoice information refers to invoice information whose impact on credit risk assessment is higher than an impact threshold; the key invoice information of multiple invoice images is integrated to obtain derived variables, wherein the derived variables are used to reflect the financial health of the target customer within the preset historical time period, and the financial health is an evaluation parameter for credit risk assessment; the derived variables are input into a risk assessment model to process the derived variables using the risk assessment model to obtain a risk assessment score for credit risk assessment of the target customer, wherein the risk assessment model is trained by using multiple groups of first training data in a machine learning manner, and each of the multiple groups of first training data includes: sample derived variables and sample risk assessment scores corresponding to the sample derived variables; and the evaluation result of credit risk assessment of the target customer is determined according to the risk threshold range of the risk assessment score. Through the above technical scheme, the purpose of extracting key information from invoice images and processing them into derivative variables is achieved, so as to use the corresponding risk assessment model to conduct credit risk assessment on target customers based on the derivative variables. The technical effect of using the model to process invoice images to analyze the credit risk of target customers is achieved, and the accuracy and effectiveness of the assessment are improved. This solves the technical problem in related technologies that credit risk assessment of customers mainly relies on manual verification, and the authenticity and effectiveness cannot be guaranteed, resulting in inaccurate assessment results. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0026] Figure 1 It is a hardware structure block diagram of a mobile terminal of a credit risk assessment method based on invoice data according to an embodiment of the present invention;

[0027] Figure 2 is a flow chart of a credit risk assessment method based on invoice data according to an embodiment of the present invention;

[0028] Figure 3 is a flowchart of an optional credit risk assessment method based on invoice data according to an embodiment of the present invention;

[0029] Figure 4 is a flowchart of modeling an invoice detection and recognition model according to an embodiment of the present invention;

[0030] Figure 5is a flow chart of invoice authenticity detection according to an embodiment of the present invention;

[0031] Figure 6 is a flowchart of building a credit scoring model and a risk target customer machine learning early warning model according to an embodiment of the present invention;

[0032] Figure 7 is a schematic diagram of a credit risk assessment device based on invoice data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0033] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0034] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0035] As introduced in the background technology, the related art mainly relies on manual verification when evaluating the credit risk of customers, and the authenticity and effectiveness cannot be guaranteed, resulting in inaccurate evaluation results. In view of the above defects, a credit risk evaluation method and device based on invoice data are provided in an embodiment of the present invention.

[0036] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0037] The method embodiments provided in the embodiments of the present invention can be executed in a mobile terminal, a computer terminal or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 1 is a hardware structure block diagram of a mobile terminal of a credit risk assessment method based on invoice data according to an embodiment of the present invention. Figure 1As shown, the mobile terminal may include one or more ( Figure 1 Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data, wherein the mobile terminal may also include a transmission device 106 and an input / output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the mobile terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0038] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the credit risk assessment method based on invoice data in the embodiment of the present invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, the above method is implemented. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of the above network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof. The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the mobile terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0039] According to an embodiment of the present invention, a method embodiment of a credit risk assessment method based on invoice data is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0040] Figure 2 is a flow chart of a credit risk assessment method based on invoice data according to an embodiment of the present invention. Figure 2As shown, the method comprises the following steps:

[0041] Step S202, obtaining a plurality of invoice images of a target customer within a preset historical time period, wherein the target customer is a subject for which a credit risk assessment is required.

[0042] In this embodiment, the credit risk of the target customer can be assessed by obtaining the invoice data of the target customer in the past period of time (assuming the past year as an example) and analyzing the invoice data.

[0043] Combine the following Figure 3 The above embodiments of the present invention are described in detail. Figure 3 is a flowchart of an optional credit risk assessment method based on invoice data according to an embodiment of the present invention.

[0044] According to the above embodiment of the present invention, in the above step S202, a plurality of invoice images of the target customer within a preset historical time period are obtained, including: obtaining a plurality of first sub-invoice images of the target customer within the preset historical time period based on a first acquisition channel, wherein the first acquisition channel refers to a channel for directly performing data interaction with a target invoice providing platform to obtain invoice images, and the target invoice providing platform refers to a platform that provides invoice images of the target customer; obtaining a plurality of second sub-invoice images of the target customer within the preset historical time period based on a second acquisition channel, wherein the second acquisition channel refers to a channel for performing data interaction with a target invoice providing platform through a third-party platform calling an API interface to obtain invoice images, and the third-party platform refers to an invoice providing platform other than the target invoice providing platform; obtaining a plurality of third sub-invoice images of the target customer within the preset historical time period based on a third acquisition channel, wherein the third acquisition channel refers to a channel for directly logging into the target invoice providing platform by calling a preset program to obtain invoice images; and determining the first sub-invoice image, the second sub-invoice image and the third sub-invoice image as the invoice images of the target customer within the preset historical time period.

[0045] like Figure 3As shown, the target customer invoice data can be collected mainly through three methods: API connection (tax data service third party, i.e. the second acquisition channel mentioned above), bank-tax interaction (system connection between the bank and the target invoice providing platform, i.e. the first acquisition channel mentioned above), and RPA (the bank uses the customer's electronic target invoice providing platform account and password to log in to the electronic target invoice providing platform, and the system simulates the customer's behavior to download tax and invoice data with the customer's authorization, i.e. the third acquisition channel mentioned above). Among them, RPA is the main invoice acquisition method, which is combined with credit application information and basic enterprise information as the main invoice data for evaluating the credit risk of the target customer; assuming that the preset time window length is T1 (T1 is 12 months in the embodiment of the present invention, of course, other values ​​can also be selected as the time window length, which is not specifically limited here), obtain the sample set Y of the target customer invoice to be evaluated within the time window T1 = (y 1 ,y 2 , ..., y n ), where n is the total number of invoices for the target customer within the time window T1.

[0046] Specifically, these three methods are described in detail below: 1) API docking: API (Application Programming Interface) docking refers to obtaining invoice images by exchanging data with the target invoice providing platform through calling the API interface through a third-party data service provider. These service providers (i.e. the above-mentioned third-party platform) usually have a direct data connection with the target invoice providing platform (i.e. the above-mentioned target invoice providing platform) and can obtain the company's invoice data in real time or near real time. By connecting with the API interface of these service providers, banks can automatically and efficiently obtain the target customer's invoice information for risk assessment; 2) Bank-tax interaction: Bank-tax interaction refers to the electronic data interaction mechanism between the bank and the target invoice providing platform. Under this mechanism, the bank can authorize the acquisition of the target customer's invoice data and other tax information in the target invoice providing platform system. Bank-tax interaction usually requires a formal cooperation agreement between the bank and the target invoice provider platform to ensure the security and legality of the data; in this way, the bank can obtain more official and accurate invoice data, which helps to improve the accuracy of risk assessment; 3) RPA (Robotic Process Automation): RPA technology allows banks to use preset programs to simulate manual operations on the electronic target invoice provider platform website, automatically log in and download the target customer's tax and invoice data. This technology does not require a direct data interface when acquiring data, but is completed through automated operations. It can adapt to the interface changes of the target invoice provider platform website, but also needs to ensure the compliance of the operation, especially when using the customer account password, the customer's explicit authorization must be obtained; the target customer invoice data collected through the above three methods has high authenticity, real-time and effectiveness. Banks can use this data to more comprehensively evaluate the target customer's operating conditions and financial health, thereby making more accurate loan decisions and risk assessments.

[0047] Step S204, performing text recognition and information extraction on each invoice image to obtain key invoice information in each invoice image, wherein the key invoice information refers to invoice information whose impact on credit risk assessment is higher than an impact threshold.

[0048] In this embodiment, the acquired invoice image data can be subjected to preprocessing operations such as noise reduction, grayscale, and binarization, and the invoice detection and recognition model (i.e., the invoice text positioning model and the invoice text recognition model in the embodiment of the present invention) can be used to perform text recognition and information extraction on each invoice image to obtain the key invoice information in each invoice image, thereby selecting representative information, thereby improving the efficiency and accuracy of credit risk assessment.

[0049] According to the above embodiment of the present invention, in the above step S204, text recognition and information extraction are performed on each invoice image respectively to obtain the invoice key information in each invoice image, including: using image processing technology to perform image preprocessing on multiple invoice images in sequence to obtain target invoice images corresponding to each invoice image; using an invoice text positioning model to determine the region type of each pixel region in each target invoice image, wherein the invoice text positioning model is trained by machine learning using multiple sets of second training data, and each of the multiple sets of second training data includes: sample target invoice images, sample region types corresponding to the sample target invoice images; determining the pixel region whose region type is a text region as the target pixel region; sequentially inputting the target pixel region in each target invoice image into the invoice text recognition model, so as to perform text recognition on the target pixel region using the invoice text recognition model to obtain the invoice key information in each target invoice image, wherein the invoice text recognition model is trained by machine learning using multiple sets of third training data, and each of the third training data includes: sample target pixel region, sample invoice key information corresponding to the sample target pixel region.

[0050] In this embodiment, the invoice sample set Y can be preprocessed, and the invoice image can be de-noised, grayed, and binarized using a Gaussian filter to generate a sample Then input it into the invoice detection and recognition model to extract the invoice key data and obtain the invoice key information set

[0051] Specifically, the following Figure 4 The construction process of the invoice detection and recognition model in the embodiment of the present invention is described in detail. Figure 4 FIG. 1 is a flowchart of modeling an invoice detection and recognition model according to an embodiment of the present invention. Figure 4 As shown in the figure, the specific process of building an invoice detection and recognition model is as follows:

[0052] (1) Collect invoice data sets, manually mark the four-point coordinates and data of key invoice information, establish data sets, and divide the training set T = {(p 1 , v 1 ), (p 2 , v 2 ),...(p m , v m )}, p and v represent all the text position information and corresponding text values ​​of an invoice respectively, both are sets, and m represents the number of training sample invoices;

[0053] (2) After preprocessing the training set by denoising, graying, and binarization, The processed invoice image is input into the DBNet model, which will classify each pixel in the image as text or not. in, represents a binary image, k is the influence factor set to 50, i and j are pixel positions, P represents the network prediction probability, T represents the network prediction threshold, and the area is judged as a text / non-text area in the following way: Among them, t is the set threshold. When the network prediction probability is greater than or equal to t, it is predicted as a text area;

[0054] (3) The invoice detection model locates the text area After that, the text area is cropped to obtain D, and the CRNN text recognition model is used to output the invoice key information extraction result V. The specific execution steps of CRNN are: ① Scale the image to 32×W×1, where W is the image width; ② After convolution, it becomes 1×(W / 4)×512; ③ Then for LSTM, set the time step T=(W / 4), D=512, and then input the feature into LSTM (long short-term memory network). LSTM has 256 hidden nodes. After LSTM, it becomes a vector of length T×n class, and then processed by the softmax function (normalized exponential function). Each element of the column vector represents the corresponding character prediction probability. Finally, the prediction results are de-redundant and merged into a complete recognition result V=(v 1 , v 2 , ..., v n ), V is the key information set of the invoice, and class here refers to the character type. When the LSTM part of the CRNN model processes the convolutional feature map, it outputs a vector of length T (time step). Each element of this vector is a vector of length n class, where nclass is the number of all possible character categories. This vector of length n class represents the probability distribution of the model predicting different character categories for the input feature at the current time step. After being processed by the softmax function, each element in this vector becomes the predicted probability of the character category being the current category, and the sum of all elements equals 1;

[0055] (4) The system obtains the extraction results of invoice key information through interface request and returns it in JSON format.

[0056] According to the above embodiment of the present invention, after the above step S204, that is, after performing text recognition and information extraction on each invoice image respectively to obtain the invoice key information in each invoice image, the credit risk assessment method based on invoice data also includes: integrating the invoice key information in multiple invoice images to obtain an invoice key information set; determining that the feature items in the invoice key information set whose frequency of occurrence is higher than the frequency threshold are target feature items, wherein the feature items are field names in the invoice key information set; analyzing the target feature items using a priori algorithms to obtain feature association relationships between any two target feature items or any multiple target feature items; analyzing the feature association relationships using a K-means clustering algorithm to obtain an abnormal fluctuation index between the target feature items; and when the abnormal fluctuation index is lower than the fluctuation threshold, determining that the authenticity of the invoice key information has passed the verification.

[0057] Combine the following Figure 5 The above embodiments of the present invention are described in detail. Figure 5 is a flow chart of invoice authenticity detection according to an embodiment of the present invention.

[0058] like Figure 5 As shown in the figure, after obtaining the key information of the invoice, it is also necessary to detect the authenticity of the extracted key information V of the invoice. The association rule mining algorithm Apriori algorithm can be used to discover the pattern of false invoices or tampered invoice data based on the association between various data on the invoice, such as date, amount, name, etc. False invoices or tampered invoice data usually manifest as anomalies in certain invoice records, such as inconsistencies or abnormal fluctuations in invoice amount, date, invoice number and other data; therefore, association rule mining can be used to discover the associations between these abnormal invoice records, and then to mine clues of false invoices or tampered invoice data. The specific operation steps of the Apriori algorithm are as follows: 1) Mining frequent item sets to mine the features such as invoice number, date, amount, etc. that frequently appear in the invoices of the enterprise to be evaluated; 2) Mining association rules to mine the association between a certain invoice number and date, or the association between a certain date and amount, etc.; 3) Anomaly detection, through K-means clustering (i.e., K-means clustering algorithm), detect abnormal fluctuations between a certain invoice number and date, or detect abnormal fluctuations between a certain date and amount; the detected abnormal fluctuations can be used to determine whether the acquired invoice key information is true. For example, if the detected abnormal fluctuation index is lower than the fluctuation threshold, it can be preliminarily considered that the authenticity of the invoice key information has been verified. If the detected abnormal fluctuation index is not lower than the fluctuation threshold, it can be preliminarily considered that the authenticity of the invoice key information has not been verified.

[0059] Step S206, integrating the key invoice information of the multiple invoice images to obtain derived variables, wherein the derived variables are used to reflect the financial health of the target customer within a preset historical time period, and the financial health is an evaluation parameter for credit risk assessment.

[0060] like Figure 3 As shown in the figure, in order to better obtain the changing trend of the target customers' operating conditions during this period, derived variables can be processed based on the basic invoice data, as follows: obtain the customer's invoice information for the past year (or longer) through API docking, bank-tax interaction or RPA, and process the following derived variables: (1) variables measuring the company's operating income, such as the total sales revenue of the invoicing company in the past 12 months, the proportion of sales revenue in the past three months, etc.; (2) variables measuring the company's operating stability, such as the number of consecutive invoicing months of the invoicing company, the number of low-income invoicing months in the past 12 months, and the number of consecutive days without invoicing; (3) variables measuring the company's upstream and downstream customers, such as whether the industry in which the invoice downstream company is located is a high-risk industry, the proportion of transaction amounts of customers in high-risk industries, etc.; (4) variables measuring whether the company's invoicing behavior is normal, such as the proportion of red and invalid invoices in the past 12 months, the proportion of invoices with large deviations between the invoice amount and the average invoice amount, etc. There are 132 derived variable characteristics in these four categories.

[0061] Step S208, input the derived variables into the risk assessment model, so as to process the derived variables using the risk assessment model to obtain a risk assessment score for credit risk assessment of the target customer, wherein the risk assessment model is trained by machine learning using multiple groups of first training data, and each of the multiple groups of first training data includes: sample derived variables and sample risk assessment scores corresponding to the sample derived variables.

[0062] In this embodiment, the generated derivative variables can be input into the risk assessment model to analyze and process them using the risk assessment model to obtain a risk assessment score for credit risk assessment of the target customer, so as to preliminarily determine whether the target customer has risks.

[0063] According to an embodiment of the present invention, in the above-mentioned step S208, the risk assessment model includes a first risk assessment model and a second risk assessment model, and the derived variables are input into the risk assessment model so as to process the derived variables using the risk assessment model to obtain a risk assessment score for credit risk assessment of the target customer, including: performing risk analysis on the derived variables to obtain the correlation between the derived variables and the credit risk; determining the derived variables whose correlation is greater than the correlation threshold as the first derived variables, and determining the derived variables whose correlation is not greater than the correlation threshold as the second derived variables; using the first risk assessment model to determine the first risk assessment score obtained by performing credit risk assessment on the first derived variable, and using the second risk assessment model to determine the second risk assessment score obtained by performing credit risk assessment on the second derived variable; weighted fusion of the first risk assessment score and the second risk assessment score to obtain a risk assessment score.

[0064] Specifically, in order to conduct a more comprehensive risk assessment of target customers, the derived variable features obtained can be divided into features with strong correlation and features with weak correlation according to the correlation between the derived variables and the credit risk, and then analyzed and processed using different models to obtain corresponding scores. Finally, the output results of the two models (i.e., the scores) are combined to obtain a risk assessment score for the credit risk assessment of the target customer; for example, the credit scoring model (i.e., the first risk assessment model mentioned above) can be used to analyze and process the derived variable features with strong correlation to obtain a first risk assessment score, and the risk target customer machine learning early warning model (i.e., the second risk assessment model) can be used to analyze and process the derived variable features with weak correlation to obtain a second risk assessment score, and then the first risk assessment score and the second risk assessment score can be weighted and fused to obtain a risk assessment score.

[0065] Combine the following Figure 6 The above embodiments of the present invention are further described in detail. Figure 6 is a flowchart of modeling a credit scoring model and a risk target customer machine learning early warning model according to an embodiment of the present invention. Figure 6 As shown in the figure, the specific process of constructing the credit scoring model and the Catboost machine learning model for early warning of risky micro-enterprises is as follows: 1) First calculate the correlation between the 132 derived variable features obtained above and the credit risk; 2) Select 32 derived variable features with high IV values ​​and process them into a credit scoring model. The weights between different rules are determined based on the recall rate and precision calculated according to actual experiments; 3) Use the remaining derived variable features to train the Catboost machine learning model for risk target customer classification to obtain the risk score of the target customer to be evaluated; 4) Weightedly combine the output results of the two models to obtain the risk assessment score.

[0066] In a specific embodiment of the present invention, risk analysis is performed on the derived variable to obtain the correlation between the derived variable and the credit risk, including: obtaining the variable characteristics of the derived variable, where the variable characteristics are used to analyze the correlation between the derived variable and the credit risk; calculating the correlation between the derived variable and the credit risk according to the variable characteristics using the first formula, where the first formula is: i represents the label of the bin interval in the variable characteristics, N represents the total number of bin intervals in the variable characteristics, Bad s represents the number of low-risk customers with a risk level not higher than the risk level threshold in the i-th bin interval, Bad t represents the total number of low-risk customers with a risk level not higher than the risk level threshold in all bin intervals, Good s represents the number of high-risk customers with a risk level higher than the risk level threshold in the i-th bin interval, Good t represents the total number of high-risk customers with a risk level lower than the risk level threshold in all bin intervals, WOE i represents the difference between the risk status of the i-th bin interval and the total risk status of all bin intervals, and the risk status is the status analyzed based on the proportion of the number of low-risk customers and high-risk customer data.

[0067] Specifically, when calculating the correlation between the derived variable and the credit risk, the formula: can be used for calculation. In the formula, i represents the label of the bin interval in the variable characteristics, N represents the total number of bin intervals in the variable characteristics, Bad s represents the number of low-risk customers with a risk level not higher than the risk level threshold in the i-th bin interval, Bad t represents the total number of low-risk customers with a risk level not higher than the risk level threshold in all bin intervals, Good s represents the number of high-risk customers with a risk level higher than the risk level threshold in the i-th bin interval, Good t represents the total number of high-risk customers with a risk level lower than the risk level threshold in all bin intervals, WOE i represents the difference between the risk status of the i-th bin interval and the total risk status of all bin intervals, and the risk status is the status analyzed based on the proportion of the number of low-risk customers and high-risk customer data.

[0068] Step S210, determining the evaluation result of the credit risk assessment for the target customer according to the risk threshold range where the risk assessment score is located.

[0069] Optionally, the risk threshold range includes: a first risk threshold range, a second risk threshold range and a third risk threshold range, and the assessment result of the credit risk assessment of the target customer is determined according to the risk threshold range in which the risk assessment score is located, including: when the risk assessment score is in the first risk threshold range, determining the assessment result as low risk, wherein the first risk threshold range is a range in which the risk assessment score is less than the first risk threshold; when the risk assessment score is in the second risk threshold range, determining the assessment result as medium risk, wherein the second risk threshold range is a range in which the risk assessment score is greater than or equal to the first risk threshold and less than the second risk threshold, and the second risk threshold is greater than the first risk threshold; when the risk assessment score is in the third risk threshold range, determining the assessment result as high risk, wherein the third risk threshold range is a range in which the risk assessment score is greater than or equal to the second risk threshold.

[0070] As can be seen from the above, through the technical solution provided by the above embodiment of the present invention, multiple invoice images of target customers within a preset historical time period can be obtained, wherein the target customers are the objects that need to be evaluated for credit risk; text recognition and information extraction are performed on each invoice image respectively to obtain the key invoice information in each invoice image, wherein the key invoice information refers to the invoice information whose influence on the credit risk evaluation is higher than the influence threshold; the key invoice information of multiple invoice images is integrated to obtain derived variables, wherein the derived variables are used to reflect the financial health of the target customers within the preset historical time period, and the financial health is an evaluation parameter for credit risk evaluation; the derived variables are input into the risk assessment model to use the risk assessment model to evaluate the derived variables Processing is performed to obtain a risk assessment score for credit risk assessment of the target customer, wherein the risk assessment model is trained by machine learning using multiple groups of first training data, and each of the multiple groups of first training data includes: sample derived variables, and sample risk assessment scores corresponding to the sample derived variables; the assessment result of the credit risk assessment of the target customer is determined according to the risk threshold range of the risk assessment score, thereby achieving the purpose of extracting key information from the invoice image and processing it into derived variables, so as to use the corresponding risk assessment model to conduct credit risk assessment on the target customer based on the derived variables, realizing the technical effect of using the model to process the invoice image to analyze the credit risk of the target customer, and improving the accuracy and effectiveness of the assessment.

[0071] Therefore, the technical solution provided by the above-mentioned embodiment of the present invention solves the technical problem in the related art that credit risk assessment of customers mainly relies on manual verification, the authenticity and effectiveness cannot be guaranteed, and the assessment results are inaccurate.

[0072] By applying the technical solutions provided in the above embodiments of the present invention, the following technical effects can be achieved: 1) By building an invoice data set and an invoice detection and recognition algorithm, it is possible to monitor the full-process income fluctuations and operating cash flow conditions of target customers during the pre-loan, in-loan, and post-loan processes using enterprise invoice data with high authenticity, easy availability, and strong real-time performance while ensuring data security; 2) In the process of identifying the authenticity of enterprise invoice information, using the key invoice information extracted by the self-built invoice detection and recognition model and applying the Apriori algorithm to mine the associated information between data to discover potential false invoices or tampered invoices, making the evaluation result of the repayment ability of target customers more credible; 3) In the process of evaluating the repayment ability and loan risk of target customers, based on the original invoice data, four types of derivative features are further generated, a credit scoring model is developed, and a machine learning early warning model is used to score the risk level of target customers, enabling automatic risk warning.

[0073] It should be noted that for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that this application is not limited by the described action sequence, because according to this application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0074] Through the description of the above embodiments, those skilled in the art can clearly understand that the method according to the above embodiments can be implemented by means of software plus a necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of this application, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions for causing a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0075] According to an embodiment of the present invention, there is also provided a credit risk assessment device based on invoice data for implementing the above credit risk assessment method based on invoice data. Figure 7 It is a schematic diagram of the credit risk assessment device based on invoice data according to an embodiment of the present invention, as Figure 7 shown. The device includes: a first acquisition unit 71, a second acquisition unit 73, a third acquisition unit 75, a fourth acquisition unit 77, and a first determination unit 79. The credit risk assessment device based on invoice data will be described in detail below.

[0076] The first acquisition unit 71 is used to acquire a plurality of invoice images of a target customer within a preset historical time period, wherein the target customer is a subject for which a credit risk assessment is required.

[0077] The second acquisition unit 73 is used to perform text recognition and information extraction on each invoice image to obtain the key invoice information in each invoice image, wherein the key invoice information refers to the invoice information whose impact on credit risk assessment is higher than the impact threshold.

[0078] The third acquisition unit 75 is used to integrate the key invoice information of multiple invoice images to obtain derived variables, wherein the derived variables are used to reflect the financial health of the target customer within a preset historical time period, and the financial health is an evaluation parameter for credit risk evaluation.

[0079] The fourth acquisition unit 77 is used to input the derived variables into the risk assessment model so as to process the derived variables using the risk assessment model to obtain a risk assessment score for credit risk assessment of the target customer, wherein the risk assessment model is trained using multiple groups of first training data through machine learning, and each of the multiple groups of first training data includes: sample derived variables and sample risk assessment scores corresponding to the sample derived variables.

[0080] The first determining unit 79 is used to determine the assessment result of the credit risk assessment of the target customer according to the risk threshold range of the risk assessment score.

[0081] It should be noted here that the above-mentioned first acquisition unit 71, second acquisition unit 73, third acquisition unit 75, fourth acquisition unit 77 and first determination unit 79 correspond to steps S202 to S210 in the above-mentioned embodiments, and the five units are the same as the instances and application scenarios implemented by the corresponding steps, but are not limited to the contents disclosed in the above-mentioned embodiments.

[0082] As can be seen from the above, in the scheme recorded in the above-mentioned embodiment of the present invention, the first acquisition unit can be used to first acquire multiple invoice images of the target customer within a preset historical time period, wherein the target customer is the object that needs to be evaluated for credit risk; then the second acquisition unit can be used to perform text recognition and information extraction on each invoice image respectively to obtain the key invoice information in each invoice image, wherein the key invoice information refers to the invoice information whose influence on the credit risk evaluation is higher than the influence threshold; then the third acquisition unit is used to integrate the key invoice information of the multiple invoice images to obtain derived variables, wherein the derived variables are used to reflect the financial health of the target customer within the preset historical time period, and the financial health is an evaluation parameter for credit risk evaluation; then the fourth acquisition unit is used to input the derived variables into the risk assessment model to use The risk assessment model processes the derived variables to obtain a risk assessment score for credit risk assessment of the target customer, wherein the risk assessment model is trained by machine learning using multiple groups of first training data, and each of the multiple groups of first training data includes: sample derived variables, and sample risk assessment scores corresponding to the sample derived variables; finally, the first determination unit is used to determine the assessment result of the credit risk assessment of the target customer according to the risk threshold range of the risk assessment score, thereby achieving the purpose of extracting key information from the invoice image and processing it into derived variables, so as to use the corresponding risk assessment model to conduct credit risk assessment on the target customer based on the derived variables, realizing the technical effect of using the model to process the invoice image to analyze the credit risk of the target customer, and improving the accuracy and effectiveness of the assessment.

[0083] Therefore, the technical solution provided by the above-mentioned embodiment of the present invention solves the technical problem in the related art that credit risk assessment of customers mainly relies on manual verification, the authenticity and effectiveness cannot be guaranteed, and the assessment results are inaccurate.

[0084] In an optional embodiment, the first acquisition unit includes: a first acquisition module, which is used to acquire multiple first sub-invoice images of the target customer within a preset historical time period based on a first acquisition channel, wherein the first acquisition channel refers to a channel for directly exchanging data with a target invoice providing platform to acquire invoice images, and the target invoice providing platform refers to a platform that provides invoice images of the target customer; a second acquisition module, which is used to acquire multiple second sub-invoice images of the target customer within a preset historical time period based on a second acquisition channel, wherein the second acquisition channel refers to a channel for exchanging data with the target invoice providing platform through a third-party platform calling an API interface to acquire invoice images, and the third-party platform refers to an invoice providing platform other than the target invoice providing platform; a third acquisition module, which is used to acquire multiple third sub-invoice images of the target customer within a preset historical time period based on a third acquisition channel, wherein the third acquisition channel refers to a channel for directly logging into the target invoice providing platform by calling a preset program to acquire invoice images; and a first determination module, which is used to determine that the first sub-invoice image, the second sub-invoice image and the third sub-invoice image are invoice images of the target customer within a preset historical time period.

[0085] In an optional embodiment, the second acquisition unit includes: a fourth acquisition module, which is used to perform image preprocessing on multiple invoice images in sequence using image processing technology to obtain target invoice images corresponding to each invoice image; a second determination module, which is used to determine the region type of each pixel region in each target invoice image using an invoice text positioning model, wherein the invoice text positioning model is trained by machine learning using multiple sets of second training data, and each of the multiple sets of second training data includes: a sample target invoice image, and a sample region type corresponding to the sample target invoice image; a third determination module, which is used to determine that a pixel region whose region type is a text region is a target pixel region; and a fifth acquisition module, which is used to input the target pixel region in each target invoice image into the invoice text recognition model in sequence, so as to perform text recognition on the target pixel region using the invoice text recognition model to obtain the invoice key information in each target invoice image, wherein the invoice text recognition model is trained by machine learning using multiple sets of third training data, and each of the third training data includes: a sample target pixel region, and sample invoice key information corresponding to the sample target pixel region.

[0086] In an optional embodiment, the credit risk assessment device based on invoice data also includes: a fifth acquisition unit, which is used to integrate the invoice key information in multiple invoice images to obtain an invoice key information set after performing text recognition and information extraction on each invoice image respectively to obtain the invoice key information in each invoice image; a second determination unit, which is used to determine that the feature items in the invoice key information set whose frequency of appearance is higher than the frequency threshold are target feature items, wherein the feature items are field names in the invoice key information set; a sixth acquisition unit, which is used to analyze the target feature items using a priori algorithms to obtain feature association relationships between any two target feature items or any multiple target feature items; a seventh acquisition unit, which is used to analyze the feature association relationships using a K-means clustering algorithm to obtain an abnormal fluctuation index between the target feature items; and a third determination unit, which is used to determine that the authenticity of the invoice key information has passed the verification when the abnormal fluctuation index is lower than the fluctuation threshold.

[0087] In an optional embodiment, the risk assessment model includes a first risk assessment model and a second risk assessment model, and the fourth acquisition unit includes: a sixth acquisition module, which is used to perform risk analysis on the derivative variables to obtain the correlation between the derivative variables and the credit risk; a fourth determination module, which is used to determine the derivative variable whose correlation is greater than the correlation threshold as the first derivative variable, and determine the derivative variable whose correlation is not greater than the correlation threshold as the second derivative variable; a seventh acquisition module, which is used to use the first risk assessment model to determine the first risk assessment score obtained by performing credit risk assessment on the first derivative variable, and use the second risk assessment model to determine the second risk assessment score obtained by performing credit risk assessment on the second derivative variable; an eighth acquisition module, which is used to weightedly fuse the first risk assessment score and the second risk assessment score to obtain a risk assessment score.

[0088] In an optional embodiment, the sixth acquisition module includes: an acquisition submodule, used to acquire variable characteristics of the derived variables, wherein the variable characteristics are used to analyze the correlation between the derived variables and the credit risk; a calculation submodule, used to calculate the correlation between the derived variables and the credit risk using a first formula according to the variable characteristics, wherein the first formula is: i represents the label of the binning interval in the variable feature, N represents the total number of binning intervals in the variable feature, Bad s Indicates the number of low-risk customers whose risk level is not higher than the risk threshold in the i-th bin interval, Bad t Indicates the total number of low-risk customers whose risk level is not higher than the risk threshold in all bin intervals, Good s Good represents the number of high-risk customers whose risk level is higher than the risk threshold in the i-th bin interval. t Indicates the total number of high-risk customers whose risk level is below the risk threshold in all bin intervals, WOEi It represents the difference between the risk status of the ith bin interval and the total risk status of all bin intervals. The risk status is obtained by analyzing the number of low-risk customers and the proportion of high-risk customer data.

[0089] In an optional embodiment, the risk threshold range includes: a first risk threshold range, a second risk threshold range and a third risk threshold range, and the first determination unit includes: a fifth determination module, which is used to determine that the assessment result is low risk when the risk assessment score is within the first risk threshold range, wherein the first risk threshold range is a range in which the risk assessment score is less than the first risk threshold; a sixth determination module, which is used to determine that the assessment result is medium risk when the risk assessment score is within the second risk threshold range, wherein the second risk threshold range is a range in which the risk assessment score is greater than or equal to the first risk threshold and less than the second risk threshold, and the second risk threshold is greater than the first risk threshold; a seventh determination module, which is used to determine that the assessment result is high risk when the risk assessment score is within the third risk threshold range, wherein the third risk threshold range is a range in which the risk assessment score is greater than or equal to the second risk threshold.

[0090] According to another aspect of an embodiment of the present invention, a credit risk assessment system based on invoice data is further provided. The credit risk assessment system based on invoice data uses any of the above-mentioned credit risk assessment methods based on invoice data.

[0091] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is further provided, the computer-readable storage medium comprising a stored program, wherein the program executes any one of the above-mentioned credit risk assessment methods based on invoice data.

[0092] Optionally, in this embodiment, the computer-readable storage medium may be located in any one of the computer terminals in a computer terminal group in a computer network, or in any one of the communication devices in a communication device group.

[0093] Optionally, in this embodiment, the computer-readable storage medium is configured to store program codes for executing the following steps: obtaining multiple invoice images of a target customer within a preset historical time period, wherein the target customer is an object that needs to be subject to credit risk assessment; performing text recognition and information extraction on each invoice image respectively to obtain key invoice information in each invoice image, wherein the key invoice information refers to invoice information whose impact on credit risk assessment is higher than an impact threshold; integrating the key invoice information of multiple invoice images to obtain derived variables, wherein the derived variables are used to reflect the financial health of the target customer within a preset historical time period, and the financial health is an assessment parameter for credit risk assessment; inputting the derived variables into a risk assessment model to process the derived variables using the risk assessment model to obtain a risk assessment score for credit risk assessment of the target customer, wherein the risk assessment model is trained using multiple sets of first training data by machine learning, and each of the multiple sets of first training data includes: sample derived variables and sample risk assessment scores corresponding to the sample derived variables; and determining an assessment result of credit risk assessment of the target customer according to the risk threshold range of the risk assessment score.

[0094] Optionally, in this embodiment, the computer-readable storage medium is configured to store program codes for executing the following steps: obtaining multiple first sub-invoice images of the target customer within a preset historical time period based on a first acquisition channel, wherein the first acquisition channel refers to a channel for directly interacting with the target invoice providing platform for data to obtain invoice images, and the target invoice providing platform refers to a platform that provides invoice images of the target customer; obtaining multiple second sub-invoice images of the target customer within a preset historical time period based on a second acquisition channel, wherein the second acquisition channel refers to a channel for interacting with the target invoice providing platform for data to obtain invoice images through a third-party platform calling an API interface, and the third-party platform refers to an invoice providing platform other than the target invoice providing platform; obtaining multiple third sub-invoice images of the target customer within a preset historical time period based on a third acquisition channel, wherein the third acquisition channel refers to a channel for directly logging into the target invoice providing platform by calling a preset program to obtain invoice images; determining the first sub-invoice image, the second sub-invoice image and the third sub-invoice image as invoice images of the target customer within the preset historical time period.

[0095] Optionally, in this embodiment, the computer-readable storage medium is configured to store program codes for executing the following steps: using image processing technology to perform image preprocessing on multiple invoice images in sequence to obtain target invoice images corresponding to each invoice image; using an invoice text positioning model to determine the region type of each pixel region in each target invoice image, wherein the invoice text positioning model is trained using multiple sets of second training data by machine learning, and each of the multiple sets of second training data includes: sample target invoice images, sample region types corresponding to the sample target invoice images; determining that the pixel region whose region type is a text region is the target pixel region; sequentially inputting the target pixel region in each target invoice image into the invoice text recognition model to perform text recognition on the target pixel region using the invoice text recognition model to obtain the invoice key information in each target invoice image, wherein the invoice text recognition model is trained using multiple sets of third training data by machine learning, and each of the third training data includes: sample target pixel region, sample invoice key information corresponding to the sample target pixel region.

[0096] Optionally, in this embodiment, the computer-readable storage medium is configured to store program codes for executing the following steps: integrating the invoice key information in multiple invoice images to obtain an invoice key information set; determining that the feature items in the invoice key information set whose frequency of appearance is higher than a frequency threshold are target feature items, wherein the feature items are field names in the invoice key information set; analyzing the target feature items using a priori algorithms to obtain feature association relationships between any two target feature items or any multiple target feature items; analyzing the feature association relationships using a K-means clustering algorithm to obtain an abnormal fluctuation index between the target feature items; and determining that the authenticity of the invoice key information has passed verification when the abnormal fluctuation index is lower than the fluctuation threshold.

[0097] Optionally, in this embodiment, the computer-readable storage medium is configured to store program codes for executing the following steps: performing risk analysis on the derived variables to obtain the correlation between the derived variables and the credit risk; determining the derived variable whose correlation is greater than the correlation threshold as the first derived variable, and determining the derived variable whose correlation is not greater than the correlation threshold as the second derived variable; using the first risk assessment model to determine a first risk assessment score obtained by performing credit risk assessment on the first derived variable, and using the second risk assessment model to determine a second risk assessment score obtained by performing credit risk assessment on the second derived variable; and weightedly fusing the first risk assessment score and the second risk assessment score to obtain a risk assessment score.

[0098] Optionally, in this embodiment, the computer-readable storage medium is configured to store program codes for executing the following steps: obtaining variable characteristics of the derived variables, wherein the variable characteristics are used to analyze the correlation between the derived variables and the credit risk; and calculating the correlation between the derived variables and the credit risk using a first formula according to the variable characteristics, wherein the first formula is: i represents the label of the binning interval in the variable feature, N represents the total number of binning intervals in the variable feature, Bad s Indicates the number of low-risk customers whose risk level is not higher than the risk threshold in the i-th bin interval, Bad t Indicates the total number of low-risk customers whose risk level is not higher than the risk threshold in all bin intervals, Good s Good represents the number of high-risk customers whose risk level is higher than the risk threshold in the i-th bin interval. t Indicates the total number of high-risk customers whose risk level is below the risk threshold in all bin intervals, WOE i It represents the difference between the risk status of the ith bin interval and the total risk status of all bin intervals. The risk status is obtained by analyzing the number of low-risk customers and the proportion of high-risk customer data.

[0099] Optionally, in this embodiment, the computer-readable storage medium is configured to store program code for executing the following steps: when the risk assessment score is within a first risk threshold range, determining that the assessment result is low risk, wherein the first risk threshold range is a range in which the risk assessment score is less than the first risk threshold; when the risk assessment score is within a second risk threshold range, determining that the assessment result is medium risk, wherein the second risk threshold range is a range in which the risk assessment score is greater than or equal to the first risk threshold and less than the second risk threshold, and the second risk threshold is greater than the first risk threshold; when the risk assessment score is within a third risk threshold range, determining that the assessment result is high risk, wherein the third risk threshold range is a range in which the risk assessment score is greater than or equal to the second risk threshold.

[0100] According to another aspect of an embodiment of the present invention, a processor is further provided, and the processor is used to run a program, wherein when the program is run, any one of the above-mentioned credit risk assessment methods based on invoice data is executed.

[0101] According to another aspect of an embodiment of the present invention, a computer program product is provided, comprising computer instructions, which, when executed by a processor, execute any one of the above-mentioned credit risk assessment methods based on invoice data.

[0102] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0103] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0104] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0105] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0106] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0107] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for 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: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0108] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A credit risk assessment method based on invoice data, characterized in that: include: Acquire multiple invoice images of a target customer within a preset historical time period, wherein the target customer is a subject for which a credit risk assessment is required; Performing text recognition and information extraction on each of the invoice images respectively to obtain key invoice information in each of the invoice images, wherein the key invoice information refers to invoice information whose impact on credit risk assessment is higher than an impact threshold; Integrate the key invoice information of the plurality of invoice images to obtain a derived variable, wherein the derived variable is used to reflect the financial health of the target customer within the preset historical time period, and the financial health is an evaluation parameter for credit risk assessment; Inputting the derived variables into a risk assessment model, so as to process the derived variables using the risk assessment model to obtain a risk assessment score for credit risk assessment of the target customer, wherein the risk assessment model is obtained by training using multiple sets of first training data by machine learning, and each of the multiple sets of first training data includes: a sample derived variable and a sample risk assessment score corresponding to the sample derived variable; An assessment result of a credit risk assessment on the target customer is determined according to a risk threshold range within which the risk assessment score falls.

2. The credit risk assessment method based on invoice data according to claim 1, characterized in that: Get multiple invoice images of the target customer within a preset historical time period, including: Acquire multiple first sub-invoice images of the target customer within the preset historical time period based on a first acquisition channel, wherein the first acquisition channel refers to a channel for directly interacting with a target invoice providing platform to acquire the invoice images, and the target invoice providing platform refers to a platform that provides the invoice images of the target customer; Acquire multiple second sub-invoice images of the target customer within the preset historical time period based on a second acquisition channel, wherein the second acquisition channel refers to a channel for acquiring the invoice images by exchanging data with the target invoice providing platform through a third-party platform calling an API interface, and the third-party platform refers to an invoice providing platform other than the target invoice providing platform; Acquire multiple third sub-invoice images of the target customer within the preset historical time period based on a third acquisition channel, wherein the third acquisition channel refers to a channel for directly logging into the target invoice providing platform to acquire the invoice images by calling a preset program; The first sub-invoice image, the second sub-invoice image and the third sub-invoice image are determined to be the invoice images of the target customer within the preset historical time period.

3. The credit risk assessment method based on invoice data according to claim 1, characterized in that: Perform text recognition and information extraction on each of the invoice images to obtain key invoice information in each of the invoice images, including: Using image processing technology to perform image preprocessing on the plurality of invoice images in sequence to obtain a target invoice image corresponding to each invoice image; Determine the region type of each pixel region in each of the target invoice images using an invoice text positioning model, wherein the invoice text positioning model is trained by machine learning using multiple sets of second training data, each of the multiple sets of second training data including: a sample target invoice image, and a sample region type corresponding to the sample target invoice image; Determine the pixel region whose region type is a text region as a target pixel region; The target pixel area in each of the target invoice images is input into the invoice text recognition model in turn, so as to perform text recognition on the target pixel area using the invoice text recognition model to obtain the invoice key information in each of the target invoice images, wherein the invoice text recognition model is trained by machine learning using multiple groups of third training data, and each group of the third training data includes: a sample target pixel area and sample invoice key information corresponding to the sample target pixel area.

4. The credit risk assessment method based on invoice data according to claim 1, characterized in that: After performing text recognition and information extraction on each invoice image to obtain the key invoice information in each invoice image, the method further includes: Integrate the key invoice information in the plurality of invoice images to obtain a key invoice information set; Determine a feature item in the invoice key information set whose occurrence frequency is higher than a frequency threshold as a target feature item, wherein the feature item is a field name in the invoice key information set; Analyzing the target feature items by using a priori algorithms to obtain a feature association relationship between any two target feature items or any multiple target feature items; The feature association relationship is analyzed using a K-means clustering algorithm to obtain an abnormal fluctuation index between the target feature items; When the abnormal fluctuation index is lower than the fluctuation threshold, it is determined that the authenticity of the key invoice information has passed the verification.

5. The credit risk assessment method based on invoice data according to claim 1, characterized in that: The risk assessment model includes a first risk assessment model and a second risk assessment model. The derived variables are input into the risk assessment model to process the derived variables using the risk assessment model to obtain a risk assessment score for credit risk assessment of the target customer, including: Performing risk analysis on the derivative variables to obtain the correlation between the derivative variables and the credit risk; Determine the derived variable whose correlation is greater than a correlation threshold as a first derived variable, and determine the derived variable whose correlation is not greater than the correlation threshold as a second derived variable; Determine a first risk assessment score obtained by performing a credit risk assessment on the first derivative variable using the first risk assessment model, and determine a second risk assessment score obtained by performing a credit risk assessment on the second derivative variable using the second risk assessment model; The first risk assessment score and the second risk assessment score are weightedly fused to obtain the risk assessment score.

6. The credit risk assessment method based on invoice data according to claim 5, characterized in that: Performing risk analysis on the derivative variables to obtain the correlation between the derivative variables and the credit risk, including: Acquiring variable characteristics of the derived variable, wherein the variable characteristics are used to analyze the correlation between the derived variable and the credit risk; The correlation between the derived variable and the credit risk is calculated using a first formula according to the variable characteristics, wherein the first formula is: i represents the number of the binning interval in the variable feature, N represents the total number of binning intervals in the variable feature, Bad s Indicates the number of low-risk customers whose risk level is not higher than the risk threshold in the i-th bin interval, Bad t Indicates the total number of low-risk customers whose risk level is not higher than the risk level threshold in all the binning intervals, Good s represents the number of high-risk customers whose risk level in the i-th bin interval is higher than the risk level threshold, Good t Indicates the total number of high-risk customers whose risk level is lower than the risk level threshold in all the binning intervals, WOE i It represents the difference between the risk status of the ith sub-box interval and the total risk status of all the sub-box intervals, and the risk status is obtained by analyzing the number of low-risk customers and the proportion of high-risk customer data.

7. The credit risk assessment method based on invoice data according to claim 1, characterized in that: The risk threshold range includes: a first risk threshold range, a second risk threshold range and a third risk threshold range. The assessment result of the credit risk assessment of the target customer is determined according to the risk threshold range in which the risk assessment score is located, including: In the case where the risk assessment score is within the first risk threshold range, determining that the assessment result is low risk, wherein the first risk threshold range is a range in which the risk assessment score is less than the first risk threshold; When the risk assessment score is within the second risk threshold range, determining that the assessment result is medium risk, wherein the second risk threshold range is a range in which the risk assessment score is greater than or equal to the first risk threshold and less than a second risk threshold, and the second risk threshold is greater than the first risk threshold; When the risk assessment score is within the third risk threshold range, the assessment result is determined to be high risk, wherein the third risk threshold range is a range in which the risk assessment score is greater than or equal to the second risk threshold.

8. A credit risk assessment device based on invoice data, characterized in that: include: A first acquisition unit is used to acquire multiple invoice images of a target customer within a preset historical time period, wherein the target customer is a subject for which a credit risk assessment is required; A second acquisition unit is used to perform text recognition and information extraction on each of the invoice images to obtain key invoice information in each of the invoice images, wherein the key invoice information refers to invoice information whose impact on credit risk assessment is higher than an impact threshold; A third acquisition unit is used to integrate the invoice key information of the multiple invoice images to obtain a derived variable, wherein the derived variable is used to reflect the financial health of the target customer in the preset historical time period, and the financial health is an evaluation parameter for credit risk evaluation; a fourth acquisition unit, configured to input the derived variables into a risk assessment model, so as to process the derived variables using the risk assessment model to obtain a risk assessment score for credit risk assessment of the target customer, wherein the risk assessment model is obtained by training using a plurality of sets of first training data in a machine learning manner, and each of the plurality of sets of first training data includes: a sample derived variable and a sample risk assessment score corresponding to the sample derived variable; The first determining unit is used to determine an assessment result of the credit risk assessment of the target customer according to the risk threshold range of the risk assessment score.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein the program executes the credit risk assessment method based on invoice data as described in any one of claims 1 to 7.

10. A computer program product comprising computer instructions, characterized in that: When the computer instructions are executed by a processor, the credit risk assessment method based on invoice data described in any one of claims 1 to 7 is performed.

Images