Abnormal document identification method and device

Abstract

The invention discloses an abnormal document identification method and device. The method comprises the steps that a target document is acquired, and the document type of the target document is recognized; m pieces of feature information in the target document are recognized according to the document type, and M is a positive integer; an auditing model and an auditing document are selected according to the document type, and the auditing model and the auditing document are used for auditing the content of the target document; comparing the target file with the auditing document to obtain a first auditing result, and inputting the M pieces of feature information into an auditing model to obtain a second auditing result; and determining an abnormal auditing result of the target document according to the first auditing result and the second auditing result. By means of the method and device, the problem that in the related technology, the efficiency of manually checking the document is low is solved.

Description

Technical Field

[0001] This application relates to the field of big data, and more specifically, to a method and apparatus for identifying abnormal documents. Background Technology

[0002] As enterprises become increasingly information-based, the number of business documents has grown dramatically, encompassing a range of important information such as contracts, policy documents, financial statements, and invoices. These documents are not only the foundation of daily operations but also core elements for ensuring compliance with regulations and maintaining business reputation.

[0003] However, in terms of document management and compliance review, enterprises primarily rely on manual review. This process is cumbersome and time-consuming, especially when dealing with large volumes of documents, making it inefficient and unable to meet the demands of rapid business growth. Furthermore, human error is inevitable when processing large amounts of information, leading to low accuracy in extracting key information and potentially causing errors in compliance judgments.

[0004] There is currently no effective solution to the problem of low efficiency in manually reviewing documents in related technologies. Summary of the Invention

[0005] This application provides a method and apparatus for identifying abnormal documents, in order to solve the problem of low efficiency in the related art of manually reviewing documents.

[0006] According to one aspect of this application, a method for identifying abnormal documents is provided. The method includes: acquiring a target document and identifying its document type; identifying M feature information in the target document based on the document type, where M is a positive integer; selecting an auditing model and an auditing document based on the document type, wherein the auditing model and the auditing document are used to audit the content of the target document; comparing the target document and the auditing document to obtain a first audit result; inputting the M feature information into the auditing model to obtain a second audit result; and determining the abnormal audit result of the target document based on the first and second audit results.

[0007] Optionally, identifying the document type of the target document includes: determining multiple preset locations and sequentially determining whether document information exists at each preset location in the target document; for N target locations where document information exists, sequentially determining the baseline information at each target location to obtain N sets of baseline information, where N is a positive integer; comparing a set of baseline information and document information at the same target location to obtain N comparison results, where the comparison results are used to characterize whether there is matching information associated with the document information in the set of baseline information; and determining the document type of the target document based on the N comparison results.

[0008] Optionally, determining the document type of the target document based on N comparison results includes: if there are comparison results indicating that there is matching information associated with the document information in a set of benchmark information, obtaining the document type indicated by the matching information to obtain H document type sets, where H is a positive integer and H is less than or equal to N; if H is a preset number, determining the document types contained in the document type sets as the document type of the target document; if H is greater than the preset number, obtaining the document types at the intersection of the H document type sets to obtain the document type of the target document; if the intersection of the H document type sets is empty, determining the document type with the largest number in the H document type sets as the document type of the target document.

[0009] Optionally, identifying M feature information in the target document based on the document type includes: determining the document style of the target document based on the document type; obtaining a preset template corresponding to the document style from a template library based on the document style, wherein the style of the preset template is the same as the document style; determining the features to be acquired marked in the preset template, and obtaining feature information under each feature to be acquired in the target document through a recognition model to obtain M feature information.

[0010] Optionally, selecting the audit model and audit document based on the document type includes: when the document type is type 1, determining the document used for consistency detection with the target document as the audit document and determining the model used for consistency detection of the target document as the audit model; when the document type is type 2, determining the document used for content audit of the target document as the audit document and determining the model used for compliance detection of the target document as the audit model.

[0011] Optionally, when the document type is the first type, comparing the target document and the review document to obtain the first review result includes: identifying the style of the target document and determining whether the style of the target document is the same as the standard style of the review document; if the style of the target document is different from the standard style, determining the first review result as abnormal; if the style of the target document is the same as the standard style, obtaining the standard information corresponding to each feature information in the review document; if each feature information is consistent with the corresponding standard information, determining the first review result as normal; if there are inconsistent feature information and standard information, determining the first review result as abnormal.

[0012] Optionally, when the document type is the first type, the review model is trained as follows: multiple first historical documents are obtained, and the second historical document corresponding to each first historical document is obtained, as well as the difference information between each group of first historical documents and second historical documents; a group of first historical documents, second historical documents and difference information are determined as a group of sample data to obtain multiple groups of first sample data, and the preset neural network model is trained using multiple groups of first sample data to obtain the review model.

[0013] Optionally, when the document type is the second type, comparing the target document and the audit document to obtain the first audit result includes: obtaining the audit content in the audit document, and auditing the target document according to each audit content in turn to obtain multiple audit results; if multiple audit results indicate that the target document has no abnormalities, the first audit result is determined to be normal; if any audit result indicates that the target document is abnormal, the first audit result is determined to be abnormal.

[0014] Optionally, when the document type is the second type, the review model is trained as follows: obtain content review documents for each document type to obtain multiple content review documents, and configure the multiple content review documents in a preset neural network model; obtain multiple second historical documents and the abnormal information of each second historical document; use each second historical document and the corresponding abnormal information as a set of sample data to obtain multiple sets of second sample data; use the multiple sets of second sample data to train the preset neural network model to obtain the review model.

[0015] According to another aspect of this application, an anomalous document identification device is provided. The device includes: a first identification unit for acquiring a target document and identifying its document type; a second identification unit for identifying M feature information in the target document based on its document type, where M is a positive integer; a selection unit for selecting an auditing model and an auditing document based on the document type, wherein the auditing model and the auditing document are used to audit the content of the target document; a comparison unit for comparing the target document and the auditing document to obtain a first audit result, and inputting the M feature information into the auditing model to obtain a second audit result; and a first determination unit for determining the anomalous audit result of the target document based on the first and second audit results.

[0016] According to another aspect of the present invention, a computer program product is also provided, comprising a computer program that, when executed by a processor, implements a method for identifying abnormal documents provided in the foregoing embodiments of the present application.

[0017] According to another aspect of the present invention, an electronic device is also provided, comprising one or more processors and a memory; the memory stores computer-readable instructions, and the processor is used to execute the computer-readable instructions, wherein the computer-readable instructions, when executed, perform an abnormal document identification method provided in the foregoing embodiments.

[0018] This application employs the following steps: acquiring a target document and identifying its document type; identifying M feature information points in the target document based on its document type, where M is a positive integer; selecting an auditing model and an auditing document based on the document type, wherein the auditing model and auditing document are used to audit the content of the target document; comparing the target document and the auditing document to obtain a first audit result; inputting the M feature information points into the auditing model to obtain a second audit result; and determining any abnormal audit results for the target document based on the first and second audit results. This solves the problem of low efficiency in manual document auditing in related technologies. By identifying the document type and determining the document and model for auditing the target document based on its feature information, and then processing the feature information in the target document using the document and model to obtain the audit result, the technical effect of improving the efficiency of document auditing while ensuring the accuracy of document auditing is achieved. Attached Figure Description

[0019] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0020] Figure 1 This is a flowchart of an abnormal document identification method provided according to an embodiment of this application;

[0021] Figure 2 This is a flowchart of an optional method for identifying abnormal documents according to an embodiment of this application;

[0022] Figure 3 This is a schematic diagram of an abnormal document identification device provided according to an embodiment of this application;

[0023] Figure 4 This is a schematic diagram of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0024] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0025] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present application.

[0026] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0027] It should be noted that the abnormal document identification method and apparatus defined in this disclosure can be used in the field of big data, or in any field other than big data. The application field of the abnormal document identification method and apparatus defined in this disclosure is not limited.

[0028] It should be noted that all information, user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, and displayed data) used in this application are information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with the relevant laws, regulations, and standards of the relevant regions, have taken necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse use. If the user chooses to refuse, the process proceeds to the expert decision-making process. For example, this system has interfaces with relevant users or institutions. Before obtaining relevant information, a request to obtain the information needs to be sent to the aforementioned user or institution through the interface, and the relevant information is obtained only after receiving consent from the aforementioned user or institution.

[0029] The embodiments or examples disclosed herein are not exhaustive, but merely illustrative of some embodiments or examples, and are not intended to limit the scope of protection of this disclosure. Unless otherwise specified, each step in a particular embodiment or example can be implemented as an independent embodiment, and the steps can be arbitrarily combined. For example, a solution after removing some steps in a particular embodiment or example can also be implemented as an independent embodiment, and the order of the steps in a particular embodiment or example can be arbitrarily interchanged. Furthermore, optional methods or examples in a particular embodiment or example can be arbitrarily combined; moreover, embodiments or examples can be arbitrarily combined. For example, some or all steps of different embodiments or examples can be arbitrarily combined, and a particular embodiment or example can be arbitrarily combined with optional methods or examples of other embodiments or examples.

[0030] For ease of description, the following explains some of the nouns or terms used in the embodiments of this application:

[0031] Robotic Process Automation (RPA) technology uses software robots to automatically perform repetitive and routine tasks. It simulates tasks performed by humans on computers, accelerating and improving the efficiency and accuracy of business processes through automation.

[0032] NLP (Natural Language Processing) technology: Using NLP technology to parse contract documents and extract key information, such as contract terms, amounts, and dates.

[0033] Machine learning algorithms: Use machine learning algorithms to intelligently classify documents, identify and distinguish different types of contract documents, and automatically assign them the corresponding processing flow.

[0034] According to embodiments of this application, a method for identifying abnormal documents is provided.

[0035] Figure 1 This is a flowchart of an abnormal document identification method provided according to an embodiment of this application. For example... Figure 1 As shown, the method includes the following steps:

[0036] Step S101: Obtain the target document and identify the document type of the target document.

[0037] It should be noted that the execution entity in this embodiment can be an anomaly detection system, the target document refers to the document that needs to be automatically processed and reviewed, and the document type is classified according to the content and format of the document, such as contract documents, financial records, invoices, etc.

[0038] Specifically, since documents may exist in various types, and different methods are used for anomaly detection for different document types, when processing a target document, it is first necessary to identify the document type of the target document, and then determine how to perform anomaly detection operations on the document based on the document type.

[0039] It should be noted that when there are multiple documents that need to be detected for anomalies, RPA technology can be used to automatically traverse the database storing the documents, identify the documents to be processed, and add the documents to a queue. Then, when detecting the target documents one by one, the target documents can be directly retrieved from the queue.

[0040] Step S102: Identify M feature information in the target document according to the document type, where M is a positive integer.

[0041] It should be noted that the feature information is key information extracted from the target document. This key information can be used to detect the document and thus determine whether there are any anomalies.

[0042] Specifically, after determining the document type of the target document, since the feature information extraction methods differ for different document types, it is necessary to determine the extraction method based on the document type and use the corresponding extraction method to extract the feature information.

[0043] For example, if the document type is a contract document, the system will identify and extract characteristic information such as the names of both parties to the contract, the contract number, the amount, the signing date, and the payment terms.

[0044] Step S103: Select the audit model and audit document according to the document type. The audit model and audit document are used to audit the content of the target document.

[0045] It should be noted that the audit model refers to the algorithmic model used to compare and evaluate the compliance of a target document. The audit document is a document containing compliance rules and standards, used to compare the content of the target document.

[0046] Specifically, after obtaining the document type of the target document, the system can automatically select the matching review model and review document based on the document type. The review model can be based on deep learning technology for intelligent comparison and analysis of document content, while the review document can be a pre-set compliance rule base containing specific compliance policies and standards, thereby determining whether there are any anomalies in the feature information of the target document based on the review document.

[0047] For example, when the target document is a contract document, the contract review model and contract review rules document can be invoked. The contract review model is used to identify risk points in the contract, such as automatically checking for loopholes in the contract terms. The contract review rules document contains all the standards for enterprise contract compliance, such as the range of contract amount and the legality of payment terms, thereby determining whether the target document is accurate.

[0048] Step S104: Compare the target document and the review document to obtain the first review result. Input the M feature information into the review model to obtain the second review result.

[0049] Specifically, the system can compare the target document and the review document through an automated process to check for inconsistencies and obtain the first review result. At the same time, the system inputs the document's feature information into the review model and uses algorithms to conduct in-depth analysis to obtain the second review result, which assesses the compliance risks of the target document and thus jointly determines whether the target document has any anomalies from two perspectives.

[0050] For example, when processing contract documents, the system first compares the contract amount with the corresponding amount in the financial records. If they match, the first audit result is "amount compliant". Next, M characteristic information (such as the names of the contracting parties, contract amount, signing date, etc.) are input into the contract audit model. The model may analyze historical compliance cases to determine whether there are potential compliance risks and arrive at a second audit result, such as "payment terms are risky".

[0051] Step S105: Determine the abnormal review result of the target document based on the first review result and the second review result.

[0052] Specifically, given the first and second audit results, the system needs to comprehensively assess the compliance status of the target document based on both results. If any discrepancies are found in either result, an abnormal audit result will be generated, indicating a compliance issue. The system will automatically mark these anomalies and generate warning messages to facilitate subsequent processing or correction. Only when both the first and second audit results indicate that the target document is free of anomalies can it be determined that the target document is risk-free, thereby ensuring its security.

[0053] For example, if the system finds that the contract amount is inconsistent with the amount in the financial records (abnormal first audit result), and / or the contract audit model detects that there is a legal risk in the payment terms (abnormal second audit result), the system determines that the contract has "abnormal amount compliance" or "abnormal payment terms compliance" and sends this abnormal audit result to the corresponding business department for processing.

[0054] Furthermore, after obtaining the audit results, evaluation criteria can be set for the audit results based on business standards. These evaluation criteria can be used to assess the risk level or risk coefficient of the abnormal content found during the audit. For example, a compliance risk assessment model covering multiple dimensions can be established, including but not limited to compliance, adaptability, and robustness. In addition, the model can be configured with automatic early warning and response measure recommendation functions. For different compliance risk levels, corresponding response measures can be formulated, such as risk avoidance strategies, risk reduction plans, risk transfer methods, and risk acceptance plans, to help customers better identify potential compliance risks and take corresponding measures to deal with them.

[0055] Furthermore, after receiving the audit results, the system can automatically generate a compliance report. This report can include a summary of the original data, trends in compliance status, key indicator compliance status, potential risk areas, and improvement suggestions. To meet the diverse needs of different departments, the system offers robust report customization features. Users can adjust the report layout, colors, and font size according to their responsibilities, focuses, and preferences. Users can also add specific analytical dimensions, data filtering conditions, or annotations to make the report more tailored to actual needs. The system also supports multiple export formats, facilitating report viewing and sharing across different platforms and devices, thus ensuring the widespread circulation and effective use of information.

[0056] The abnormal document identification method provided in this application embodiment obtains a target document and identifies its document type; identifies M feature information in the target document based on the document type, where M is a positive integer; selects an audit model and an audit document based on the document type, wherein the audit model and audit document are used to audit the content of the target document; compares the target document and the audit document to obtain a first audit result; inputs the M feature information into the audit model to obtain a second audit result; and determines the abnormal audit result of the target document based on the first and second audit results. This solves the problem of low efficiency in manual document auditing in related technologies. By identifying the document type and determining the document and model for auditing the target document based on the feature information in the target document, and then processing the feature information in the target document through the document and model to obtain the audit result of the target document, the method achieves the technical effect of improving the efficiency of document auditing while ensuring the accuracy of document auditing.

[0057] To accurately determine the document type of a target document, optionally, in the abnormal document identification method provided in this application embodiment, identifying the document type of a target document includes: determining multiple preset locations, and sequentially determining whether document information exists at each preset location in the target document; for N target locations where document information exists, sequentially determining the reference information at each target location to obtain N sets of reference information, where N is a positive integer; comparing a set of reference information and document information at the same target location to obtain N comparison results, where the comparison results are used to characterize whether there is matching information associated with document information in the set of reference information; and determining the document type of the target document based on the N comparison results.

[0058] It should be noted that a preset location refers to a specific location or area within a document that may contain key information, pre-defined according to the document type. Document information can be information that may exist in the specific preset location, and this information is usually associated with the document type. Baseline information can be specific information used to determine the document type at the preset location.

[0059] Specifically, when identifying document types, a feature comparison method can be used. First, multiple preset locations are determined. These preset locations may contain special symbols or text used to determine the document type. For example, the title of the document can be identified, and it can be determined whether the word "contract" is present in the title, thereby determining whether the document is a contract. Alternatively, the upper right corner of the document can be identified to determine the document type based on the seal.

[0060] For example, assuming the target document is a contract, preset locations might include headers, footers, titles, and the beginnings of clauses. The system checks these locations for contract-related document information such as "Contract Number," "Party A," and "Party B."

[0061] After determining multiple preset locations, it is possible to determine whether document information exists in each preset location of the target document. For example, it is possible to determine whether there is document information such as a seal in the upper right corner of the target document.

[0062] Given N target locations where document information exists, multiple reference information can be obtained for each target location, resulting in N sets of reference information. Each set of reference information can include multiple pieces of information for identifying document types.

[0063] For example, a set of reference information corresponding to the upper right corner of a document may include multiple stamps of different styles, each stamp corresponding to a document type.

[0064] After identifying the target locations where document information exists and the baseline information at each target location, the document information at the same target location can be compared with each baseline information in a set of baseline information to determine whether there is content identical to the document information in the set of baseline information, thereby obtaining the comparison result. The comparison result can be no matching information or matching information, and the document type indicated by the matching baseline information is listed.

[0065] After obtaining the comparison results corresponding to each target location, the document type of the target document can be determined based on the N comparison results.

[0066] This embodiment can quickly locate areas that may contain key information by determining and checking document information at preset locations. By comparing the benchmark information with the document information, the document type can be accurately determined based on N comparison results, avoiding the subjectivity and error of manual judgment and significantly improving the automation level and accuracy of document processing.

[0067] To accurately determine the document type of the target document, optionally, in the abnormal document identification method provided in this application embodiment, determining the document type of the target document based on N comparison results includes: when there are comparison results indicating that there is matching information associated with the document information in a set of benchmark information, obtaining the document type indicated by the matching information to obtain H document type sets, where H is a positive integer and H is less than or equal to N; when H is a preset number, determining the document types contained in the document type sets as the document type of the target document; when H is greater than the preset number, obtaining the intersection document types of the H document type sets to obtain the document type of the target document; when the intersection of the H document type sets is empty, determining the document type with the largest number in the H document type sets as the document type of the target document.

[0068] Specifically, after obtaining the comparison results, since each comparison result contains content that is typeless (i.e. empty) or at least one document type, the document type indicated by the matching information (i.e., matching information) in the comparison results that represent matching information associated with document information in a set of benchmark information can be obtained, resulting in H document type sets. Wherein, if there is a comparison result that is typeless, H is less than N, and if each comparison result contains at least one document type, N is equal to H.

[0069] Furthermore, when obtaining H document type sets, since the number of document types and the specific type content contained in different document type sets may be different, when determining the document type of the target document, it is first necessary to determine whether the number indicated by H is the preset number, where the preset number can be 1.

[0070] When H equals 1, it indicates that there is valid document type information in only one set of document types. In this case, the document types in that set of document types can be determined as the document type of the target document. For example, if the document type in the set of document types is "contract," then the document type of the target document can be directly determined as "contract."

[0071] When H is greater than 1, it indicates that there are multiple sets of document types containing valid document type information. In this case, it is necessary to obtain the intersection document types of the H document type sets, and then determine the intersection document types as the document type of the target document. For example, if the document types in document type set 1 are A and B, and the document types in document type set 2 are C and B, then the document type of the target document can be determined to be type B.

[0072] It should be noted that when H is greater than 1, there may be cases where document types do not overlap, meaning that the document types in each document type set are different. In this case, the document type with the most occurrences can be determined as the document type of the target document, thus ensuring accurate document type determination. For example, if document type A is in document type set 1, document types A and B are in document type set 2, and document types C and D are in document type set 3, then the document type can be determined as A.

[0073] This embodiment accurately determines the document type through intersection analysis or majority voting principles, enabling the system to handle not only documents with obvious characteristics but also complex information coexisting, ensuring accurate document type identification. This not only improves the efficiency of document processing but also ensures the accuracy and relevance of subsequent automated review processes.

[0074] To accurately determine the feature information of the target document, optionally, in the abnormal document identification method provided in this application embodiment, identifying M feature information in the target document according to the document type includes: determining the document style of the target document according to the document type; obtaining a preset template corresponding to the document style from a template library according to the document style, wherein the style of the preset template is the same as the document style; determining the features to be acquired marked in the preset template, and obtaining the feature information under each feature to be acquired in the target document through an identification model to obtain M feature information.

[0075] It should be noted that a document style is the standard format or layout for a specific document type, which may include elements such as page layout, headings, and tables. A template library is a database that stores preset templates corresponding to various document types. Preset templates are templates that match the document style and are used to guide the system on how to accurately extract feature information from the document.

[0076] Specifically, when identifying feature information, it is necessary to first determine the document style of the document type based on the document type being identified. Since the document style reflects the basic structure and layout of the document, it is necessary to first identify the document style and then determine from which locations the corresponding feature information needs to be obtained based on the document style.

[0077] Furthermore, after determining the document style, a preset template with the same document style can be retrieved from the template library. Based on the determined document style, the preset template can guide the system on how to extract specific feature information from the target document, ensuring the accuracy and completeness of the information.

[0078] After obtaining the preset template, since the preset template is marked with the feature information that the document using the template needs to be identified, the features to be obtained marked in the preset template can be determined, and the feature information at the corresponding position can be obtained from the target document according to the position of each feature to be obtained in the template, so as to obtain M feature information.

[0079] For example, the system can use NLP and machine learning models to automatically extract key feature information such as "names of both parties to the contract," "contract amount," and "signing date" from the corresponding locations in the target contract document based on the annotations on the pre-set contract template. For instance, the model identifies the contract amount as "1 million" and uses it as one of the M feature information.

[0080] This embodiment uses a preset template and recognition model to identify and obtain M feature information from the target document, which not only reduces the burden of manual document processing and improves the accuracy and efficiency of information extraction, but also provides a reliable data foundation for subsequent automated comparison and review.

[0081] To accurately select the review model and review document, optionally, in the abnormal document identification method provided in this application embodiment, selecting the review model and review document according to the document type includes: when the document type is a first type, determining the document used for consistency detection with the target document as the review document, and determining the model used for consistency detection of the target document as the review model; when the document type is a second type, determining the document used for content review of the target document as the review document, and determining the model used for compliance detection of the target document as the review model.

[0082] It should be noted that the first type represents document types that require consistency checks, such as the consistency of invoices or financial documents. The second type represents document types that require content compliance checks, such as the compliance of contracts.

[0083] Specifically, when determining the audit model and audit document for the target document, it is necessary to classify them according to document type. When the target document is identified as belonging to the first type, the system will automatically select an audit document for consistency detection, which is usually another document closely related to the target document, such as an invoice. At the same time, the system will call an audit model specifically for consistency detection, which is used to determine whether the content between the target document and the audit document is consistent, ensuring the accuracy and integrity of the data.

[0084] For example, suppose the target document is an invoice, and its type is identified as Type 1. The system will automatically retrieve the financial records related to this invoice from the database and use them as the audit document. Simultaneously, a contract consistency detection model will be activated as the audit model. This model is trained to identify discrepancies between the invoice's amount, content, and other information and the corresponding information in the financial records, thereby determining whether the invoice is abnormal.

[0085] For documents identified as Category II, such as contracts, the system selects audit documents for content compliance checks. These are typically internal corporate compliance policy documents, industry standards, laws, and regulations. Simultaneously, the system activates a compliance check model to ensure the target document's content complies with the company's or industry's compliance requirements.

[0086] For example, if the target document is a contract, it is identified as Type II. The system will retrieve the company's compliance policy document from the document library for review. Subsequently, a compliance detection model is invoked as the review model. This model analyzes each clause of the contract to check for any violations of the company's compliance policy, thereby ensuring the accuracy and compliance of the contract content.

[0087] This embodiment achieves precise document review by flexibly selecting appropriate review documents and models based on the type of the target document. For "Type 1" documents, the system focuses on consistency detection to ensure that the document data is consistent with the company's records; for "Type 2" documents, the system focuses on content compliance to ensure that the document content meets company or industry standards. Through differentiated review strategies, the efficiency and accuracy of compliance review are significantly improved, while enabling companies to promptly identify and correct potential data inconsistencies and compliance risks, ensuring the accuracy of compliance checks on target documents.

[0088] Figure 2 This is a flowchart of an optional abnormal document identification method provided according to an embodiment of this application, such as... Figure 2As shown, the document type of the target document is first identified, and the review document and review model are selected according to the document type. The first review result is determined by the review document, and the second review result is determined by the review model. The review result of the target document is determined by combining the first review result and the second review result. Then, the review result is used to determine whether there are any anomalies in the target document and an anomaly identification result is generated.

[0089] Optionally, in the abnormal document identification method provided in this application embodiment, when the document type is a first type, comparing the target document and the review document to obtain a first review result includes: identifying the style of the target document and determining whether the style of the target document is the same as the standard style of the review document; if the style of the target document is different from the standard style, determining that the first review result is abnormal; if the style of the target document is the same as the standard style, obtaining the standard information corresponding to each feature information in the review document; if each feature information is consistent with the corresponding standard information, determining that the first review result is normal; if there are inconsistent feature information and standard information, determining that the first review result is abnormal.

[0090] Specifically, the standard style, which refers to the predefined layout and format of the audit document, is used for comparison with the style of the target document. Upon obtaining the initial audit result, the target document's style must first be identified, including its page layout, heading format, font size, and other elements. This style is then compared with the standard style of the audit document to determine if they are consistent, thus ensuring that the target and audit documents are format-comparable and preventing errors in information extraction or comparison due to style mismatches.

[0091] If inconsistencies are found between the target document and the standard style of the audited document during style comparison, it indicates that the target document has an anomaly in terms of style, such as the document style not conforming to regulations. In this case, the first audit result is automatically marked as "abnormal". This step can promptly identify formatting issues that may affect subsequent substantive content comparisons, avoiding invalid or erroneous comparison operations.

[0092] Once the style of the target document and the audit document are confirmed to be consistent, the system extracts standard information corresponding to the feature information of the target document from the audit document. The system then compares the feature information in the target document with the standard information obtained from the audit document one by one. If all feature information is consistent with the standard information, the first audit result is determined to be "normal", indicating that no problems were found in the consistency detection of the target document. Conversely, if any inconsistent feature information exists, the first audit result is marked as "abnormal", indicating that further inspection or correction is needed to determine whether there are any abnormalities in the content of the target document, and thus obtain the first audit result.

[0093] For example, assuming the target document is an invoice, the system identifies the invoice's format, including the location of the invoice amount and the invoice date, and retrieves the corresponding audited invoice. The system then compares the target document with the audited invoice's format. If the format is found to be correct, the content of the target document and the audited invoice are compared. If the content matches, the target document is considered correct; otherwise, an anomaly is identified, resulting in the first audit result.

[0094] This embodiment ensures a high degree of matching between the target document and the review document in terms of format and content through style recognition and consistency detection, thereby improving the accuracy and efficiency of automated review.

[0095] Optionally, in the abnormal document identification method provided in this application embodiment, when the document type is a first type, the review model is trained in the following manner: multiple first historical documents are obtained, and a second historical document corresponding to each first historical document is obtained, as well as the difference information between each group of first historical documents and second historical documents; a group of first historical documents, second historical documents and difference information are determined as a group of sample data to obtain multiple groups of first sample data, and the multiple groups of first sample data are used to train a preset neural network model to obtain the review model.

[0096] Specifically, before using the review model to review target documents, the model needs to be trained. First, the system extracts a large number of previously processed documents from the database as the first set of historical documents. Next, it finds the corresponding record document for each document, which serves as the second set of historical documents. Then, the system analyzes these paired historical documents, extracting the differences between them for subsequent modeling training. This accumulates a large dataset for training the review model, i.e., the training set, ensuring that the model can learn the consistency and difference features between documents from past real-world cases.

[0097] For example, the system collects 1,000 historical invoice documents and their corresponding financial records. Some details in the invoices (such as payment method and scope of services) may have slight differences from the corresponding descriptions in the financial records, while information such as invoice amount and invoice number should be completely consistent. The system records these discrepancies, including cases of mismatched amounts and inconsistent dates, as training material.

[0098] Furthermore, the previously collected historical documents and their corresponding financial records, along with discrepancy information, are packaged into multiple sets of sample data. Each set of sample data is a complete case, containing all necessary inputs (historical contracts and financial records) and expected outputs (discrepancy information). This step is crucial for building the model's training dataset, ensuring that the model can learn comprehensive comparison logic and discrepancy recognition capabilities from diverse cases.

[0099] Finally, the system uses multiple sets of initial sample data to train a pre-set neural network model, enabling the model to learn to identify and analyze differences between a set of documents, thus automatically detecting the consistency of new documents in the future. This process involves inputting a large amount of data and adjusting the model parameters until the model can accurately predict the difference information.

[0100] This embodiment constructs an audit model for consistency detection by collecting and utilizing a large number of historical documents and corresponding comparison documents as training data. Through this audit model, subtle differences between contract documents and financial records can be automatically identified, significantly improving the efficiency and accuracy of automated auditing.

[0101] Optionally, in the abnormal document identification method provided in the embodiments of this application, when the document type is the second type, comparing the target document and the review document to obtain the first review result includes: obtaining the review content in the review document, and reviewing the target document according to each review content in turn to obtain multiple review results; when multiple review results indicate that the target document is not abnormal, determining that the first review result is normal; when any review result indicates that the target document is abnormal, determining that the first review result is abnormal.

[0102] Specifically, when reviewing the second type of target documents, all review content can be extracted from the review documents. This content consists of pre-defined compliance check standards, including but not limited to financial standards, legal and regulatory requirements, or internal corporate policies. The target documents are then compared with the review content item by item, a compliance check is performed on each review item, and the review results for each check item are recorded. This process aims to ensure that the target documents meet the requirements in all compliance check items through meticulous item-by-item comparison, avoiding any potential violations being overlooked.

[0103] If all audit results for all items are "normal," the first audit result is determined to be "normal," indicating that the target document fully complies with compliance requirements. Conversely, if any item shows an "abnormal" result, even if the other items are normal, the first audit result will be marked as "abnormal," indicating that the target document is non-compliant in some aspect and requires further review or correction.

[0104] This embodiment obtains specific audit content from the audit document and audits the target document based on the audit content, enabling the system to conduct a comprehensive and in-depth compliance check on the target document, ensuring that every standard is followed, thereby guaranteeing the accuracy and compliance of the target document's content.

[0105] Optionally, in the abnormal document identification method provided in this application embodiment, when the document type is the second type, the review model is trained in the following manner: obtaining content review documents under each document type to obtain multiple content review documents, and configuring the multiple content review documents in a preset neural network model; obtaining multiple second historical documents and abnormal information of each second historical document; using each second historical document and the corresponding abnormal information as a set of sample data to obtain multiple sets of second sample data; using the multiple sets of second sample data to train the preset neural network model to obtain the review model.

[0106] Specifically, before using the review model to review the target document, the review model needs to be trained. First, the system collects all compliance standard documents related to the second type of document, which constitute a set of content review documents. Then, these content review documents are integrated into a preset neural network model. The purpose is to allow the model to be exposed to all compliance standards in the early stages of training, ensuring that the model can identify and judge the compliance of documents based on these standards during subsequent training.

[0107] Furthermore, the system extracts multiple second-type documents from the historical database, especially those documents that were found to have compliance issues in previous audits, along with the document's anomaly information, to provide the model with real-world examples. These examples form the training set for the training model, enabling it to learn the characteristics of compliance and non-compliance from historical data, thus achieving higher accuracy in identifying similar issues in the future.

[0108] Finally, the system uses the aforementioned second sample data to train the preset neural network model. By adjusting the model parameters, the model can learn how to identify abnormal information in the second historical document, thereby judging the compliance of the new document and ensuring that the model can accurately detect compliance issues in the second type of document.

[0109] This embodiment provides an intelligent solution for compliance review of second-type documents by training a neural network model under a deep learning framework. The model not only absorbs the standard information on compliance in the content review documents, but also learns to identify and judge the characteristics of non-compliant behavior through training on historical abnormal cases, so that the review of second-type documents can be carried out efficiently and accurately in the future.

[0110] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0111] This application also provides an abnormal document identification device. It should be noted that the abnormal document identification device of this application can be used to execute the abnormal document identification method provided in this application. The abnormal document identification device provided in this application is described below.

[0112] Figure 3 This is a schematic diagram of an abnormal document identification device provided according to an embodiment of this application. For example... Figure 3 As shown, the device includes: a first identification unit 31, a second identification unit 32, a selection unit 33, a comparison unit 34, and a first determination unit 35.

[0113] The first identification unit 31 is used to acquire the target document and identify the document type of the target document.

[0114] The second identification unit 32 is used to identify M feature information in the target document according to the document type, where M is a positive integer.

[0115] Unit 33 is used to select an audit model and an audit document based on the document type. The audit model and audit document are used to audit the content of the target document.

[0116] The comparison unit 34 is used to compare the target document and the review document to obtain the first review result, and input M feature information into the review model to obtain the second review result.

[0117] The first determining unit 35 is used to determine the abnormal review result of the target document based on the first review result and the second review result.

[0118] The abnormal document identification device provided in this application embodiment includes: a first identification unit 31 for acquiring a target document and identifying its document type; a second identification unit for identifying M feature information in the target document based on its document type, where M is a positive integer; a selection unit 32 for selecting an audit model and an audit document based on the document type, wherein the audit model and audit document are used to audit the content of the target document; a comparison unit 33 for comparing the target document and the audit document to obtain a first audit result, and inputting the M feature information into the audit model to obtain a second audit result; and a first determination unit 34 for determining the abnormal audit result of the target document based on the first and second audit results. This solves the problem of low efficiency in manual document auditing in related technologies. By identifying the document type and determining the document and model for auditing the target document based on its feature information, and then processing the feature information in the target document using the document and model to obtain the audit result, the device achieves the technical effect of improving the efficiency of document auditing while ensuring the accuracy of document auditing.

[0119] Optionally, in the abnormal document identification device provided in this application embodiment, the first identification unit 31 includes: a first determining module, used to determine multiple preset locations and sequentially determine whether document information exists at each preset location in the target document; a second determining module, used to sequentially determine the reference information at each of the N target locations where document information exists, to obtain N sets of reference information, where N is a positive integer; a comparison module, used to compare a set of reference information and document information at the same target location to obtain N comparison results, wherein the comparison results are used to characterize whether there is matching information associated with document information in the set of reference information; and a third determining module, used to determine the document type of the target document based on the N comparison results.

[0120] Optionally, in the abnormal document identification device provided in this application embodiment, the third determining module includes: a first obtaining submodule, used to obtain the document type indicated by the matching information when there is a comparison result representing a set of benchmark information containing matching information associated with document information, to obtain H document type sets, where H is a positive integer and H is less than or equal to N; a first determining submodule, used to determine the document types contained in the document type sets as the document type of the target document when H is a preset number; a second obtaining submodule, used to obtain the intersection document types of the H document type sets when H is greater than a preset number, to obtain the document type of the target document; and a second determining submodule, used to determine the document type with the largest number in the H document type sets as the document type of the target document when the intersection of the H document type sets is empty.

[0121] Optionally, in the abnormal document identification device provided in this application embodiment, the second identification unit 32 includes: a fourth determining module, used to determine the document style of the target document according to the document type; a first obtaining module, used to obtain a preset template corresponding to the document style from the template library according to the document style, wherein the style of the preset template is the same as the document style; and a second obtaining module, used to determine the features to be obtained marked in the preset template, and obtain the feature information under each feature to be obtained in the target document through the identification model to obtain M feature information.

[0122] Optionally, in the abnormal document identification device provided in this application embodiment, the selection unit 33 includes: a fifth determining module, used to determine the document used for consistency detection with the target document as an audit document and the model used for consistency detection of the target document as an audit model when the document type is a first type; and a sixth determining module, used to determine the document used for content review of the target document as an audit document and the model used for compliance detection of the target document as an audit model when the document type is a second type.

[0123] Optionally, in the abnormal document identification device provided in this application embodiment, when the document type is a first type, the comparison unit 34 includes: an identification module, used to identify the style of the target document and determine whether the style of the target document is the same as the standard style of the audited document; a seventh determination module, used to determine that the first audit result is abnormal when the style of the target document is not the same as the standard style; a third acquisition module, used to acquire the standard information corresponding to each feature information in the audited document when the style of the target document is the same as the standard style; an eighth determination module, used to determine that the first audit result is normal when each feature information is consistent with the corresponding standard information; and a ninth determination module, used to determine that the first audit result is abnormal when there are inconsistent feature information and standard information.

[0124] Optionally, in the abnormal document identification device provided in this application embodiment, when the document type is a first type, the review model is trained in the following manner: a first acquisition unit is used to acquire multiple first historical documents, and acquire a second historical document corresponding to each first historical document, as well as the difference information between each group of first historical documents and second historical documents; a second determination unit is used to determine a group of first historical documents, second historical documents and difference information as a group of sample data, obtain multiple groups of first sample data, and use the multiple groups of first sample data to train a preset neural network model to obtain the review model.

[0125] Optionally, in the abnormal document identification device provided in this application embodiment, when the document type is the second type, the comparison unit 34 includes: a fourth acquisition module, used to acquire the review content in the review document, and sequentially review the target document according to each review content to obtain multiple review results; a tenth determination module, used to determine the first review result as normal when multiple review results indicate that the target document is not abnormal; and an eleventh determination module, used to determine the first review result as abnormal when any one review result indicates that the target document is abnormal.

[0126] Optionally, in the abnormal document identification device provided in this application embodiment, when the document type is the second type, the review model is trained in the following manner: a second acquisition unit is used to acquire content review documents under each document type, obtain multiple content review documents, and configure the multiple content review documents in a preset neural network model; a third acquisition unit is used to acquire multiple second historical documents and abnormal information of each second historical document; a third determination unit is used to take each second historical document and the corresponding abnormal information as a set of sample data to obtain multiple sets of second sample data; and a training unit is used to train the preset neural network model using the multiple sets of second sample data to obtain the review model.

[0127] The aforementioned abnormal document identification device includes a processor and a memory. The first identification unit 31, the second identification unit 32, the selection unit 33, the comparison unit 34, the first determination unit 35, etc., are all stored in the memory as program units. The processor executes the aforementioned program units stored in the memory to realize the corresponding functions.

[0128] The processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured; adjusting kernel parameters addresses the inefficiency of manual document review in related technologies.

[0129] The memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0130] This invention provides a computer-readable storage medium storing a program that, when executed by a processor, implements the method for identifying abnormal documents.

[0131] This invention provides a processor for running a program, wherein the program executes the method for identifying the abnormal document during runtime.

[0132] Figure 4This is a schematic diagram of an electronic device provided according to an embodiment of this application, such as... Figure 4 As shown, this embodiment of the invention provides an electronic device 40, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the above-described method for identifying abnormal documents. The device in this document can be a server, PC, PAD, mobile phone, etc.

[0133] This application also provides a computer program product that, when executed on a data processing device, is adapted to perform the steps of initializing the identification method for the aforementioned abnormal documents.

[0134] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0135] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0136] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0137] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0138] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0139] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, like read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0140] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0141] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0142] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for identifying abnormal documents, characterized in that, include: Obtain the target document and identify its document type; Based on the document type, identify M feature information in the target document, where M is a positive integer; A review model and a review document are selected based on the document type, wherein the review model and the review document are used to review the content of the target document; The target document and the document to be reviewed are compared to obtain a first review result. The M feature information are then input into the review model to obtain a second review result. The abnormal review result of the target document is determined based on the first review result and the second review result.

2. The method according to claim 1, characterized in that, The document types for identifying the target document include: Multiple preset locations are determined, and the presence of document information at each preset location is sequentially determined in the target document. For N target locations where the document information exists, the reference information under each target location is determined sequentially to obtain N sets of reference information, where N is a positive integer; A set of benchmark information and document information at the same target location are compared to obtain N comparison results, wherein the comparison results are used to characterize whether there is matching information in the set of benchmark information that is associated with the document information; The document type of the target document is determined based on the N comparison results.

3. The method according to claim 2, characterized in that, The document type of the target document is determined based on the N comparison results, including: If there is a comparison result in which matching information associated with the document information exists in a set of benchmark information, the document type indicated by the matching information is obtained, resulting in a set of H document types, where H is a positive integer and H is less than or equal to N; When H is a preset quantity, the document types contained in the document type set are determined as the document types of the target document; If H is greater than the preset number, obtain the intersection document types of the H document type sets to obtain the document type of the target document; If the intersection of the H document type sets is empty, the document type with the largest number of occurrences in the H document type sets shall be determined as the document type of the target document.

4. The method according to claim 1, characterized in that, Identifying M feature information in the target document based on the document type includes: The document style of the target document is determined based on the document type; According to the document style, a preset template corresponding to the document style is obtained from the template library, wherein the style of the preset template is the same as the document style; The features to be acquired are identified in the preset template, and the feature information under each feature to be acquired in the target document is obtained through the recognition model, thus obtaining the M feature information.

5. The method according to claim 1, characterized in that, The review model and review documents selected based on the document type include: When the document type is the first type, the document used for consistency detection with the target document is determined as the audit document, and the model for consistency detection of the target document is determined as the audit model; When the document type is the second type, the document used to review the content of the target document is identified as the review document, and the model used to perform compliance testing on the target document is identified as the review model.

6. The method according to claim 5, characterized in that, When the document type is the first type, comparing the target document and the review document to obtain the first review result includes: Identify the style of the target document and determine whether the style of the target document is the same as the standard style of the document to be reviewed; If the style of the target document is different from the standard style, the first review result is determined to be abnormal. If the style of the target document is the same as the standard style, obtain the standard information corresponding to each feature information in the audit document; If each feature matches the corresponding standard information, the first audit result is determined to be normal. In the presence of inconsistent feature information and standard information, the first audit result is determined to be abnormal.

7. The method according to claim 5, characterized in that, When the document type is the first type, the review model is trained in the following manner: Retrieve multiple first historical documents, and retrieve the second historical document corresponding to each first historical document, as well as the difference information between each group of first historical documents and second historical documents; A set of first historical documents, second historical documents, and difference information are identified as a set of sample data to obtain multiple sets of first sample data. The multiple sets of first sample data are then used to train a preset neural network model to obtain the audit model.

8. The method according to claim 5, characterized in that, When the document type is the second type, comparing the target document and the review document to obtain the first review result includes: The review content in the document to be reviewed is obtained, and the target document is reviewed in turn according to each review content to obtain multiple review results; If all the audit results indicate that the target document is without abnormalities, the first audit result is determined to be normal. If any audit result indicates that the target document is abnormal, the first audit result is determined to be abnormal.

9. The method according to claim 5, characterized in that, When the document type is the second type, the review model is trained in the following manner: Obtain content-reviewed documents for each document type to obtain multiple content-reviewed documents, and configure the multiple content-reviewed documents in a preset neural network model; Retrieve multiple second historical documents, and the exception information for each second historical document; Each second historical document and its corresponding anomaly information are used as a set of sample data to obtain multiple sets of second sample data. The preset neural network model is trained using the multiple sets of second sample data to obtain the audit model.

10. An anomalous document identification device, characterized in that, include: The first identification unit is used to acquire the target document and identify the document type of the target document; The second identification unit is used to identify M feature information in the target document according to the document type, where M is a positive integer; The selection unit is used to select an audit model and an audit document according to the document type, wherein the audit model and the audit document are used to audit the content of the target document; The comparison unit is used to compare the target document and the review document to obtain a first review result, and input the M feature information into the review model to obtain a second review result; The first determining unit is used to determine the abnormal review result of the target document based on the first review result and the second review result.

Images