Financial institution electronic account bill identification method and system

Abstract

The invention belongs to the technical field of electronic account lists of financial institutions. A financial institution electronic account bill identification method comprises the following steps: obtaining a business order request; matching the business order request with a historical task order to obtain a matching result; generating an exclusive mailbox according to the matching result; acquiring an electronic flow account list sent to the exclusive mailbox, and identifying the electronic flow account list by utilizing OCR (Optical Character Recognition) to obtain standardized structure data of the electronic flow account list; processing the standardized structure data of the electronic account bill by using a decision model to obtain a decision result message; and performing risk control auditing according to the decision result message to obtain a risk control auditing reference result. Uniform data standards and processing flows are established, and the efficiency and accuracy of data processing are improved. Big data and machine learning technologies are utilized to deeply analyze bill data, and potential risk points are found in time, so that the auditing quality and speed are improved.

Description

Technical Field

[0001] The present application belongs to the technical field of electronic statement of financial institutions, and in particular, relates to a method and system for identifying electronic statement of financial institutions. Background Art

[0002] Manual processing of electronic bills is common in current technology, but this method is vulnerable to bill tampering and forgery. Users may modify bills for various purposes, causing the data received by financial institutions to lose authenticity. Due to the lack of effective technical means to identify these tampering, financial institutions face greater risks during the audit process.

[0003] In addition, the electronic bill templates of different financial institutions vary greatly, which poses a challenge to the unified processing and analysis of data. The inconsistency of templates not only complicates automated processing, but also limits the ability to conduct in-depth analysis of data. Financial institutions need a standardized and structured data management system to be able to analyze and utilize this information more accurately. Summary of the invention

[0004] Based on this, it is necessary to provide a financial institution electronic statement recognition system and method to address the above technical issues.

[0005] In a first aspect, the present application provides a method for identifying an electronic statement of a financial institution, the method comprising:

[0006] Get business order request;

[0007] Matching the business order request with the historical task order to obtain a matching result;

[0008] Generate a dedicated mailbox according to the matching results;

[0009] Obtaining the electronic statement sent to the dedicated mailbox, identifying the electronic statement by using OCR recognition, and obtaining standardized structure data of the electronic statement;

[0010] Using the decision model, the standardized structure data of the electronic statement is processed to obtain a decision result message;

[0011] Conduct risk control review based on the decision result message to obtain a risk control review reference result.

[0012] In some practicable manners, the step of obtaining a business order request includes:

[0013] Generate a sharing page according to the business order request formed by the business request end, and present an operation guide on the sharing page;

[0014] The execution of the business order request is guided according to the operation guidance presented in the sharing page.

[0015] In some practicable manners, the step of matching the business order request with the historical task order to obtain a matching result includes:

[0016] Constructing a task order database, wherein the task order database stores a number of historical task orders generated according to the historical business order requests;

[0017] The business order request is matched with the historical task orders in the task order database to obtain a matching result.

[0018] In some practicable manners, the step of generating a dedicated mailbox according to the matching result includes:

[0019] If the matching result is successful, a dedicated mailbox for receiving electronic statement is generated, and the dedicated mailbox is sent to the service request end;

[0020] If the matching result fails, a request task order and a task order number are generated, a dedicated mailbox is generated according to the request business order, and the dedicated mailbox is sent to the business request end, and the task order number is associated with the request task order.

[0021] In some practicable methods, the step of obtaining the electronic statement sent to the dedicated mailbox, identifying the electronic statement by using OCR recognition, and obtaining the standardized structure data of the electronic statement includes:

[0022] Obtaining the electronic statement sent to the dedicated mailbox;

[0023] Using OCR recognition to adjust the image angle of the electronic statement to obtain a first electronic statement image, wherein the image angle adjustment means adjusting the electronic statement in four directions of 0°, 90°, 180° or 270°;

[0024] Performing color enhancement on the first electronic statement image to obtain a second electronic statement image, wherein the color enhancement represents the application of the generalization capability of the Gaussian noise ratio enhancement model and the random channel transformation enhancement model to enhance the recognition of different colors in the first electronic statement image;

[0025] Performing image preprocessing on the second electronic statement image to obtain a third electronic statement image, wherein the image preprocessing means converting the second electronic statement image into a grayscale image, and scaling the length and width in proportion to align the image to a uniform length and width dimension;

[0026] Performing image detection on the third electronic statement image to obtain a plurality of text regions, wherein the image detection means cutting the third electronic statement image to form the text regions, and outputting the four corner coordinates of the text regions and the confidence level;

[0027] Binarizing the plurality of text regions to obtain the plurality of binarized text regions;

[0028] Performing image recognition on the binarized text regions to obtain text information of the electronic statement, wherein the text information includes categories and corresponding statement information under the categories;

[0029] Mapping the text information of the electronic statement to the same category in a pre-configured electronic statement template to obtain a target electronic statement with statement information, wherein the pre-configured electronic statement template has the same category as the electronic statement and a filling area for the statement information under the category, and the unmapped category in the electronic statement template is processed by zero padding;

[0030] The standardized structure data of the electronic statement is obtained according to the target electronic statement with statement information.

[0031] In some practicable manners, the step of using the decision model to process the standardized structure data of the electronic statement to obtain a decision result message includes:

[0032] Extracting features from the standardized structure data of the electronic statement to obtain feature labels for the standardized structure data;

[0033] According to the feature labels of the standardized structure data, in the decision model, a derivative indicator library matching the feature labels is selected to obtain a target decision model, wherein the decision model pre-stores a number of comparable derivative indicator libraries for flow characteristic analysis;

[0034] The target decision model is used to compare and process the standardized structure data of the electronic statement to obtain the decision result message.

[0035] In some practicable manners, the step of performing risk control review according to the decision result message to obtain a reference result of the risk control review includes:

[0036] According to the preset risk rule engine of bank flow derivative indicators, the decision result message is matched and judged to obtain a judgment result, wherein the risk rule engine of bank flow derivative indicators at least includes a rule set consisting of income and loan discrepancy, abnormal transaction amount, high-risk transaction behavior, hit on the black and gray product list, and suspected fraud of bank flow, and the judgment result indicates the number of rules in the rule set that are hit;

[0037] According to the judgment result, a risk control review rating is performed to obtain a risk control review reference result, wherein the risk control review rating represents a grade division according to the number of rules in the rule set that are hit.

[0038] In a second aspect, the present application provides a financial institution electronic statement recognition system, which applies the aforementioned financial institution electronic statement recognition method, and the system includes:

[0039] An acquisition unit, used for acquiring a business order request;

[0040] A matching unit, used for matching the business order request with the historical task order to obtain a matching result;

[0041] A generating unit, configured to generate a dedicated mailbox according to the matching result;

[0042] The acquisition unit is further used to acquire the electronic statement sent to the dedicated mailbox, identify the electronic statement by using OCR recognition, and obtain the standardized structure data of the electronic statement;

[0043] A decision-making unit, used to process the standardized structure data of the electronic statement using a decision model to obtain a decision result message;

[0044] The result unit is used to perform risk control review according to the decision result message to obtain a risk control review reference result.

[0045] In a third aspect, the present application provides a computer storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the aforementioned method for identifying electronic statement of a financial institution.

[0046] In a fourth aspect, the present application provides a computer program, which, when executed by a processor, implements the steps of the aforementioned financial institution electronic statement identification method.

[0047] Beneficial Effects

[0048] The present application provides a method for identifying electronic bank statements of financial institutions. First, a business order request is obtained; then, the business order request is matched with the historical task order to obtain a matching result; based on the matching result, a dedicated mailbox is generated; the electronic bank statement sent to the dedicated mailbox is obtained, and the electronic bank statement is identified by OCR recognition to obtain the standardized structure data of the electronic bank statement; next, the standardized structure data of the electronic bank statement is processed by a decision model to obtain a decision result message; based on the decision result message, a risk control review is performed to obtain a risk control review reference result. Through the above method, a unified data standard and processing flow are established, which can improve the efficiency and accuracy of data processing. At the same time, big data and machine learning technology are used to conduct in-depth analysis of bill data, and potential risk points are discovered in a timely manner, thereby improving the quality and speed of the review. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the conventional technology, the drawings required for use in the embodiments or the conventional technology descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0050] Figure 1 The present invention is a flowchart of a method for identifying an electronic statement of a financial institution in one embodiment.

[0051] Figure 2 The present invention is a logic block diagram of a method for identifying an electronic statement of a financial institution in one embodiment. DETAILED DESCRIPTION

[0052] In order to facilitate understanding of the present application, the present application will be described more fully below with reference to the relevant drawings. Embodiments of the present application are provided in the drawings. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive.

[0053] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. The term "and / or" used herein includes any and all couplings of one or more related listed items.

[0054] It can be understood that the terms "first", "second", etc. used in the present application can be used in this article to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from another element.

[0055] Some terms involved in this application are explained below to facilitate understanding of this application:

[0056] Confidence is a statistical term used to express the degree of confidence in a statistical result or estimate.

[0057] OCR recognition: OCR recognition is the abbreviation of Optical Character Recognition, which is optical character recognition technology. This is a technology that converts the text on the document into a digital text format that can be edited and processed by a computer by scanning and recognizing the text on the document. This technology mainly relies on image processing and pattern recognition algorithms. It captures the character features in the document, such as strokes, shapes, sizes, spacing, etc., and compares them with the preset character library to recognize the corresponding text information.

[0058] Gaussian Noise Augmentation Model is a data augmentation technology commonly used in machine learning and deep learning to improve the generalization ability and robustness of the model. The core idea is to add Gaussian noise to the original data to generate new training samples, so that the model can better adapt to and cope with noisy or uncertain data.

[0059] Random Channel Transformation Augmentation Model is a technique for image data augmentation, especially for computer vision tasks. It generates diverse training samples by randomly changing the color channels of the image, thereby improving the model's generalization ability and adaptability to different environmental conditions.

[0060] Binarization is a basic technique in image processing and computer vision that is used to convert an image into a form containing only two colors (usually black and white). This process is done by comparing the pixel values ​​of the image with a certain threshold. If the pixel value is higher than the threshold, it is set to one color (usually white); otherwise, it is set to another color (usually black).

[0061] Convolutional Neural Network (CNN) is a deep learning model that is particularly suitable for processing data with grid structures, such as images and videos. CNN achieves excellent performance in computer vision tasks such as image classification, object detection, and semantic segmentation by automatically learning feature representation.

[0062] Objective Decision Model is a mathematical model used for decision analysis and optimization. The model helps decision makers make the best decision when faced with multiple choices by setting specific goals or objective functions. Objective Decision Model is used to solve problems such as resource allocation, strategy formulation and complex system control.

[0063] The Derived Indicator Library is a database that collects and stores derived indicators generated by calculating, transforming, combining or analyzing basic data or original indicators. The purpose of the Derived Indicator Library is to provide richer and more in-depth data analysis results to support decision-making, business monitoring and strategic planning.

[0064] like Figure 1 and Figure 2 As shown, in the first aspect, the present application provides a method for identifying an electronic statement of a financial institution, the method comprising:

[0065] S100, obtaining a business order request.

[0066] Specifically, obtaining a business order request includes the following steps:

[0067] Generate a sharing page according to the business order request formed by the business request end, and present an operation guide on the sharing page;

[0068] The execution of the business order request is guided according to the operation guidance presented in the sharing page.

[0069] It should be noted that after forming a business order request using the business request end, an H5 sharing page can be generated. This sharing page is specially customized for users and is used to display operation guides related to business order requests. Among them, the H5 sharing page refers to a web page developed using HTML5 technology, which can be run on different devices and operating systems, including smartphones, tablets, laptops and desktop computers, etc. Users can access it without downloading specific applications. It supports rich interactive functions such as form submission, animation effects, video playback, etc., providing a richer user experience than traditional static web pages.

[0070] The sharing page will include clear instructions on how to submit electronic statements as required, including how to send the statement to a dedicated email address generated by the system, what information needs to be provided, and the format requirements of the statement.

[0071] The operation guide can be presented in the form of pictures and texts as needed, so that users can easily understand and follow the steps. This helps reduce the confusion and errors of users in the process of submitting electronic statements.

[0072] S200, matching the business order request with the historical task order to obtain a matching result.

[0073] Specifically, obtaining the matching result includes the following steps:

[0074] Constructing a task order database, wherein the task order database stores a number of historical task orders generated according to the historical business order requests;

[0075] The business order request is matched with the historical task orders in the task order database to obtain a matching result.

[0076] It should be noted that the task order database is used to store historical task orders generated according to historical business order requests. These historical task orders are instances of historical business order requests for specific business needs. Each historical task order contains detailed information required to complete the business order request. The historical task order represents the business type that can be processed and stores all necessary business parameters and conditions. When a user submits a business order request, the business order request will be received and parsed to extract key information such as user identity, business type, and specific content of the request. The current business order request is compared with the historical task orders in the task order database to find the most suitable task order to form a match. If the match is unsuccessful, it is necessary to create a task order for the current business order request. Several historical task orders stored in the task order database represent task orders generated by business order requests. However, due to non-compliance with the business order request, the task order is not completed. The historical task order formed can match the current business order request. If the match is successful, the historical task order can continue to be executed. If a matching historical task order is found in the task order database, a dedicated mailbox will be created for the user, and operation guidance will be provided to help the user complete the order request. A task order number can be constructed for each task order in the task order database. In this way, by matching the order number corresponding to the user's business order with the order number in the task order database, the corresponding task order can be called to obtain a matching result.

[0077] S300: Generate a dedicated mailbox according to the matching result.

[0078] Specifically, generating a dedicated mailbox includes the following steps:

[0079] If the matching result is successful, a dedicated mailbox for receiving electronic statement is generated, and the dedicated mailbox is sent to the service request end;

[0080] If the matching result fails, a request task order and a task order number are generated, a dedicated mailbox is generated according to the request business order, and the dedicated mailbox is sent to the business request end, and the task order number is associated with the request task order.

[0081] It should be noted that if the match is successful, it means that a suitable task order has been found to respond to the user's business order request. At this time, the following steps will be performed:

[0082] a. Generate a dedicated email address: Create a dedicated email address for the user to receive electronic statements related to the business order request.

[0083] b. Send exclusive email information: Send the address of this exclusive email to the business order requester, usually through the contact information or user interface provided by the user when making the request.

[0084] If the match fails, it means that there is no task order in the history that can directly meet the user's business order request. At this time, the following steps will be performed:

[0085] a. Generate Task Order: Create a new task order to record the user's business order request and prepare for subsequent processing.

[0086] b. Generate task order number: Assign a unique task order number to the newly created task order for tracking and management.

[0087] c. Generate a dedicated email address: Even if the match fails, a dedicated email address will still be generated for the user to receive possible electronic statement bills.

[0088] d. Send exclusive email information: Send the newly generated exclusive email address to the business order requester.

[0089] e. Associate the task ticket number: Associate the task ticket number with the newly created task order to ensure that all information submitted by the user can be correctly associated with their business request.

[0090] S400, obtaining the electronic statement sent to the dedicated mailbox, identifying the electronic statement by using OCR recognition, and obtaining standardized structure data of the electronic statement.

[0091] Specifically, obtaining the standardized structure data of the electronic statement includes the following steps:

[0092] Obtaining the electronic statement sent to the dedicated mailbox;

[0093] The electronic statement is image-angle adjusted by utilizing OCR recognition to obtain a first electronic statement image, wherein the image angle adjustment represents adjusting the electronic statement in four directions of 0°, 90°, 180° or 270°.

[0094] It should be noted that most electronic banknotes are rectangular, so the rotation direction of the electronic banknote can be adjusted in four directions: 0°, 90°, 180° or 270°, so as to ensure the consistency of the image direction. Exemplarily, the image angle adjustment can be performed by comparing the text or pattern in the electronic banknote with the preset text or pattern, that is, adjusting the angle of the electronic banknote in 0°, 90°, 180° or 270° in sequence until the text or pattern in the electronic banknote overlaps with the preset text or pattern, or most of it overlaps. Exemplarily, the preset text is "Liushi" and the pattern is an official seal or signature.

[0095] The first electronic statement image is color enhanced to obtain a second electronic statement image, wherein the color enhancement represents the generalization ability of applying the Gaussian noise ratio enhancement model and the random channel transformation enhancement model to enhance the recognition of different colors in the first electronic statement image.

[0096] It should be noted that the purpose of color enhancement is to improve the accuracy and robustness of the OCR system in recognizing the first electronic bill of lading image. By enhancing the color features of the image, different texts and backgrounds in the image can be better distinguished. In other words, the first electronic bill of lading image may contain various colors and backgrounds, which need to be processed to improve the recognition rate. The generalization ability of the Gaussian noise ratio enhancement model is applied to process the first electronic bill of lading image to enhance the color. In addition, the color can be further enhanced by combining with random channel transformation to obtain a second electronic bill of lading image with bright colors. Among them, the Gaussian noise enhancement model works by adding random noise that conforms to the Gaussian distribution to the first electronic bill of lading image (noise interference can be regarded as random changes in the image, which conforms to the Gaussian normal distribution). This enhancement can simulate various noise interferences that may be encountered in the image acquisition process in the real world, and help the Gaussian noise enhancement model learn to better process color features under different image conditions. By improving the color contrast, text and background can be more easily distinguished, which is convenient for improving the recognition accuracy of text in subsequent steps. In addition, when combined with random channel transformation (such as the red, green, and blue channels in RGB), the color channels of the image are randomly adjusted. This transformation can change the hue and brightness of the image, adapt to different color conditions, and improve the recognition ability of different color backgrounds.

[0097] It should also be noted that when the Gaussian noise enhancement model and random channel transformation are two different image enhancement techniques used in combination, first, Gaussian noise is added to the first electronic flow bill image. The intensity of Gaussian noise is randomly distributed in the image and conforms to the normal distribution. This step can simulate the natural noise that may be encountered during the acquisition of the first electronic flow bill image, forcing the model to learn to recognize image features in the presence of noise. Next, the RGB channels of the first electronic flow bill image are randomly transformed. This includes randomly adjusting the gain or offset of each channel, or randomly swapping channels to change the hue and brightness of the image. This transformation helps learn to recognize the same object under different color representations. In addition, Gaussian noise enhancement and random channel transformation can be applied alternately multiple times, and a new variant of the first electronic flow bill image is generated after each transformation. This alternating application can further increase the diversity of the image and can learn to recognize features under a wider range of color and noise conditions. By combining these two enhancement methods, the generalization ability is improved, enabling it to better handle images under various color and noise conditions in practical applications.

[0098] The second electronic statement image is subjected to image preprocessing to obtain a third electronic statement image, wherein the image preprocessing means converting the second electronic statement image into a grayscale image, and scaling the length and width in proportion to align the image to a uniform length and width dimension.

[0099] It should be noted that the second electronic statement image is converted from color to a grayscale image. Grayscale images only contain brightness information, not color information, which can reduce the complexity of processing while retaining sufficient information to recognize text. The second electronic statement image is resized so that the aspect ratio of the second electronic statement image is consistent with the preset standard size. Proportional scaling ensures that the text in the image will not affect the recognition effect due to size deformation. All images are adjusted to a uniform resolution or size so that all images can be processed with the same parameters, simplifying subsequent processing steps. After the above preprocessing steps, the third electronic statement image obtained has a uniform format and improved quality, providing better input for text recognition. The method used for converting to grayscale is a conventional method, and this application has not made any changes to this.

[0100] Image detection is performed on the third electronic statement image to obtain a plurality of text regions, wherein image detection means cutting the third electronic statement image to form the text region, and outputting the four corner coordinates of the text region and the confidence level.

[0101] It should be noted that the image detection process starts with cutting the preprocessed image to identify and separate the areas containing text. These areas may include text information such as account information, transaction records, dates, etc. on the bill. Through image cutting, several text areas are formed, each of which contains a part of the text to be recognized. For each text area, its four corner coordinates in the original image are output, namely the coordinate positions of the upper left corner, upper right corner, lower left corner, and lower right corner. It helps to accurately locate the text area. Confidence is an indicator that indicates the confidence in the accuracy of the detected text area. High confidence means that the model believes that the detection result is very reliable.

[0102] Binarization is performed on the plurality of text regions to obtain the plurality of binarized text regions.

[0103] Image recognition is performed on the binarized text regions to obtain text information of the electronic statement, wherein the text information includes categories and corresponding statement information under the categories.

[0104] It should be noted that each pixel in the several text areas is adaptively binarized to convert the image into a format that is easier to recognize text. In addition, during image recognition, the binarized several text areas are processed using a convolutional neural network. Specifically, the convolutional neural network includes three parts: a convolutional layer, a loop layer, and a decoding layer. The function of the convolutional layer is to extract features from the input image. The extracted feature map will enter the loop layer. The loop layer outputs the prediction results for each frame of the feature sequence and extracts the semantic information therein. Finally, the decoding layer converts the obtained prediction probability distribution into a tag sequence to obtain the final recognition result. The decoding layer is actually the loss function in the model. By minimizing the loss function, the final image recognition result is obtained through training, that is, the text information of the electronic flow bill. Among them, the text information contains categories and the corresponding bill information under the categories. In this way, in the subsequent steps, the categories are matched first, and then the bill information corresponding to the matched categories is filled in the filling area.

[0105] It should also be noted that in the aforementioned step of obtaining the text information of the electronic statement bill, after image recognition, the information obtained is often scattered. At this time, it is necessary to use post-processing tools, such as table reconstruction methods, indicator calculation methods, image processing methods, text format methods and other conventional reconstruction methods to reconstruct the recognition results according to information such as the template location of the bill recognition field, and finally output the electronic statement bill corresponding to the field. The structured information of the electronic statement bill is obtained, and the text information of the electronic statement bill is obtained.

[0106] The text information of the electronic statement bill is mapped to the same category in a pre-configured electronic statement bill template to obtain a target electronic statement bill with bill information, wherein the pre-configured electronic statement bill template has the same category as the electronic statement bill and a filling area for bill information under the category, and the unmapped category in the electronic statement bill template is processed by filling with zeros.

[0107] It should be noted that the pre-configured electronic statement template refers to a preset template formed according to the statement flow of each bank institution, that is, before processing the electronic statement, the statement flow form of the bank institution has been pre-acquired and the statement flow template has been configured. The pre-configured electronic statement template can include multiple categories as needed, and corresponding filling areas under the categories. It can be understood that according to the complex situations such as the presence or absence of table lines, multiple header names, various sorting, and interlaced rows and columns in the statement of each financial institution, the statement positioning is divided into fixed-size areas through the pre-configured electronic statement template, and the location and mapping of the pre-configured statement identification field (such as name, ID number, statement time interval, serial number, transaction time, transaction object, transaction nature, transaction amount, etc.) in the area are pre-configured. Exemplarily, the text information of the pre-configured electronic statement includes the name: Zhang San, where the name is the category and Zhang San is the statement information. The name category is found in the electronic statement template, and Zhang San is written into the filling area.

[0108] It should also be noted that the unmapped categories in the pre-configured electronic statement template are processed by zero padding. Zero padding ensures the integrity of the data and avoids incomplete information due to missing data, which is crucial for subsequent data analysis and processing. Zero padding not only ensures that all records meet the preset format requirements and facilitates data integration and storage; it can also reduce potential errors caused by incomplete data to a certain extent.

[0109] The standardized structure data of the electronic statement is obtained according to the target electronic statement with statement information.

[0110] It should be noted that, in the aforementioned steps, the original electronic statement bill has been converted into a target electronic statement bill that conforms to the pre-configured electronic statement bill template. That is to say, by converting the original electronic statement bill in the above manner, an electronic statement bill with standardized structural data is obtained. This standardized structural data can facilitate analysis in subsequent steps.

[0111] S500, using the decision model, processing the standardized structure data of the electronic statement to obtain a decision result message.

[0112] Specifically, obtaining the decision result message includes the following steps:

[0113] Extracting features from the standardized structure data of the electronic statement to obtain feature labels for the standardized structure data;

[0114] According to the feature labels of the standardized structure data, in the decision model, a derivative indicator library matching the feature labels is selected to obtain a target decision model, wherein the decision model pre-stores a number of comparable derivative indicator libraries for flow characteristic analysis;

[0115] The target decision model is used to compare and process the standardized structure data of the electronic statement to obtain the decision result message.

[0116] It should be noted that the feature extraction is performed on the standardized structure data of the electronic flow bill. The feature extraction method is a conventional method and is not limited in this application. The feature extraction process involves identifying and marking key information in the standardized data, such as transaction amount, date, frequency, counterparty, etc., which are converted into feature tags. These feature tags are a high-level summary of the original data and can capture the important characteristics and patterns of the bill data. After the feature extraction is completed, these feature tags will be matched in the decision model to select a suitable derivative indicator library. The decision model pre-stores a series of derivative indicator libraries for flow characteristic analysis. These indicator libraries are comparable and are specially designed for different types of flow data. According to the matching degree between the feature tag and the derivative indicator library, the derivative indicator library is selected to form the target decision model to prepare for the next step of comparison processing.

[0117] Next, the selected target decision model is used to perform comparison processing on the standardized structure data of the electronic statement. This process includes comparing and analyzing the statement data with the indicators in the derivative indicator library one by one to obtain the comparison analysis results, that is, the decision result message. In the subsequent steps, the user's risk can be evaluated based on the decision result message.

[0118] It should also be noted that derivative indicators are usually combined according to dimensions such as transaction time, transaction object, transaction nature, transaction amount, transaction frequency, and transaction messages of customer bill flows, and bank flow data is processed and classified into a comparable derivative indicator system. The derivative indicator library based on bank flow tags mainly includes a transaction amount derivative indicator library, a transaction behavior derivative indicator library, and a transaction attribute derivative indicator library.

[0119] S600: Perform risk control review according to the decision result message to obtain a risk control review reference result.

[0120] Specifically, obtaining the risk control review reference results includes the following steps:

[0121] According to the preset risk rule engine of bank flow derivative indicators, the decision result message is matched and judged to obtain a judgment result, wherein the risk rule engine of bank flow derivative indicators at least includes a rule set consisting of income and loan discrepancy, abnormal transaction amount, high-risk transaction behavior, hit on the black and gray product list, and suspected fraud of bank flow, and the judgment result indicates the number of rules in the rule set that are hit;

[0122] According to the judgment result, a risk control review rating is performed to obtain a risk control review reference result, wherein the risk control review rating represents a grade division according to the number of rules in the rule set that are hit.

[0123] It should be noted that, based on the multi-dimensional derivative indicator output, the relevant risk indicator engine library is configured in the risk rule engine based on the bank flow derivative indicator, and the user is judged whether the risk rule is hit according to the rule trigger. The rule configuration includes income and loan discrepancies, abnormal transaction amounts, high-risk transaction behaviors, hits on the black and gray product lists, and suspected fraud in bank flow. For example, hitting one rule is low risk, hitting two rules is medium risk, and hitting three rules is high risk. In addition, each rule can also be weighted to determine the importance of the rule. For example, the rule that bank flow is suspected of fraud is a high-risk rule. In other words, hitting only this rule can also be considered high risk.

[0124] It should also be noted that the risk control review rating represents the classification according to the number of rules in the rule set that are hit, and the user risk level and income level stratification can be output.

[0125] To evaluate the consistency, rationality and potential risk points of billing data. The target decision model and the risk rule engine based on bank flow derivative indicators constitute the bank flow verification model. The built-in rules and algorithms of the bank flow verification model will conduct in-depth analysis of the data, identify any abnormal patterns or behaviors that do not meet expectations, and form user risk registration and income level stratification based on the output of the flow verification model. In other words, the structured bank flow data identified by OCR sends data to the bank flow verification model. After processing the data, the bank flow verification model obtains the user risk registration and income level stratification based on the output of the flow verification model, and uses the user risk registration and income level stratification based on the output of the flow verification model to obtain the user's relevant financial information.

[0126] For example, some indicator rules are as follows:

[0127] Indicator name 1: The proportion of income from individuals in the past three months;

[0128] Indicator name 2: Average account balance in the past three months;

[0129] Indicator Name 3: Proportion of large amounts of funds before and after festivals in the past 6 months;

[0130] Indicator Name 4: Proportion of lump-sum deposit and withdrawal of funds in the past 3 months;

[0131] Model Configuration Rule 1: Indicator Name 1 > X%, and the personal names of income and expenditure overlap;

[0132] Model Configuration Rule 2: Indicator Name 2 < X yuan;

[0133] Model Configuration Rule 3: Indicator Name 3 > X%;

[0134] Model Configuration Rule 4: Indicator Name 4 > X%;

[0135] Note: X is a variable and is adjusted immediately according to business objectives.

[0136] In summary, a method for identifying electronic bank statements of financial institutions provided by this application has main technical nodes including receiving the bank statement file by email and parsing the bank statement pdf file. The adjustment at the system architecture level is to add a new email service for receiving the bank statement emails sent by customers.

[0137] The main breakdown of requirements is as follows:

[0138] a. The email system needs to support dynamically creating emails for customers. For example, after Zhang San logs in to the system, an email zhangsan@xxxx.cn will be automatically created. Therefore, the email system needs to support user operations through the api method;

[0139] b. The email has only the receiving scenario and no scenario of sending emails externally;

[0140] c. The received email information needs to be stored in the bank;

[0141] d. When parsing the pdf file of the customer's bank statement, the file cannot be transferred to the external network through an external network interface and needs to be completed within the internal network.

[0142] Since it involves the domain name resolution of external network emails, dns resolution needs to be performed at the network level to achieve the domain name resolution of external network emails.

[0143] In addition, security measures for a method for identifying electronic bank statements of financial institutions:

[0144] 1. Use the domain name black and white list: Use the domain name black and white list of the email service to control which email servers can send emails to the system. By configuring the relevant settings of the mail transfer agent (MTA) such as Postfix, emails from unauthorized domains can be rejected;

[0145] 2. Content filtering and spam prevention: Configure email content filters and spam prevention systems, such as SpamAssassin or Rspamd, to detect and block suspicious or malicious emails;

[0146] 3. Increase the complexity of email account passwords: Email account passwords are generated by the system and can be set to more complex rules;

[0147] 4. Scheduled cleanup of email accounts: Currently, the mailbox is only needed to receive emails, and the emails can be deleted after the content is parsed.

[0148] In a second aspect, the present application provides a financial institution electronic statement recognition system, which applies the aforementioned financial institution electronic statement recognition method, and the system includes:

[0149] An acquisition unit, used for acquiring a business order request;

[0150] The matching unit is used to match the business order request with the historical task order to obtain a matching result:

[0151] A generating unit, configured to generate a dedicated mailbox according to the matching result;

[0152] The acquisition unit is further used to acquire the electronic statement sent to the dedicated mailbox, identify the electronic statement by using OCR recognition, and obtain the standardized structure data of the electronic statement;

[0153] A decision-making unit, used to process the standardized structure data of the electronic statement using a decision model to obtain a decision result message;

[0154] The result unit is used to perform risk control review according to the decision result message to obtain a risk control review reference result.

[0155] In a third aspect, the present application provides a computer storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the aforementioned method for identifying electronic statement of a financial institution.

[0156] In a fourth aspect, the present application provides a computer program, which, when executed by a processor, implements the steps of the aforementioned financial institution electronic statement identification method.

[0157] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0158] The various embodiments in the present disclosure are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0159] The protection scope of the present disclosure is not limited to the above-mentioned embodiments. Obviously, those skilled in the art can make various changes and modifications to the present disclosure without departing from the scope and spirit of the present disclosure. If these changes and modifications fall within the scope of the claims of the present disclosure and their equivalents, the intention of the present disclosure also includes these changes and modifications.

Claims

1. A method for identifying electronic bank statements of financial institutions. It is characterized in that The method comprises: Get business order request; Matching the business order request with the historical task order to obtain a matching result; Generate a dedicated mailbox according to the matching results; Obtaining the electronic statement sent to the dedicated mailbox, identifying the electronic statement by using OCR recognition, and obtaining standardized structure data of the electronic statement; Using the decision model, the standardized structure data of the electronic statement is processed to obtain a decision result message; Conduct risk control review based on the decision result message to obtain a risk control review reference result.

2. The method for identifying the electronic statement of a financial institution according to claim 1, It is characterized in that The step of obtaining a business order request includes: Generate a sharing page according to the business order request formed by the business request end, and present an operation guide on the sharing page; The execution of the business order request is guided according to the operation guidance presented in the sharing page.

3. The method for identifying electronic bank statements of a financial institution according to claim 2, It is characterized in that The step of matching the business order request with the historical task order to obtain a matching result includes: Constructing a task order database, wherein the task order database stores a number of historical task orders generated according to the historical business order requests; The business order request is matched with the historical task orders in the task order database to obtain a matching result.

4. The method for identifying electronic bank statements of a financial institution according to claim 1, It is characterized in that The step of generating a dedicated mailbox according to the matching result includes: If the matching result is successful, a dedicated mailbox for receiving electronic statement is generated, and the dedicated mailbox is sent to the service request end; If the matching result fails, a request task order and a task order number are generated, a dedicated mailbox is generated according to the request business order, and the dedicated mailbox is sent to the business request end, and the task order number is associated with the request task order.

5. The method for identifying electronic bank statements of a financial institution according to claim 1, It is characterized in that The step of obtaining the electronic statement sent to the dedicated mailbox, identifying the electronic statement by using OCR recognition, and obtaining the standardized structure data of the electronic statement includes: Obtaining the electronic statement sent to the dedicated mailbox; Using OCR recognition to adjust the image angle of the electronic statement to obtain a first electronic statement image, wherein the image angle adjustment means adjusting the electronic statement in four directions of 0°, 90°, 180° or 270°; Performing color enhancement on the first electronic statement image to obtain a second electronic statement image, wherein the color enhancement represents the application of the generalization capability of the Gaussian noise ratio enhancement model and the random channel transformation enhancement model to enhance the recognition of different colors in the first electronic statement image; Performing image preprocessing on the second electronic statement image to obtain a third electronic statement image, wherein the image preprocessing means converting the second electronic statement image into a grayscale image, and scaling the length and width in proportion to align the image to a uniform length and width dimension; Performing image detection on the third electronic statement image to obtain a plurality of text regions, wherein the image detection means cutting the third electronic statement image to form the text regions, and outputting the four corner coordinates of the text regions and the confidence level; Binarizing the plurality of text regions to obtain the plurality of binarized text regions; Performing image recognition on the binarized text regions to obtain text information of the electronic statement, wherein the text information includes categories and corresponding statement information under the categories; Mapping the text information of the electronic statement to the same category in a pre-configured electronic statement template to obtain a target electronic statement with statement information, wherein the pre-configured electronic statement template has the same category as the electronic statement and a filling area for the statement information under the category, and the unmapped category in the electronic statement template is processed by zero padding; The standardized structure data of the electronic statement is obtained according to the target electronic statement with statement information.

6. The method for identifying electronic bank statements of a financial institution according to claim 1, It is characterized in that The step of using the decision model to process the standardized structure data of the electronic statement to obtain a decision result message includes: Extracting features from the standardized structure data of the electronic statement to obtain feature labels for the standardized structure data; According to the feature labels of the standardized structure data, in the decision model, a derivative indicator library matching the feature labels is selected to obtain a target decision model, wherein the decision model pre-stores a number of comparable derivative indicator libraries for flow characteristic analysis; The target decision model is used to compare and process the standardized structure data of the electronic statement to obtain the decision result message.

7. The method for identifying electronic bank statements of a financial institution according to claim 1, It is characterized in that The step of performing risk control review according to the decision result message to obtain a risk control review reference result includes: According to the preset risk rule engine of bank flow derivative indicators, the decision result message is matched and judged to obtain a judgment result, wherein the risk rule engine of bank flow derivative indicators at least includes a rule set consisting of income and loan discrepancy, abnormal transaction amount, high-risk transaction behavior, hit on the black and gray product list, and suspected fraud of bank flow, and the judgment result indicates the number of rules in the rule set that are hit; According to the judgment result, a risk control review rating is performed to obtain a risk control review reference result, wherein the risk control review rating represents a grade division according to the number of rules in the rule set that are hit.

8. A financial institution electronic statement recognition system, It is characterized in that The method for identifying an electronic statement of a financial institution according to any one of claims 1 to 7 is applied, wherein the system comprises: An acquisition unit, used for acquiring a business order request; A matching unit, used for matching the business order request with the historical task order to obtain a matching result; A generating unit, configured to generate a dedicated mailbox according to the matching result; The acquisition unit is further used to acquire the electronic statement sent to the dedicated mailbox, identify the electronic statement by using OCR recognition, and obtain the standardized structure data of the electronic statement; A decision-making unit, used to process the standardized structure data of the electronic statement using a decision model to obtain a decision result message; The result unit is used to perform risk control review according to the decision result message to obtain a risk control review reference result.

9. A computer storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the steps of the method for identifying an electronic statement of a financial institution according to any one of claims 1 to 7 are implemented.

10. A computer program, It is characterized in that When the computer program is executed by a processor, the steps of the method for identifying an electronic statement of a financial institution according to any one of claims 1 to 7 are implemented.

Images