Form checking method and device, computing device, storage medium and program product

Abstract

The application provides a form checking method and device. The form checking method comprises the following steps: receiving a form checking request, wherein the form checking request carries to-be-checked form data; calling a preset checking rule to check the to-be-checked form data, and obtaining a checking result, wherein the preset checking rule comprises at least one of a compliance checking rule, a consistency checking rule and an integrity checking rule; based on attribute information of the to-be-checked form data, searching a historical form database to find historical form data matched with the form checking request, and determining a historical checking result of the historical form data; performing risk prediction on the to-be-checked form data, and obtaining a risk prediction result; and performing result comprehensive analysis based on the checking result, the historical checking result and the risk prediction result, and obtaining a target checking result for the to-be-checked form data. Through the method, the form can be checked from multiple dimensions, so that the accuracy of the result is improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a form verification method. This application also relates to a form verification device, a computing device, a computer-readable storage medium, and a computer program product. Background Technology

[0002] With the development of internet technology and the acceleration of enterprise digital transformation, the demand for automation and intelligentization of financial processes is becoming increasingly urgent.

[0003] In the existing technology, traditional form verification systems, as core tools for enterprise management, have long relied on manual input and fixed rule-driven processes. While such systems have improved the efficiency of manual review to some extent, their technical limitations are particularly significant in complex scenarios, making it impossible to achieve true intelligence. In practical applications, they mainly rely on manual review.

[0004] Therefore, there is an urgent need for a form verification method that can improve verification efficiency and ensure the accuracy of results, in order to break through the bottlenecks of traditional technology. Summary of the Invention

[0005] In view of this, embodiments of this application provide a form verification method to address the technical deficiencies existing in the prior art. Embodiments of this application also provide a form verification apparatus, a computing device, a computer-readable storage medium, and a computer program product.

[0006] According to a first aspect of the embodiments of this application, a form verification method is provided, including:

[0007] Receive a form verification request, wherein the form verification request carries form data to be verified;

[0008] The form data to be verified is validated by calling preset validation rules to obtain validation results. The preset validation rules include at least one of compliance validation rules, consistency validation rules, and integrity validation rules.

[0009] Based on the attribute information of the form data to be verified, search the historical form database for historical form data that matches the form verification request, and determine the historical verification result of the historical form data;

[0010] Perform risk prediction on the form data to be verified, and obtain the risk prediction results;

[0011] Based on the verification results, the historical verification results, and the risk prediction results, a comprehensive analysis is performed to obtain the target verification result for the form data to be verified.

[0012] According to a second aspect of the embodiments of this application, a form verification device is provided, comprising:

[0013] The receiving module is configured to receive form verification requests, wherein the form verification requests carry form data to be verified;

[0014] The verification module is configured to call preset verification rules to verify the form data to be verified and obtain the verification result. The preset verification rules include at least one of compliance verification rules, consistency verification rules, and integrity verification rules.

[0015] The search module is configured to search for historical form data that matches the form verification request from the historical form database based on the attribute information of the form data to be verified, and to determine the historical verification result of the historical form data;

[0016] The risk prediction module is configured to perform risk prediction on the form data to be verified and obtain the risk prediction result;

[0017] The obtaining module is configured to perform a comprehensive analysis of the verification results, the historical verification results, and the risk prediction results to obtain the target verification result for the form data to be verified.

[0018] According to a third aspect of the embodiments of this application, a computing device is provided, including: a memory and a processor;

[0019] The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer programs / instructions are executed by the processor, they implement a form verification method.

[0020] According to a fourth aspect of the embodiments of this application, a computer-readable storage medium is provided that stores a computer program / instructions, which, when executed by a processor, implement a form verification method.

[0021] According to a fifth aspect of the embodiments of this application, a computer program product is provided, including a computer program / instructions, which, when executed by a processor, implement a form verification method.

[0022] The form verification method provided in this application performs strong verification on the form data to be verified carried in the received form verification request, that is, it calls the preset verification rules for verification, and finds the historical verification results of the corresponding historical form data and risk prediction based on the attribute information of the form data to be verified. Based on the verification results, historical verification results and risk prediction results obtained from the above verification, the method performs a comprehensive analysis of the results and determines the target verification result from a multi-dimensional perspective. It does not have to rely entirely on manual review, thus improving verification efficiency and ensuring the accuracy of the results. Attached Figure Description

[0023] Figure 1 This is a flowchart of a form verification method provided in one embodiment of this application;

[0024] Figure 2 This is a flowchart illustrating a form verification method for expense reimbursement provided in one embodiment of this application;

[0025] Figure 3 This is a hierarchical diagram of a form verification method for expense reimbursement provided in one embodiment of this application;

[0026] Figure 4 This is a schematic diagram of the structure of a form verification device provided in one embodiment of this application;

[0027] Figure 5 This is a structural block diagram of a computing device provided in one embodiment of this application. Detailed Implementation

[0028] Many specific details are set forth in the following description to provide a full understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this application; therefore, this application is not limited to the specific embodiments disclosed below.

[0029] The terminology used in one or more embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the scope of one or more embodiments of this application. The singular forms “a,” “the,” and “the” used in one or more embodiments of this application and in the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” used in one or more embodiments of this application refers to and includes any or all possible combinations of one or more associated listed items.

[0030] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this application, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this application, and similarly, second may also be referred to as first.

[0031] First, the terminology used in one or more embodiments of the present invention will be explained.

[0032] Reimbursement form: An internal document used by a company to apply for expense reimbursement. It must be accompanied by relevant supporting documents (such as invoices) for verification.

[0033] Invoice: A legally binding document issued by the seller to the buyer to confirm the transaction details, amount, and tax declaration.

[0034] This application provides a form verification method. This application also relates to a form verification device, a computing device, a computer-readable storage medium, and a computer program product, which will be described in detail in the following embodiments.

[0035] Figure 1 A flowchart of a form verification method according to an embodiment of this application is shown, specifically including steps 102-110:

[0036] Step 102: Receive a form verification request, wherein the form verification request carries form data to be verified.

[0037] The form verification request is the instruction carrier that triggers the form verification process. It includes at least the form verification data, i.e., the dataset that needs to be verified, and the applicant information of the form verification request. The form data to be verified is the dataset that needs to be verified. The form data to be verified contains at least one of the following types of data: structured data, such as form field key-value pairs; unstructured data, such as attachment files (scanned copies, images, PDF documents, etc.); and time-series data: form filling trajectory (number of field modifications, duration of time spent, etc.).

[0038] For example, in the expense reimbursement scenario, the form verification request is an expense reimbursement request. The form data to be verified uploaded by the user may include expense reimbursement bills, invoices, sales lists, orders, contracts, goods acceptance forms, etc. After the user uploads the data and clicks submit, a form verification request is generated, and then the form data to be verified carried in the form verification request is verified.

[0039] Based on the above description of the form data to be verified, its type can include various types. In the case of unstructured data, the form data to be verified cannot be directly verified. It needs to be structured, that is, the form data to be verified needs to be preprocessed in order to achieve subsequent verification.

[0040] In one possible implementation of this application, after receiving the form verification request, the following is included:

[0041] The form data to be verified is subjected to text recognition to extract each text content and the corresponding coordinate information of each text content; the layout feature vector of the form data to be verified is determined based on each text content and the corresponding coordinate information; similar templates are matched from the template library according to the layout feature vector, and the target data in the form data to be verified is identified based on the similar templates.

[0042] The form data to be verified can be scanned documents, images, PDF documents, and other documents that require text recognition, and OCR can be used for text recognition.

[0043] Before text recognition, the form data to be verified can be preprocessed to improve the accuracy of text recognition. Specifically, when the form data to be verified is an image or other file, the following processing can be performed on the form data during text recognition: 1. Denoising, binarization, and skew correction; 2. Segmenting the user-submitted file into different regions to improve the accuracy of subsequent text recognition, dividing it into different regions such as title area, QR code area, and table area, and recording the bounding box coordinates of each region relative to the entire file. These bounding box coordinates can then be used as the coordinate information of each text content. Region segmentation can use a trained detection model or other methods. This application embodiment only illustrates the use of a detection model for region segmentation as an example of an achievable method, and other implementation methods are not limited here; for example, for the table area, a table line detection algorithm can be used to extract the row and column structure, and the coordinate matrix of each cell can be recorded as coordinate information.

[0044] After extracting the text content and its corresponding coordinate information from the form data to be verified through text recognition, the following processing can be performed:

[0045] 1. Structured processing. This involves data cleaning, rule extraction, and field mapping of the text content to transform it into data in a specific format, such as: {Form Category: "XX", Amount: "XX", Item Name: "XX"}.

[0046] 2. Input the structured data into the NLP model (Natural Language Processing model). The NLP model performs word segmentation, part-of-speech tagging, and other processing on the structured data, and then performs semantic understanding. Specifically, it can perform entity recognition and extract names, times, locations, etc. For example, taking an invoice as an example, it can extract entities such as company name, amount, invoice date, product name, quantity, and unit price; it can also perform relation extraction, etc.

[0047] 3. Determine the layout feature vector of the form to be verified based on the text content and the coordinate information corresponding to each text content; match similar templates from the template library according to the layout feature vector, and identify the target data in the form data to be verified based on the similar templates.

[0048] The layout feature vector can be understood as the feature vector of each text content and its corresponding coordinate information. When generating the layout feature vector, the text content and its corresponding coordinate information can be concatenated and embedded.

[0049] A template library can be understood as a database that stores predefined form templates. This template library contains templates for different form categories and stores metadata of key form information contained in each form category, such as: data type, validation rule binding (regular expression, numerical range, associated fields), business tags (sensitive field markers), etc. Key form information can be understood as key fields.

[0050] Taking invoice categories as an example: the template library can contain various invoice types, such as: 04, paper VAT general invoice; 08, electronic VAT special invoice; 10, electronic VAT general invoice (other); 10c, electronic VAT general invoice (passenger transport); 10d, electronic VAT general invoice (non-taxable); 11, VAT general invoice (roll type); 14, electronic VAT general invoice (toll); 16, blockchain invoice; 31, electronic invoice (VAT special invoice); 32, electronic invoice (…). Common invoices include: 88, Highway / Waterway Passenger Ticket (Name); 89, Highway / Waterway Passenger Ticket (Name); 91, Taxi Receipt; 92, Train Ticket; 93, Airline Ticket Itinerary; 94, Highway / Waterway Passenger Ticket (No Name); 94b, International Passenger Transport Invoice; 95, Fixed-Amount Invoice; 97, General Machine-Printed Invoice; 98, Paper Toll Invoice (Bridge / Gate); 98a, Paper Toll Invoice (Other); NA, No Invoice; NB, Foreign Invoice, etc., along with the key form information for each invoice type. For example: If invoice A is a general VAT invoice, it will not be marked, and only the key form information such as invoice amount, tax amount, invoice issuer, buyer, product name, and product unit price will be identified; if invoice B is a construction service invoice, it can be marked, and the content in the remarks column of the construction service invoice can be identified, and the service location name and project name can be obtained and marked as well; if invoice C is a transportation invoice such as an airplane ticket or train ticket, the departure station and arrival station fields can be obtained and marked.

[0051] In practical applications, the layout feature vector of the form to be verified is compared with the templates in the template library to calculate similarity. The template with the highest similarity is identified as the similar template. In addition, the template library can store the form category, key form information, and coordinate information of key form information. When searching based on the layout feature vector of the form to be verified, the key form information and coordinate information are embedded and then similarity is judged to determine the similar template. Similarly, the key form information and coordinate information of key form information in the template library can be embedded in advance, and similarity matching can be performed directly when the layout feature vector of the form to be verified is obtained.

[0052] In practical applications, the template with the highest similarity can be used as the similar template, or the template with the highest similarity and higher than the preset similarity can be used as the similar template.

[0053] After identifying similar templates, target data is identified in the form data to be verified based on the corresponding key form information in the similar templates. For example, if the form data to be verified contains data 1, data 2, data 3, data 4, and data 5, and the key form information only contains fields a, b, and c, then data 2 corresponding to field a, data 4 corresponding to field b, and data 5 corresponding to field c in the form data to be verified are identified as target data. The key form information may be the same or different for different form categories. For example, the key form information for most invoice types includes fields such as invoice amount, tax amount, invoice issuer, buyer, product name, and unit price, while the key form information for construction service invoices must at least include the service location name and project name.

[0054] In practical applications, to improve the adaptability to recognize different form categories, the template library can be continuously updated. In a possible implementation of this disclosure, the step of matching similar templates from the template library based on the layout feature vector includes:

[0055] If no similar template is found, the form category and key form information of the form to be verified are marked; based on the marking results, the template library is updated.

[0056] Specifically, if no similar template is found, the form data to be verified carried in the current form verification request can be sent to the annotators, who can then annotate the form data to be verified and generate annotation results. Alternatively, other annotation tools can be used to annotate the form data to be verified carried in the current form verification request and generate annotation results. Then, it can be determined whether the annotation results need to be reviewed by the annotators as needed.

[0057] The specific annotation results can be stored and processed according to the above description of the template library, which will not be elaborated further in this application.

[0058] Upon receiving a form verification request, the form data to be verified carried in the form verification request is verified in multiple dimensions. The verification includes steps 104-108:

[0059] Step 104: Call the preset verification rules to verify the form data to be verified and obtain the verification result. The preset verification rules include at least one of compliance verification rules, consistency verification rules and integrity verification rules.

[0060] Among them, the preset validation rules are the rules for basic validation of the form data to be verified, and the preset validation rules are strong validation rules. The compliance validation rules can be understood as the rules for determining whether the form data to be verified conforms to the regulations, the consistency validation rules can be understood as the rules for whether the form data to be verified is consistent, and the integrity validation rules are the rules for whether the form data to be verified is complete.

[0061] In practical applications, a rule engine can be configured. This engine contains at least one rule for validating the form data to be checked. The corresponding preset validation rule can be invoked based on the form category of the data. For example, a preset rule might be used to check whether the remarks column of a construction service invoice contains both the service location name and the project name. If not, the invoice does not meet the integrity validation rule.

[0062] In one embodiment of this disclosure, the compliance verification rules include at least one of the following rules: integrity verification rules for the form supervision seal and special seal, logical consistency verification rules for the form code and form number, verification rules for whether the invoice date is within the verification period, and related data verification rules associated with the form data to be verified.

[0063] In one embodiment of this disclosure, the consistency verification rules include at least one of the following rules: consistency verification rules between the tax-exclusive amount and tax amount in the form data to be verified and the target amount submitted in the form verification request; consistency verification rules between the seller name in the form data to be verified and the supplier name registered in the invoice submitted in the form verification request; and consistency verification rules between the goods name, specifications and models in the form data to be verified and the purchase contract or order.

[0064] In one embodiment of this disclosure, the integrity verification rules include at least one of the following rules: verification rules for the clarity of the form to be verified and the identifiability of key form information, verification rules for the consistency of the numbering of the attachments to the form to be verified and the header form to be verified, and verification rules for whether the remarks column of the building service form contains the service location and project name.

[0065] Taking the expense reimbursement scenario as an example, the preset engine rules may include at least one of the following rules.

[0066]

[0067] Step 106: Based on the attribute information of the form data to be verified, search for historical form data that matches the form verification request in the historical form database, and determine the historical verification result of the historical form data.

[0068] The attribute information of the form data to be verified can include form category, amount, date, etc. In practice, the attribute information of the form data to be verified can be matched with the attribute information of historical form data stored in the historical form database to determine the historical form data that meets the matching conditions, and to determine the historical verification result of the historical form data. The historical form database contains historically verified form data. In practical applications, the attribute information and verification results of the form data to be verified can be stored in the historical form database after the verification process is completed. Alternatively, they can be stored separately or according to the data structure in the template database; this application does not impose any restrictions on this.

[0069] In one embodiment of this disclosure, considering the efficiency of retrieval, the attribute information can be vectorized to obtain a high-dimensional vector or vector set. The vector or vector set is then used to search for historical form data in a historical form database. Specifically, the step of searching for historical form data matching the form verification request in the historical form database based on the attribute information of the form data to be verified includes:

[0070] Using vector retrieval technology, the attribute information of the form data to be verified is vectorized and encoded to generate a query vector; based on the query vector, a reference vector matching the form verification request is searched from the historical form database, wherein the reference vector is the feature vector corresponding to the historical form in the historical form database; the historical form data is determined based on the reference vector.

[0071] Vector retrieval technology achieves efficient searching by converting data into high-dimensional vectors and calculating similarity. Vector retrieval technology can employ KNN, ANN, FAISS, Annoy, etc. The query vector is the result of vectorizing the form data to be verified.

[0072] In practical applications, to improve retrieval efficiency, each historical form in the historical form database can be pre-vectorized to obtain a reference vector. During matching, the reference vector is directly compared with the query vector to obtain the comparison result.

[0073] For example, the historical form library contains the following historical forms:

[0074] Invoice 1: Company A, Amount: 1000 RMB, Date: 2023-01-01; Verification Result: Passed;

[0075] Invoice 2: Company B, amount 2000 yuan, date 2023-01-02; Verification result: Passed.

[0076] Invoice 3: Company A, Amount: 1500 RMB, Date: 2023-01-03; Verification Result: Passed;

[0077] Invoice 4: Company C, Amount: 500 RMB, Date: 2023-01-04. Verification Result: Passed.

[0078] Invoice 5: Company B, Amount: 2500 RMB, Date: 2023-01-05; Verification Result: Passed;

[0079] If the form data to be verified is an airplane ticket with a price of 1,000 yuan and a date of 2023-01-02, then the form data to be verified is vectorized and encoded, retrieved, and the historical form data is determined, and the historical verification result is "passed".

[0080] In practical applications, multiple similar historical forms can be retrieved. Therefore, the verification result of the historical form data in the historical form with the highest similarity can be selected as the historical verification result. Alternatively, the historical verification result can be determined based on the proportion, that is, the verification result in the historical form with the largest proportion of verification results exceeding the preset similarity can be used as the historical verification result.

[0081] Step 108: Perform risk prediction on the data in the form to be verified, and obtain the risk prediction results.

[0082] Risk prediction can be understood as implicit risk reasoning of the data in the form to be verified, such as information associated with the data and information on duplicate verifications. In the expense reimbursement scenario, risk prediction can include inferences about related transactions and duplicate reimbursements. In specific implementation, an LLM model can be considered for risk prediction.

[0083] In one embodiment of this disclosure, the step of performing risk prediction on the form data to be verified and obtaining risk prediction results includes:

[0084] Input the form data to be verified and the target task into the target model, wherein the target task includes at least one of the following: related transaction risk task, duplicate reimbursement risk task, and invoice logic contradiction risk task;

[0085] The target model is used to output corresponding risk type labels and risk prediction results based on the form data to be verified and the target task, wherein the risk type labels are determined based on the target task.

[0086] The target task is a risk prediction task, such as a related-party transaction risk task, a duplicate reimbursement risk task, and a bill logic contradiction risk task, which is the purpose of this prediction.

[0087] In the reimbursement scenario, training can be conducted in the following ways: The data to be verified, i.e., the invoice text: the text content obtained from invoice recognition, and the data structure of the target data, such as: invoice number, invoice date, amount, supplier name, etc. Contextual information: Contextual information related to the invoice, such as the purpose of the reimbursement, related projects or departments, etc. Task description: Text specifying the task, such as "Please review whether this invoice meets the reimbursement standards," "Related party transaction risk task," "Duplicate reimbursement risk task," "Invoice logic contradiction risk task," etc. Sample labels: Determined according to the type of target task. Different task types correspond to different sample labels. For categorized tasks, the label can be yes or no; for generated tasks, corresponding text can be generated. For example, the label for "Related party transaction risk task" is "Related party transaction text," the label for "Please review whether this invoice meets the reimbursement standards" is "compliant" or "non-compliant," and the label for "Duplicate reimbursement risk task" is "duplicate" or "non-duplicate," etc. This application will not provide examples of each. After inputting samples and the target task, the model makes predictions based on the samples to obtain prediction results. Based on the prediction results and labels, the model's storage is modified until the error between the prediction results and labels meets a preset threshold.

[0088] In summary, by reasoning about the implicit risks in the data to be verified, potential risks in the data to be verified are discovered, thus improving the accuracy of predictions.

[0089] After obtaining the verification results, the historical verification results, and the risk prediction results, the final verification results of the form data to be verified are determined based on the results obtained from these three perspectives.

[0090] Step 110: Based on the verification results, the historical verification results, and the risk prediction results, perform a comprehensive analysis to obtain the target verification result for the form data to be verified.

[0091] The comprehensive analysis of results can be understood as calculating and analyzing the verification results, the historical verification results, and the risk prediction results. In practical applications, the comprehensive analysis of results can be performed by weighted summation. When the calculated result is less than the first threshold, the target verification result is considered successful; when it is greater than the second threshold, the verification fails; when it is between the first and second thresholds, it is forwarded to the approval node for manual review. Alternatively, if a certain result has a problem, the target verification result can be determined to be a verification failure. In specific implementation, the settings can be configured according to the needs of those skilled in the art.

[0092] In one embodiment of this disclosure, to improve the accuracy of the comprehensive analysis of results, the step of performing a comprehensive analysis based on the verification results, the historical verification results, and the risk prediction results to obtain a target verification result for the form data to be verified includes:

[0093] Determine the result score and weight coefficient corresponding to the verification result, the historical verification result, and the risk prediction result respectively; based on the weight coefficient corresponding to the verification result, the result score corresponding to the verification result, the historical verification result, and the risk prediction result respectively are weighted and calculated to obtain the target verification score; determine the target verification result for the form data to be verified based on the target verification score.

[0094] The weight coefficients corresponding to the verification result, the historical verification result, and the risk prediction result can be configured as needed. The result scores corresponding to the verification result, the historical verification result, and the risk prediction result can be determined based on the correspondence between the result and the score. For example, if the verification result of the verification rule is "fail", the score is 1, and if it is "pass", the score is 0. Similarly, if the historical verification result is "fail", the score is 1, and if it is "pass", the score is 0. Similarly, the result score corresponding to the risk prediction result can also be determined in the above manner. In addition, the risk prediction result can also be determined based on the risk prediction probability output by the model. For example, if the model outputs "fail", the probability is 0.8, then the result score can be 0.8.

[0095] In one embodiment of this disclosure, determining the target verification result for the form data to be verified based on the verification score includes:

[0096] If the verification score is less than the first threshold, the target verification result is a successful verification.

[0097] If the verification score is greater than the second threshold, the target verification result is a verification failure, and the reason for the failure is output.

[0098] If the verification score is greater than the first threshold and less than the second threshold, the form verification request is forwarded to the approval node.

[0099] The first and second thresholds can be pre-set result thresholds, and the reasons for failure can be pre-configured. The correspondence between verification results and corresponding reasons can be pre-configured. In the event of a verification failure, the reason for failure is determined based on one of the following: the verification result, historical verification results, or risk prediction results. For example, if the verification rule is "the attached invoice number should match the header invoice number," and the verification result is "inconsistent," then the setting is "the attached invoice number does not match the header invoice number." If the verification rule is "the remarks column of the construction service invoice must include the service location name and project name," and the verification result is "does not include," then the setting is "the remarks column of the construction service invoice does not specify the service location name and / or project name."

[0100] Taking the reimbursement scenario as an example, if the invoice is configured with the validation rule "duplicate reimbursement is not allowed", its validation result is "failed", and its result score is 1. If the historical form data retrieved from the historical form database has a historical validation result of "failed", its result score is also 1. If the risk prediction result is "duplicate" and the prediction probability is 0.7, then when performing comprehensive result analysis, the result is calculated based on the corresponding result score and weight coefficient. If the result of 0.5*1+0.2*1+0.3*0.7 is greater than the second threshold, then the target validation result is determined to be a validation failure. The form data to be validated is then returned to the node that submitted the form validation request, and the failure reason of "duplicate reimbursement" is output accordingly.

[0101] When an invoice of a certain category is set with the validation rule "calculated according to travel expense standards," the output result is "passed," and the result score is 0. The historical verification result for similar cases in the historical form database, such as "special travel personnel can be reimbursed for actual expenses," is also "passed," with a result score of 0. The model prediction can then predict the personnel's status information to determine if they are special personnel or have special travel. If the result is "passed," with a result score of 0, it is less than the first threshold, and the target verification result is determined to be successful, completing the verification. If the target model inference finds that the travel is not for special personnel, and the calculated result exceeds the second threshold, the verification fails, and the reason for failure is output; or if it is between the first and second thresholds, the form verification request is forwarded to the approval node for manual review.

[0102] In addition, some complex approval nodes are difficult to approve, and business approvers may not be able to grasp the situation. They can request the model to calculate and process the data. The model will look up the rules for the approval point based on the request, and indicate whether the data of the form to be verified meets the corresponding validation rules. It will also find the corresponding historical form data and display it to the approvers, and provide its own suggestions for the approvers to refer to.

[0103] In one embodiment of this disclosure, to ensure the smooth progress of the process, the target approval node can be determined based on the load status of the approval node to ensure the efficiency of the process. Specifically, forwarding the form verification request to the approval node includes:

[0104] Determine the load status of the approval node and the target data of the form data to be verified; determine the target approval node based on the load status and the target data; forward the form verification request to the target approval node.

[0105] The target data is the data identified by key form information. The load status can be understood as the form verification requests pending approval at the approval node. The load status of the approval node can be determined based on the number of form verification requests or the size of the form data pending verification carried in them.

[0106] To ensure the accuracy of the results, priority strategies can be set to prioritize the processing of specific form verification requests or to prioritize their routing to auditors. For example, "documents with amounts exceeding a preset amount will be prioritized for routing to senior auditors", "special loan types will be prioritized for routing to senior auditors", and "payment items will be prioritized for order assignment and review".

[0107] In addition, different approval paths can be set for form verification requests of different form types. For example, when the invoice type is travel expense, the project code in the invoice belongs to the marketing department, and the amount exceeds 5,000 yuan, the approval path is department manager → financial director; if the invoice type is ordinary value-added tax invoice, the approval path can be finance personnel.

[0108] After forwarding to the approval node, the approver may also be uncertain about the verification result. Therefore, the verification results, historical verification results, risk prediction results, etc. obtained in the aforementioned steps 104-106 can be pushed to the approver for reference.

[0109] After forwarding the form verification request to the approval node, the following steps are included:

[0110] Upon receiving the approval instruction for the form verification request, the system displays the approval path and historical verification results of the historical form data corresponding to the form verification request, and / or displays the reasoning path and associated evidence of the risk prediction results.

[0111] Among them, the approval instruction is the instruction initiated by the approver to trigger the reference case; the approval path refers to which approval nodes the historical form data has passed through, the approval results of each node, and the final verification result; the reasoning path of the risk prediction result is the logical steps by which the model infers the result; and the corroborating evidence is the relevant content that supports the risk prediction result.

[0112] In summary, by showing auditors the approval path and historical verification results of the historical form data corresponding to the form verification request, and / or showing the reasoning path and related evidence of the risk prediction results, auditors can be provided with reference results, thereby improving audit efficiency.

[0113] In summary, this application improves the accuracy and efficiency of verification results by evaluating the form data to be verified from multiple perspectives.

[0114] The following is in conjunction with the appendix Figure 2 Appendix Figure 3 Taking the form verification method provided in this application as an example in an expense reimbursement scenario, the form verification method will be further explained. Among other things, Figure 2 This paper illustrates a flowchart of a form verification method for an expense reimbursement scenario, according to an embodiment of this application. Figure 3 This application illustrates a hierarchy diagram according to an embodiment, which specifically includes the following steps:

[0115] Step 202: Receive a form verification request, wherein the form verification request carries form data to be verified.

[0116] Specifically, the form verification request is an expense reimbursement request, which is submitted by the user through a webpage, client, or other means. The expense reimbursement request may carry the user's personal information and form data to be verified. The form data to be verified includes various invoices, receipts, etc. The user's personal information includes at least one of the following: user name, employee ID, company name, company code, user department name, user department code, etc. In addition, it may also carry the user's expense reimbursement amount for this instance. After the expense reimbursement request is submitted, an expense reimbursement form cover is generated based on the user's personal information and the information of the person who prepared the form carried in the expense reimbursement request. The expense reimbursement form cover may also include the generated expense reimbursement form number or its QR code, the log of the expense reimbursement form generation, and other information.

[0117] In addition, a scanning control can be provided. Users can click the scanning control to trigger the scanning program. This can be used for invoices, forms, etc. The types of invoices will not be detailed here, as described above.

[0118] Step 202 is the perception layer.

[0119] Step 204: Perform text recognition on the form data to be verified, and extract each text content and the coordinate information corresponding to each text content from the form data to be verified.

[0120] Step 206: Determine the layout feature vector of the form to be verified based on the text content and the coordinate information corresponding to each text content.

[0121] Step 208: Match similar templates from the template library based on the layout feature vector, and identify the target data in the form data to be verified based on the similar templates.

[0122] Step 210: If no similar template is found, mark the form category and key form information of the form to be verified; based on the marking results, update the template library.

[0123] Step 212: Call the preset verification rules to verify the form data to be verified and obtain the verification result. The preset verification rules include at least one of compliance verification rules, consistency verification rules and integrity verification rules.

[0124] The verification rules are shown in the table above. In addition, other rule engines can be called for verification, such as calling the tax bureau's interface for invoice verification.

[0125] Step 214: Using vector retrieval technology, the attribute information of the form data to be verified is vectorized and encoded to generate a query vector.

[0126] Step 216: Based on the query vector, find the reference vector that matches the form verification request from the historical form library, wherein the reference vector is the feature vector corresponding to the historical form in the historical form library.

[0127] Step 218: Determine the historical form data based on the reference vector, and determine the historical verification result of the historical form data.

[0128] Step 220: Input the form data to be verified and the target task into the target model, wherein the target task includes at least one of the following: related transaction risk task, duplicate reimbursement risk task, and invoice logic contradiction risk task.

[0129] Step 222: Using the target model, output the corresponding risk type label and risk prediction result based on the form data to be verified and the target task, wherein the risk type label is determined based on the target task.

[0130] Step 224: Determine the result score and weight coefficient corresponding to the verification result, the historical verification result and the risk prediction result respectively.

[0131] Step 226: Based on the weight coefficients corresponding to the verification result, the historical verification result, and the risk prediction result, perform a weighted calculation on the result scores corresponding to the verification result, the historical verification result, and the risk prediction result to obtain the target verification score.

[0132] Step 228: Determine the target verification result for the form data to be verified based on the target verification score.

[0133] Step 230: If the verification score is greater than the first threshold, the target verification result is successful.

[0134] Step 232: If the verification score is less than the second threshold, the target verification result is a verification failure, and the reason for the failure is output.

[0135] Step 234: If the verification score is less than the first threshold and greater than the second threshold, forward the form verification request to the approval node.

[0136] Step 234 includes: determining the load status of the approval node and the target data of the form data to be verified; determining the target approval node based on the load status and the target data; and forwarding the form verification request to the target approval node.

[0137] Step 236: Upon receiving the approval instruction for the form verification request, display the approval path and historical verification results of the historical form data corresponding to the form verification request, and / or display the reasoning path and associated evidence of the risk prediction results.

[0138] Steps 204-236 constitute the decision-making level.

[0139] After the target verification result is determined, if the target verification result is successful, the approval result is determined to be successful. If the target verification result is unsuccessful, the form verification request will be returned, and the user can decide whether to supplement the data and resubmit the form verification request as needed. If it is forwarded to the approval node, the success or failure will be determined according to the approval result, and the subsequent processing is the same as the processing method for the target verification result.

[0140] Once the verification is successful, it proceeds to the execution layer.

[0141] If the verification is successful, the expense reports, invoices, and other documents will be merged into a PDF. The reimbursement category will be determined based on the target data, and the corresponding data will be stored.

[0142] In addition, other rules can be set, such as setting a rule for intelligent preliminary review by a verification system, and setting a manual approval node regardless of whether the verification is successful or not.

[0143] When the process enters the intelligent initial review node, the corresponding voucher preview can be automatically generated based on the voucher engine configuration and the company, business type, scenario, account, cost center, etc. of the document. If the voucher preview is correct, it will be pushed to the SAP system to generate the voucher. If the voucher preview or generation fails, it will be automatically transferred to manual processing, and the reason for the failure will be recorded and an alarm will be issued.

[0144] By batch processing mixed invoices and directly connecting to SAP / Oracle systems, it is possible to quickly verify the authenticity of invoices, achieve instant reimbursement by taking photos with a mobile phone, and overcome the limitations of traditional template matching through deep collaboration between OCR and RAG. It is especially suitable for processing non-standard format invoices and cross-border multilingual documents.

[0145] When a voucher is generated, if the business is a payment business and automatic payment is configured, the system will automatically determine the corresponding payment method, exclude offline payment methods, and automatically make online payments. When the payment is successful and the bank returns the payment transaction record, the payment node will be completed automatically.

[0146] Once all processes are completed, the system will automatically push the batch records, form information, PDFs of all invoice forms, and vouchers generated during the process to the electronic archive for filing.

[0147] Once all information in the reimbursement system is entered into the electronic archive, due to policy requirements to retain paper archives, the paper archives will be scanned and compared before being sent to the archive. The scanned document will be OCR-generated, and then semantic analysis will be performed to extract the content. This content will be compared with the content of the original PDF that has been OCR-recognized. Only when the difference is less than 10% and the important content is consistent in the semantic analysis using the NLP model can the document be archived.

[0148] In summary, the above methods avoid the inefficiencies and subjectivity inherent in traditional manual review, thereby improving the efficiency and accuracy of expense reimbursement.

[0149] Corresponding to the above method embodiments, this application also provides an embodiment of a form verification device. Figure 4 A schematic diagram of a form verification device according to an embodiment of this application is shown. Figure 4 As shown, the device includes:

[0150] The receiving module 402 is configured to receive a form verification request, wherein the form verification request carries form data to be verified;

[0151] The verification module 404 is configured to call preset verification rules to verify the form data to be verified and obtain the verification result. The preset verification rules include at least one of compliance verification rules, consistency verification rules and integrity verification rules.

[0152] The lookup module 406 is configured to search for historical form data that matches the form verification request from the historical form database based on the attribute information of the form data to be verified, and to determine the historical verification result of the historical form data;

[0153] Risk prediction module 408 is configured to perform risk prediction on the form data to be verified and obtain risk prediction results;

[0154] The module 410 is configured to perform a comprehensive analysis of the verification results, the historical verification results, and the risk prediction results to obtain the target verification result for the form data to be verified.

[0155] The above is a schematic scheme of a form verification device according to this embodiment. It should be noted that the technical solution of this form verification device and the technical solution of the above-described form verification method belong to the same concept. Details not described in detail in the technical solution of the form verification device can be found in the description of the technical solution of the above-described form verification method. Furthermore, the components in the device embodiment should be understood as functional modules necessary to implement each step of the program flow or each step of the method; these functional modules are not actual functional divisions or separations. A device claim defined by such a set of functional modules should be understood as a functional module architecture that primarily implements the solution through the computer program described in the specification, and not as a physical device that primarily implements the solution through hardware.

[0156] Figure 5 A structural block diagram of a computing device 500 according to an embodiment of this application is shown. The components of the computing device 500 include, but are not limited to, a memory 510 and a processor 520. The processor 520 is connected to the memory 510 via a bus 530, and a database 550 is used to store data.

[0157] The computing device 500 also includes an access device 540, which enables the computing device 500 to communicate via one or more networks 560. Examples of these networks include PSTN (Public Switched Telephone Network), LAN (Local Area Network), WAN (Wide Area Network), PAN (Personal Area Network), or combinations of communication networks such as the Internet. The access device 540 may include one or more of any type of wired or wireless network interface (e.g., NIC (Network Interface Controller)), such as an IEEE 802.11 WLAN (Wireless Local Area Network) wireless interface, Wi-MAX (Worldwide Interoperability for Microwave Access) interface, Ethernet interface, USB (Universal Serial Bus) interface, cellular network interface, Bluetooth interface, and NFC (Near Field Communication).

[0158] In one embodiment of this application, the aforementioned components of the computing device 500 and Figure 5Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 5 The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this application. Those skilled in the art can add or replace other components as needed.

[0159] Computing device 500 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or PCs (Personal Computers). Computing device 500 can also be a mobile or stationary server.

[0160] The processor 520 is used to execute computer-executable instructions for the form verification method.

[0161] The above is an illustrative scheme of a computing device according to this embodiment. It should be noted that the technical solution of this computing device and the technical solution of the form verification method described above belong to the same concept. For details not described in detail in the technical solution of the computing device, please refer to the description of the technical solution of the form verification method described above.

[0162] An embodiment of this application also provides a computer-readable storage medium storing a computer program / instructions which, when executed by a processor, are used for a form verification method.

[0163] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium and the technical solution of the form verification method described above belong to the same concept. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the form verification method described above.

[0164] An embodiment of this application also provides a computer program product, including a computer program / instructions that, when executed by a processor, are used for a form verification method.

[0165] The above is an illustrative scheme of a computer program product according to this embodiment. It should be noted that the technical solution of this computer program product and the technical solution of the form verification method described above belong to the same concept. For details not described in detail in the technical solution of the computer program product, please refer to the description of the technical solution of the form verification method described above.

[0166] The computer program / instructions include computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, ROM (Read-Only Memory), RAM (Random Access Memory), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added to or subtracted according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0167] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0168] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0169] The preferred embodiments disclosed above are merely illustrative of this application. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this application. These embodiments are selected and specifically described in this application to better explain the principles and practical applications of this application, thereby enabling those skilled in the art to better understand and utilize this application. This application is limited only by the claims and their full scope and equivalents.

Claims

1. A form verification method, characterized in that, include: Receive a form verification request, wherein the form verification request carries form data to be verified; The form data to be verified is validated by calling preset validation rules to obtain validation results. The preset validation rules include at least one of compliance validation rules, consistency validation rules, and integrity validation rules. Based on the attribute information of the form data to be verified, search the historical form database for historical form data that matches the form verification request, and determine the historical verification result of the historical form data; The step of searching for historical form data matching the form verification request from the historical form database based on the attribute information of the form data to be verified includes: Using vector retrieval technology, the attribute information of the form data to be verified is vectorized and encoded to generate a query vector; based on the query vector, a reference vector matching the form verification request is searched from the historical form database, wherein the reference vector is a feature vector corresponding to a historical form in the historical form database; the historical form data is determined based on the reference vector. Perform risk prediction on the form data to be verified, and obtain the risk prediction results; A comprehensive analysis is performed based on the verification results, the historical verification results, and the risk prediction results to obtain a target verification result for the form data to be verified; the comprehensive analysis of the verification results, the historical verification results, and the risk prediction results to obtain a target verification result for the form data to be verified includes: Determine the result score and weight coefficient corresponding to the verification result, the historical verification result, and the risk prediction result; Based on the weight coefficients corresponding to the verification result, the historical verification result, and the risk prediction result, the result scores corresponding to the verification result, the historical verification result, and the risk prediction result are weighted and calculated to obtain the target verification score. If the target verification score is less than the first threshold, the target verification result is successful; if the target verification score is greater than the second threshold, the target verification result is unsuccessful, and the reason for failure is output; if the target verification score is greater than the first threshold and less than the second threshold, the form verification request is forwarded to the approval node.

2. The form verification method according to claim 1, characterized in that, After receiving the form verification request, the process includes: Perform text recognition on the form data to be verified, and extract each text content and the coordinate information corresponding to each text content from the form data to be verified. The layout feature vector of the form to be verified is determined based on the text content and the coordinate information corresponding to each text content. Match similar templates from the template library based on the layout feature vector, and identify the target data in the form data to be verified based on the similar templates.

3. The form verification method according to claim 2, characterized in that, The step of matching similar templates based on the layout feature vector includes: If no similar template is found, mark the form category and key form information of the form to be verified; The template library is updated based on the annotation results.

4. The form verification method according to claim 1, characterized in that, The integrity verification rules include at least one of the following rules: the clarity and identifiability of key form information of the form to be verified, the consistency of the numbering of the attachments to the form to be verified and the header of the form to be verified, and the verification rules for whether the remarks column of the building service form contains the service location and project name.

5. The form verification method according to claim 1, characterized in that, The step of performing risk prediction on the form data to be verified and obtaining the risk prediction result includes: Input the form data to be verified and the target task into the target model, wherein the target task includes at least one of the following: related transaction risk task, duplicate reimbursement risk task, and invoice logic contradiction risk task; The target model is used to output corresponding risk type labels and risk prediction results based on the form data to be verified and the target task, wherein the risk type labels are determined based on the target task.

6. A form verification device, characterized in that, include: The receiving module is configured to receive form verification requests, wherein the form verification requests carry form data to be verified; The verification module is configured to call preset verification rules to verify the form data to be verified and obtain the verification result. The preset verification rules include at least one of compliance verification rules, consistency verification rules, and integrity verification rules. The search module is configured to search for historical form data that matches the form verification request from the historical form database based on the attribute information of the form data to be verified, and to determine the historical verification result of the historical form data; The search module is further configured to use vector retrieval technology to vectorize the attribute information of the form data to be verified, and generate a query vector; based on the query vector, search for a reference vector that matches the form verification request from the historical form database, wherein the reference vector is a feature vector corresponding to a historical form in the historical form database; and determine the historical form data based on the reference vector. The risk prediction module is configured to perform risk prediction on the form data to be verified and obtain risk prediction results; The obtaining module is configured to perform a comprehensive analysis of the verification results, the historical verification results, and the risk prediction results to obtain the target verification result for the form data to be verified. The obtaining module is further configured to determine the result score and weight coefficient corresponding to each of the verification result, the historical verification result, and the risk prediction result; based on the weight coefficient corresponding to each of the verification result, the historical verification result, and the risk prediction result, a weighted calculation is performed on the result scores corresponding to each of the verification result, the historical verification result, and the risk prediction result to obtain a target verification score; if the target verification score is less than a first threshold, the target verification result is a successful verification; if the target verification score is greater than a second threshold, the target verification result is a failed verification, and the reason for the failure is output; if the target verification score is greater than the first threshold and less than the second threshold, the form verification request is forwarded to the approval node.

7. A computing device, characterized in that, include: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, wherein when the computer programs / instructions are executed by the processor, they implement the method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program / instructions that, when executed by a processor, implement the method described in any one of claims 1 to 5.

9. A computer program product, characterized in that, Includes a computer program / instruction that, when executed by a processor, implements the method according to any one of claims 1 to 5.

Images