Document attachment risk detection method, device and equipment and readable storage medium

Abstract

The invention relates to a document attachment risk detection method and device, computer equipment, a computer readable storage medium and a computer program product. The method comprises the steps of obtaining a target attachment associated with a target document, and determining an attachment content type of the target attachment; determining an information detection element and a risk detection rule matched with the attachment content type; through a large language model, accessory detection information is extracted from the target accessory according to the information detection elements, and the accessory detection information comprises information, corresponding to the information detection elements, in the target accessory; and performing information risk detection based on the attachment detection information, the risk detection rule and the document information of the target document through a large language model to obtain a risk detection result of the target attachment. By adopting the method, the risk detection efficiency of the receipt attachments can be improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a document attachment risk detection method, apparatus, computer equipment, computer-readable storage medium and computer program product. Background Art

[0002] With the expansion of enterprise scale and the complexity of business, the number of financial reimbursement documents has increased dramatically. As an important part of reimbursement vouchers, the authenticity, integrity and compliance of document attachments are crucial to ensure the accuracy of financial information and prevent fraud, and are of great significance to the risk control of enterprises and organizations. However, there are many types of document attachments, which makes it difficult to identify risks. The risk detection operation for document attachments is cumbersome, resulting in low efficiency of risk detection for document attachments. Summary of the invention

[0003] Based on this, it is necessary to provide a document attachment risk detection method, device, computer equipment, computer readable storage medium and computer program product that can improve the risk detection efficiency of document attachments in response to the above technical problems.

[0004] In a first aspect, the present application provides a document attachment risk detection method, comprising:

[0005] Obtain the target attachment associated with the target document, and determine the attachment content type of the target attachment;

[0006] Determine information detection elements and risk detection rules that match the type of attachment content;

[0007] Extracting attachment detection information from the target attachment according to the information detection element through the large language model, where the attachment detection information includes information in the target attachment corresponding to the information detection element;

[0008] Through the large language model, information risk detection is performed based on attachment detection information, risk detection rules and document information of the target document to obtain the risk detection result of the target attachment.

[0009] In a second aspect, the present application also provides a document attachment risk detection device, comprising:

[0010] The attachment acquisition module is used to acquire the target attachment associated with the target document and determine the attachment content type of the target attachment;

[0011] A content type processing module, used to determine information detection elements and risk detection rules that match the attachment content type;

[0012] A detection information extraction module, used to extract attachment detection information from the target attachment according to the information detection elements through a large language model, where the attachment detection information includes information in the target attachment corresponding to the information detection elements;

[0013] The risk detection module is used to perform information risk detection based on attachment detection information, risk detection rules and document information of the target document through a large language model to obtain the risk detection result of the target attachment.

[0014] In a third aspect, the present application further provides a computer device, wherein the computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the document attachment risk detection method are implemented.

[0015] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the document attachment risk detection method described above are implemented.

[0016] In a fifth aspect, the present application further provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps of the document attachment risk detection method are implemented.

[0017] The document attachment risk detection method, device, computer equipment, computer-readable storage medium and computer program product described above determine the attachment content type of the target attachment associated with the target document, extract attachment detection information from the target attachment according to the information detection elements that match the attachment content type through a large language model, and perform information risk detection based on the attachment detection information, risk detection rules that match the attachment content type and the document information of the target document to obtain the risk detection result of the target attachment. In the risk detection processing of the document attachment, the attachment detection information is extracted from the target attachment according to the information detection elements through a large language model, and information risk detection is performed based on the attachment detection information, risk detection rules and the document information of the target document, so that accurate risk detection can be performed by only configuring the information detection elements and risk detection rules that match the attachment content type, simplifying the document attachment detection operation, thereby improving the risk detection efficiency of the document attachment. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 This is an application environment diagram of a document attachment risk detection method in an embodiment;

[0020] Figure 2 A schematic diagram of a process of detecting risks of document attachments in one embodiment;

[0021] Figure 3 A schematic diagram of a process for extracting attachment detection information in one embodiment;

[0022] Figure 4 A schematic diagram of a flow chart of a document attachment risk detection method in yet another embodiment;

[0023] Figure 5 A schematic diagram of setting details of audit elements in an embodiment;

[0024] Figure 6 A schematic diagram of an interface for reviewing factor extraction results in an embodiment;

[0025] Figure 7 A schematic diagram of a process for extracting audit elements from different types of attachments in one embodiment;

[0026] Figure 8 A schematic diagram of a process for reviewing risk points according to different rules in an embodiment;

[0027] Fig. 9 A schematic diagram of an interface of risk detection results in one embodiment;

[0028] Fig.10 A schematic diagram of display configuration of picture type attachments in one embodiment;

[0029] Fig.11 is a structural block diagram of a document attachment risk detection device in an embodiment;

[0030] Fig.12 is an internal structure diagram of a computer device in one embodiment;

[0031] Fig.13 FIG. 4 is a diagram showing the internal structure of a computer device in yet another embodiment. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0033] The document attachment risk detection method provided in the embodiment of the present application can be applied to Figure 1In the application environment shown, the terminal 102 communicates with the server 104 through a network. The data storage system can store data that the server 104 needs to process. The data storage system can be integrated on the server 104, or placed on the cloud or other network servers.

[0034] The user can import the target attachment associated with the target document through the terminal 102. Specifically, the user can import the target attachment through the human-computer interaction interface of the terminal 102. The terminal 102 can upload the target attachment to the server 104 so that the server 104 can perform risk detection on the target attachment. Specifically, the server 104 can determine the attachment content type of the target attachment, and determine the information detection elements and risk detection rules that match the attachment content type. The information detection elements and risk detection rules can be configured by the user in advance according to actual needs. The server 104 can extract attachment detection information from the target attachment according to the information detection elements that match the attachment content type through a large language model. The attachment detection information includes information corresponding to the information detection elements in the target attachment. The server 104 performs information risk detection based on the attachment detection information, the risk detection rules that match the attachment content type, and the document information of the target document to obtain the risk detection result of the target attachment. The server 104 can also return the risk detection result of the target attachment to the terminal 102 so that the terminal 102 can display the risk detection result to the user through the human-computer interaction interface.

[0035] In some embodiments, the document attachment risk detection method can also be implemented directly by the terminal 102 or the server 104 alone, that is, the terminal 102 can directly perform risk detection processing on the target attachment imported by the user, or the server 104 can obtain the target attachment associated with the target document from the data storage system, and perform risk detection processing on the obtained target attachment.

[0036] The terminal 102 may be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, IoT devices, and portable wearable devices. The IoT devices may be smart speakers, smart TVs, smart air conditioners, smart car devices, projection devices, etc. Portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc. Head-mounted devices may be virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, etc. The server 104 may be an independent physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides cloud computing services.

[0037] In an exemplary embodiment, Figure 2As shown, a document attachment risk detection method is provided, and the method is applied to Figure 1 The server in the example is used to illustrate, including the following steps 202 to 208. Among them:

[0038] Step 202: Acquire the target attachment associated with the target document, and determine the attachment content type of the target attachment.

[0039] Among them, the document is a written proof obtained or filled in when the economic business occurs, which records the actual situation of the transaction and the matter. It is the original data and important basis for accounting. The document can independently explain the full picture of the matter and has independent legal effect. It can be used as an original voucher and directly used in accounting and auditing. The document can specifically include but is not limited to at least one of invoices, receipts, material requisition forms, warehouse entry forms, warehouse exit forms or expense reimbursement forms. Attachments refer to supporting documents that are closely related to a certain subject and supplement the functions of the subject. Attachments are documents formulated together with the main document. They are supplementary explanations of the main document and cannot exist independently. They must be used together with the main document to explain the full picture of the matter. Attachments cannot be used as original vouchers independently. They must be used together with the main document to be used as original vouchers. The main function of the attachment is to supplement and explain the contents of the main document to make the contents of the main document more complete and accurate. The attachments of documents are usually used to supplement the details of matters such as expense reimbursement and income recognition in order to better carry out accounting and auditing. Attachments can include but are not limited to at least one of contracts, bank receipts, invoice lists, approval forms, photos or pictures. Document attachments are important vouchers for economic activities such as financial reimbursement and business transactions. Their authenticity, integrity and compliance are directly related to the financial security, business compliance and information security of the enterprise. Risk detection of document attachments is an important measure to protect the interests of the enterprise and prevent risks. Risk detection of document attachments may include but is not limited to authenticity verification, integrity check, compliance review or abnormal behavior monitoring.

[0040] The target document is the document that needs to be tested for attachment risk, and the target attachment is the attachment associated with the target document, which can be a document that provides supplementary explanations for the document. The attachment content type refers to the type divided according to the attachment content, which can include train tickets, flight itineraries, hotel bills, value-added tax invoices, taxi tickets, or online car-hailing itineraries.

[0041] Optionally, the server can obtain the target attachment that needs to be processed for risk detection. The target attachment can be a file that provides supplementary instructions for the target document. In some embodiments, the server can directly receive the target attachment uploaded and sent by the user from the terminal, and the server can also query based on the target document to query the target attachment associated with the target document. The server can classify the content types of the target attachment, specifically according to the name, source, attachment content, etc. of the target attachment, to determine the attachment content type of the target attachment. For attachments including different types of content, risk detection can be performed according to different risk detection rules.

[0042] Step 204: Determine information detection elements and risk detection rules that match the type of attachment content.

[0043] Among them, the information detection factor is the information item in the attachment that needs to be detected. The information detection factor matches the type of attachment content, that is, attachments belonging to different types of attachment content need to be risk-detected according to the information of different information items. The risk detection rule is the rule for risk determination of the information in the attachment. The risk detection rule can be used to determine whether there is risk information in the attachment. The information detection factor and risk detection rule can be set according to actual needs, and can be specifically configured according to the type of attachment content.

[0044] Exemplarily, the server can determine the corresponding matching information detection elements and risk detection rules based on the type of attachment content. In some embodiments, the user can configure the corresponding information detection elements and risk detection rules for different types of attachment content according to the needs of the actual application scenario, and establish a mapping relationship between the attachment content type and the information detection elements and risk detection rules. After the server determines the attachment content type of the target attachment, the server can query based on the mapping relationship according to the attachment content type to determine the information detection elements and risk detection rules that match the attachment content type. For example, if the attachment content type is an airplane itinerary, the corresponding matching information detection elements may include information items such as the traveler, time, departure and arrival city, seat, total amount, whether there is an invoice, invoice type, invoice number, and total invoice amount. The risk detection rules may include determination conditions for risk determination of information for each information item, such as whether the total amount exceeds the amount threshold, whether the seat belongs to a specified type, and other determination conditions for risk determination of information for various information items.

[0045] In some embodiments, the server can also determine the application scenario to which the target attachment belongs, such as determining the user account that triggers risk detection for the target attachment, and determining the corresponding information detection elements and risk detection rules based on the user account and the type of attachment content, so that different information detection elements and risk detection rules can be configured for different user accounts to adapt to the actual application needs of various users.

[0046] Step 206: extract attachment detection information from the target attachment according to the information detection element through the large language model. The attachment detection information includes information in the target attachment corresponding to the information detection element.

[0047] Among them, Large Language Model (LLM) is a type of deep learning model with a large number of parameters. They learn language patterns, grammar and semantics by processing a large amount of text data, so as to understand and generate human language. Large language models can specifically include but are not limited to various models including GPT (Generative Pretrained Transformer), PaLM (Pathways Language Model), LLaMA, etc. Attachment detection information is information in the target attachment used for risk detection, specifically including information corresponding to information detection elements, that is, attachment detection information is information extracted from the target attachment according to the information detection elements.

[0048] Optionally, the server can extract attachment detection information that needs to be risk-detected from the target attachment by calling the large language model. Specifically, the server can extract information from the target attachment according to the information detection elements through the large language model to extract the attachment detection information. The attachment detection information includes information in the target attachment corresponding to the information detection elements, that is, the attachment detection information includes information corresponding to the information items that need to be risk-detected. In a specific application, the information detection element may include at least one, and the large language model can extract information from the target attachment one by one according to the information detection elements to extract element information belonging to each information detection element from the target attachment. The large language model can obtain the attachment detection information based on the element information combination of each information detection element. For example, if the target attachment is an itinerary for user A's business trip by plane, the corresponding matching information detection elements may include information items such as the traveler, time, departure and arrival cities, seats, and total amount, and the attachment detection information may include specific information corresponding to each information detection element in user A's itinerary, such as the traveler's name (such as user A), specific travel time (such as September 1, 2024), seat type (such as business class), total amount (such as 1,500 yuan), and other information.

[0049] Step 208 , using a large language model, information risk detection is performed based on the attachment detection information, risk detection rules, and document information of the target document to obtain a risk detection result of the target attachment.

[0050] The document information refers to the information carried in the target document, which can be extracted from the target document, and can include but is not limited to various types of information including invoice name, invoice code, invoice number, invoice date, transaction content, tax payment information, etc. The risk detection result is the detection result obtained by performing risk detection on the target attachment, and the risk detection result can indicate whether the target attachment has risks.

[0051] Exemplarily, the server can perform information risk detection on the attachment detection information of the target attachment through the large language model to determine whether the target attachment has risks based on the attachment detection information and obtain corresponding risk detection results. The server can perform information risk detection based on the attachment detection information, risk detection rules and document information of the target document by calling the large language model. Specifically, the large language model can perform information comparison on the attachment detection information in accordance with the risk detection rules and in combination with the document information of the target document to determine whether the internal information in the attachment detection information is consistent and whether the attachment detection information is consistent with the document information, thereby achieving information risk detection on the target attachment and obtaining corresponding risk detection results.

[0052] In some embodiments, different risk weights can be configured for different risk detection rules. After the large language model performs information risk detection according to the risk detection rules, it can perform weighted fusion based on the detection results of different risk detection rules and their respective risk weights to obtain weighted risk detection results. The risk detection results can be represented in the form of values ​​or levels. For example, the risk detection results can include various forms such as a risk score of 90 points and a risk level of low risk.

[0053] In the above document attachment risk detection method, the attachment content type of the target attachment associated with the target document is determined, and the attachment detection information is extracted from the target attachment according to the information detection elements that match the attachment content type through the large language model, and information risk detection is performed based on the attachment detection information, the risk detection rules that match the attachment content type, and the document information of the target document to obtain the risk detection result of the target attachment. For the risk detection processing of document attachments, the attachment detection information is extracted from the target attachment according to the information detection elements through the large language model, and information risk detection is performed based on the attachment detection information, the risk detection rules, and the document information of the target document, so that accurate risk detection can be performed by only configuring the information detection elements and risk detection rules that match the attachment content type, which simplifies the document attachment detection operation and improves the risk detection efficiency of document attachments.

[0054] In an exemplary embodiment, Figure 3 As shown, the process of extracting attachment detection information, that is, extracting attachment detection information from the target attachment according to information detection elements through a large language model, includes steps 302 to 306. Among them:

[0055] Step 302: extracting original attachment information from the target attachment in accordance with an information extraction method adapted to the attachment format type of the target attachment.

[0056] Among them, the information extraction method refers to the method of performing preliminary information extraction on the target attachment, and the information extraction method can be adapted to the attachment format type of the target attachment. The attachment format type refers to the type of the file format of the target attachment, which can specifically include image type and non-image type. The non-image type can include various types such as text format and database format. The original attachment information is the information extracted from the target file by the information extraction method. The original attachment information can include information in the target attachment that needs to be detected for risks, and can also include information that does not need to be detected for risks, such as advertising information in the target attachment.

[0057] Optionally, the server may determine the attachment format type of the target attachment. Specifically, the server may make a determination based on the file format of the target attachment, such as determining the attachment format type of the target attachment based on the file name of the target attachment, the attachment content, etc. Based on the attachment format type of the target attachment, the server may determine an appropriate information extraction method, thereby extracting the original attachment information from the target attachment through the appropriate information extraction method. For example, for a target attachment of a picture type, the server may perform information extraction based on OCR (Optical Character Recognition) technology to obtain the original attachment information in the target attachment.

[0058] Step 304: extract prompt words based on the attachment format type and the information detection element configuration information.

[0059] Among them, the information extraction prompt words are used as the model input of the large language model, and specifically may include sentence text expressed in natural language. Exemplarily, the server may configure the information extraction prompt words for the large language model based on the attachment format type and the information detection elements, so as to input them into the large language model, and the large language model will perform information extraction processing on them. In specific applications, different attachment format types may correspond to different prompt word templates, that is, the server may obtain the prompt word template corresponding to the attachment format type, and update the information detection element configuration to the prompt word template, thereby obtaining the information extraction prompt word. For example, the information extraction prompt words may include a natural language expression sentence of "Please extract the corresponding information from the original attachment information and output it according to the following information detection elements".

[0060] Step 306: Input the information extraction prompt word and the original attachment information into the large language model to extract detection information, and obtain the attachment detection information output by the large language model.

[0061] Optionally, the server may input the information extraction prompt word and the original attachment information into the large language model to instruct the large language model to extract the detection information. Specifically, the large language model may extract the original attachment information according to the information extraction requirements in the information extraction prompt word to obtain the attachment detection information. The large language model may also output the extracted attachment detection information according to the information output requirements in the information extraction prompt word. For example, the large language model may typeset and output the attachment detection information in the order specified in the information extraction prompt word.

[0062] In this embodiment, the server can extract the original attachment information from the target attachment in accordance with an information extraction method that is compatible with the attachment format type of the target attachment, extract prompt words based on the attachment format type and information detection element configuration information, and input the information extraction prompt words and the original attachment information into the large language model to obtain the attachment detection information output by the large language model, so that the original attachment information can be accurately and quickly extracted, and the attachment detection information can be extracted based on the large language model, which simplifies the operation of information extraction in document attachment detection, thereby improving the risk detection efficiency of document attachments.

[0063] In an exemplary embodiment, original attachment information is extracted from the target attachment in accordance with an information extraction method adapted to the attachment format type of the target attachment, including: determining the attachment format type of the target attachment; when the attachment format type includes an image type, extracting original attachment information from the target attachment based on character recognition technology; when the attachment format type includes a non-image type, extracting original attachment information from the target attachment based on file vectorization technology.

[0064] Among them, the attachment format type may include image type and non-image type. The image type may include but is not limited to jpg, jpeg, png, bmp, OFD or single-page pdf format types, and the non-image type may include but is not limited to multi-page pdf, txt, json library and other types. Character recognition technology, namely OCR technology, is a technology that converts printed or handwritten text into editable text. OCR technology converts the text in the image into computer-recognizable text by scanning or photographing the image and then using image processing and pattern recognition algorithms. File vectorization technology is a technology that converts non-numeric data into numerical vectors. Specifically, it can convert unstructured data into structured numerical vector representation. Through vectorization, the data in the file can be encoded into a series of numerical values ​​that can reflect the characteristics and semantic information of the data, so that the data in the file can be effectively processed and analyzed by computers and machine learning models.

[0065] Exemplarily, the server can determine the attachment format type of the target attachment. Specifically, the server can analyze the attachment format of the target attachment, and specifically analyze the file suffix name, file content, etc. of the template attachment to determine the attachment format type of the target attachment. When it is determined that the attachment format type of the target attachment includes a picture type, it indicates that the target attachment is a picture or a file in a single-page PDF format. The server can extract information from the target attachment based on character recognition technology to obtain the original attachment information of the target attachment. When it is determined that the attachment format type of the target attachment includes a non-picture type, it indicates that the target attachment is a file in a multi-page PDF, txt, json library, etc. format. The server can extract information from the target attachment based on file vectorization technology to obtain the original attachment information of the target attachment.

[0066] In this embodiment, for target attachments of image type, the server can extract the original attachment information based on character recognition technology, and for target attachments of non-image type, the server can extract the original attachment information based on file vectorization technology. This can accurately and quickly extract the original attachment information of the target attachment for target files of image type and non-image type, which is beneficial to improving the risk detection efficiency of document attachments.

[0067] In an exemplary embodiment, information risk detection is performed through a large language model based on attachment detection information, risk detection rules, and document information of a target document to obtain a risk detection result for the target attachment, including: obtaining the document information of the target document; configuring risk detection prompt words based on the attachment detection information, risk detection rules, and document information; inputting the risk detection prompt words into the large language model to perform information risk detection, and obtaining a risk detection result for the target attachment output by the large language model.

[0068] The document information is information extracted from the target document. The risk detection prompt words are used as model inputs of the large language model, and may specifically include sentence texts expressed in natural language. The risk detection prompt words may be configured based on the attachment detection information, risk detection rules, and document information.

[0069] Exemplarily, the server can obtain the document information of the target document, and specifically can extract information from the target document in advance to extract the document information from the target document. The server can configure prompt words based on the attachment detection information, risk detection rules and document information to configure risk detection prompt words. For example, the risk detection prompt words can include an expression sentence of "please follow the risk detection rules, refer to the document information and the attachment detection information, and compare the attachment detection information one by one to determine the risk of the attachment detection information." The server can use the risk detection prompt words as the input of the large language model and input them into the large language model so that the large language model can perform information risk detection on the risk detection prompt words and output the risk detection result for the target attachment. In some embodiments, when the document information of the target document is not obtained, the server can directly configure the risk detection prompt words based on the attachment detection information and the risk detection rules, and input the configured risk detection prompt words into the large language model for information risk detection, and obtain the risk detection result for the target attachment output by the large language model.

[0070] In this embodiment, the server configures risk detection prompt words based on attachment detection information, risk detection rules and document information of the target document, and obtains risk detection results for the target attachment by inputting the risk detection prompt words into a large language model for information risk detection. This allows accurate risk detection to be performed by only configuring information detection elements and risk detection rules that match the type of attachment content, thereby simplifying the document attachment detection operation and improving the risk detection efficiency of document attachments.

[0071] In an exemplary embodiment, the target attachments include at least two, and the document attachment risk detection method further includes: configuring risk detection prompt words based on the attachment detection information and risk detection rules of each target attachment; inputting the risk detection prompt words into the large language model for information risk detection, and obtaining the risk detection results for each target attachment output by the large language model.

[0072] Exemplarily, the target attachments may include multiple ones, and the document attachment risk detection may be directly performed based on the multiple target attachments. After the server traverses each target attachment, it may obtain the attachment detection information of each target attachment. The server may directly configure the risk detection prompt words based on the attachment detection information and risk detection rules of each target attachment, and input the configured risk detection prompt words into the large language model for information risk detection, and obtain the risk detection results for each target attachment output by the large language model, so that accurate risk detection may be performed directly based on the attachment detection information and risk detection rules of each target attachment, thereby improving the risk detection efficiency of the document attachments.

[0073] In an exemplary embodiment, the document attachment risk detection method also includes: when the attachment format type of the target attachment includes a picture type, adding the attachment identifier of the target attachment to the attachment detection information to obtain attachment summary information for the target attachment; extracting the attachment identifier from the attachment summary information, and intercepting the attachment style information from the target attachment according to the attachment identifier; and displaying the attachment style information and the attachment detection information in the model output display area of ​​the large language model.

[0074] Among them, the attachment identifier is used to identify the target attachment, and specifically may include various identification information such as the ID number of the target attachment, the attachment name, etc. The attachment summary information is obtained by adding the attachment identifier to the attachment detection information, that is, the attachment summary information includes the attachment detection information and the attachment identifier of the target attachment. The attachment style information is information that reflects the intuitive style of the target attachment, and specifically may include but is not limited to style information such as the title and amount of the target attachment. The model output display area is an area used to display the output of the large language model, and specifically may include the human-computer interaction interface displayed in the terminal.

[0075] Optionally, for a target attachment of a picture type, that is, when it is determined that the attachment format type of the target attachment includes a picture type, the server can determine the attachment identifier of the target attachment, and add the attachment identifier to the attachment detection information, thereby binding the attachment detection information with the attachment identifier to obtain the attachment summary information. The attachment summary information includes both the attachment detection information and the attachment identifier of the target attachment. When the attachment detection information needs to be displayed, the server can extract the attachment identifier from the attachment summary information, and intercept the attachment style information from the target attachment according to the attachment identifier. In some embodiments, the server can extract the attachment identifier from the attachment summary information, and query from the attachment library including each attachment according to the attachment identifier to query the target attachment identified by the attachment identifier. For the queried target attachment, the server can intercept the attachment style information, such as the title and numerical information in the target attachment. The server can determine the model output display area of ​​the large language model, and display the attachment style information and the attachment detection information in the model output display area to intuitively display the attachment summary information for the target attachment.

[0076] In this embodiment, the server obtains attachment summary information by adding the attachment identifier of the target attachment to the attachment detection information. When the attachment summary information needs to be displayed intuitively, the server determines the target attachment according to the attachment identifier in the attachment summary information, and extracts the attachment style information therefrom, thereby displaying the attachment style information and the attachment detection information in the model output display area of ​​the large language model. The style of the target attachment can be retained to intuitively display the information in the target attachment, which is conducive to improving the information display effect of the target attachment.

[0077] In an exemplary embodiment, the document attachment risk detection method further includes: displaying risk detection rules and risk detection results in a first output display area of ​​the large language model; and displaying attachment detection information in a second output display area of ​​the large language model.

[0078] Among them, the output of the large language model can be displayed in its associated model output display area. The model output display area of ​​the large language model may include a first output display area and a second output display area to respectively display risk detection rules, risk detection results and attachment detection information.

[0079] Exemplarily, the server may determine a first output display area of ​​the large language model, and render and display the risk detection rules and risk detection results in the first output display area. Specifically, for each risk detection rule, the server may display the corresponding risk detection result accordingly. For attachment detection information, the server may render and display the attachment detection information in a second output display area of ​​the large language model. In some embodiments, the first output display area and the second output display area may belong to the same interface or to different interfaces.

[0080] In this embodiment, the server displays the risk detection rules, risk detection results, and attachment detection information in the first output display area and the second output display area of ​​the large language model, respectively, so as to intuitively display the output of the large language model.

[0081] In an exemplary embodiment, the document attachment risk detection method further includes: in response to an information statistics instruction for the attachment detection information in the second output display area, displaying statistical results for the attachment detection information; wherein the statistical results are obtained by the large language model performing information statistics on the attachment detection information in accordance with the information statistics instruction.

[0082] Among them, the information statistics instruction can be an instruction triggered by the user, which is specifically used to perform further statistical analysis and processing on the attachment detection information. The statistical result is the result obtained by the large language model performing information statistics on the attachment detection information according to the information statistics instruction. Exemplarily, the user triggers an interaction with the attachment detection information displayed in the second output display area, and can specifically perform further statistical analysis and processing on the attachment detection information by inputting an information statistics instruction. The server responds to the user's information statistics instruction for the attachment detection information, performs information statistics on the attachment detection information according to the information statistics instruction by calling the large language model, and displays the statistical results output by the large language model. Specifically, the server can render and display the statistical results in the second output display area.

[0083] In this embodiment, the server can respond to the user's information statistics instruction and display the statistical results obtained by performing information statistics on the attachment detection information through the large language model according to the information statistics instruction, thereby supporting the user to perform further statistical analysis on the attachment detection information.

[0084] The present application also provides an application scenario, which applies the above-mentioned document attachment risk detection method. Specifically, the application of the document attachment risk detection method in the application scenario is as follows:

[0085] Currently, risk detection for document attachments includes a purely manual review method, that is, the reviewer reviews the information of the document, such as the reimbursement amount, business type, and department to which it belongs, to determine whether there is any violation information. However, the types of attachments are diverse, and the use of vectorization technology alone cannot extract unstructured data. In addition, for various types of documents, the risk point review is different. For example, the attachment of the expense reimbursement form requires whether the start and end time of the itinerary is within the departure time-arrival time of the document; whether the hotel name and unit on the hotel receipt on the reimbursement work order are consistent with the invoice seller, and whether the amount is consistent, which makes the review cumbersome.

[0086] Therefore, currently there are many risk points in the key elements of attachments, which are easy to be missed, such as the review of the consistency between the contract content and the document payment information, the review of the matching degree between the contract content and the document collection progress, etc., which are easy to be missed in manual review. Moreover, the workload of attachment review is large, the efficiency of manual review is low, and the labor cost is high; the larger the scale of the enterprise, the more types of documents are involved, the more types and numbers of attachments are involved, and the greater the workload of shared financial personnel during the review, and more manpower is needed for review. In addition, business scenarios are diverse, and the audit points are numerous and complex; for example, the types of expense reimbursement attachments for different businesses are different, and it is easy for incomplete attachments to lead to the problem of being unable to prove the relevance of expenditures to the business, resulting in low efficiency in risk detection of document attachments.

[0087] Based on this, the document attachment risk detection method provided in this embodiment realizes attachment information extraction through OCR recognition technology combined with file vectorization technology capabilities, and realizes complex rule checks on unstructured attachments / pictures / texts through the intention recognition and semantic understanding capabilities of the large model; compares the information on the attachment or the document through GPT prompt words to see if it complies with the risk rules, thereby reducing manual review costs; realizes unstructured data extraction; the inspection rules may be complex, and through risk rule configuration and natural language expression, the configuration workload is simplified and the implementation difficulty is reduced, thereby improving the risk detection efficiency of attachments.

[0088] Specifically, Figure 4As shown, in this embodiment, after creating a task, that is, creating a document attachment risk detection task, it is determined whether an attachment review scheme is configured. If not, the task is terminated directly. If the attachment review scheme is configured, it is further determined whether the attachment type is a picture. Specifically, the attachment type of the attachment can be determined. For picture types, such as jpg, jpeg, png, bmp, OFD or single-page pdf format types, information extraction can be performed based on character recognition technology to obtain the original attachment information. For non-picture types, such as multi-page pdf, txt, json library and other types, information extraction can be performed based on vectorization technology to obtain the original attachment information. For the extracted original attachment information, the audit elements are extracted through a large language model. Specifically, the audit elements can be refined based on GPT picture prompts or GPT long text prompts to obtain the key information of the attachment. It is determined whether the key information of the attachment contains risk warning items. If not, the risk review is terminated. If included, a risk warning review is performed based on the document information + attachment key information through a large language model to obtain a risk detection result for the attachment, such as a risk warning summary output by GPT.

[0089] Specifically, for the big model's processing of the summary of attachment information, the attachment is extracted through OCR recognition technology or file vectorization technology; the attachment type is distinguished according to the GPT prompt word requirements; the key information to be extracted for each attachment type is configured according to the attachment type; and finally, the attachment information is summarized through the big model's intent recognition capability. Specifically, it is necessary to configure the audit elements of the attachment review plan. For example, the key elements required for each attachment type of the attachment review plan can be preset according to the customer's attachment type, that is, according to the type of attachment content; for example, if the attachment type is a train ticket, the key elements are the traveler, time, station, seat, total amount, etc. If Figure 5 As shown in FIG. 1 , it is a detailed description of the audit elements configured according to actual needs in a specific application. Specifically, different audit elements are configured for different types of attachment content. Figure 6 As shown, for attachment 1 of the picture type and attachment 2 of the non-picture type, different key information of the attachments that need to be subject to risk review can be extracted respectively.

[0090] Furthermore, it is also necessary to configure the GPT prompt words used in the attachment review solution. GPT prompt words refer to the text entered by the user when interacting with the GPT model, which are used to guide the model to generate corresponding replies. These prompt words can be questions, instructions, descriptions, etc., and they play a key role in guiding the model to understand user needs and generate appropriate answers. It is a neural network model with ultra-large-scale parameters (usually more than one billion). These models are trained using deep learning technology, large amounts of data and computing resources to achieve optimal performance in various tasks. GPT prompt words distinguish between image categories and long text categories (non-image categories). Depending on the type of attachment, different GPT prompt words are used to extract key elements. For example Figure 7 As shown, for different attachment types, after creating the task, determine whether to configure the attachment review plan, and configure two types of GPT prompt words and review elements for image type and non-image type (file type) respectively; perform OCR recognition on image type attachments to obtain the corresponding attachment content, and perform vectorization processing on non-image types to obtain the corresponding attachment content; the obtained attachment content is processed according to the image type GPT prompt words and non-image type GPT prompt words respectively to obtain the corresponding summary content, that is, the key information of the attachment that needs to be subject to risk review is obtained.

[0091] In the previous processing, we obtained the summary of attachment information. In the risk review process, we expanded the review plan, added review risk warning rules, added GPT risk warning words, and used the big model to compare attachments with attachments (whether the hotel name and unit on the hotel receipt are consistent with the invoice seller, and whether the amount is consistent) based on the summary of attachment information + review risk warning rules; compare attachments with document information to see if there are any discrepancies (whether the start and end times of the itinerary are within the departure time - arrival time of the document (such as the reimbursement form)).

[0092] Specifically, for audit risk warning rules, that is, risk detection rules, risk detection rules may be very complex. Through natural language expression, the configuration workload for risk detection rules can be simplified and the implementation difficulty can be reduced; for example, for train ticket rules, "First-class and business-class seats are considered violations for train travel" can be configured in the corresponding risk detection rules, and "The amount of hotel bills (accommodation) cannot be greater than 500" can be configured in the risk detection rules for hotel bills. For the constructed GPT risk warning words, the warning words specify the output of the large model, specify the delineation of model responsibilities, and improve the warning word information before using the model; the information of the documents and attachments is supplemented.

[0093] For example, the format of the risk detection prompt word corresponding to the large language model can be:

[0094] The document information is as follows:

[0095]

[0096] The attachment information is as follows:

[0097]

[0098] The above is the document information and attachment information. If the document information is not provided, it will not be referenced. You are an auditor. The user input is the checkpoint. According to the checkpoint, a concise result basis and inference result are given. For the inference result, if it passes, it is true, otherwise it is false. It is required to be returned in json format. The json format is as follows: {"result": result basis, "pass": inference result}

[0099] No other information needs to be output. "

[0100] In a specific application, such as Figure 8 As shown, for the key information of the attachment and the document information, the GPT risk warning words can be compared one by one according to each rule in the audit risk warning check rule, so that the GPT can output the risk point summary one by one, thereby obtaining the risk audit result for the attachment. In a specific application, such as Fig. 9 As described, the risk review results output by GPT include the risk review results for user Zhang San's train, specifically including the various inspection results of train ticket inspection item 1 and train ticket inspection item 2, and displaying the train ticket information extracted from the picture of train ticket 2.

[0101] Furthermore, for the processing of the summary information output by the image format class, the image class summarizes the OCR image information one by one when summarizing, so as to solve the problem of missing tickets in the overall summary of the model. That is, for the target attachments including the image type, the corresponding risk detection prompt words can be further configured to obtain the image class GPT prompt words. Specifically, the responsibilities can be limited in the image class GPT prompt words, so that the large model can extract the attachment summary information such as the title and amount from the image information to provide a specialized style scheme, which can be specifically displayed and optimized according to the corresponding attachment style information. In a specific application, when the model return information is obtained after the summary is completed, attachmentid: attachment id is appended to the information header, that is, the attachment identifier of the target attachment of the image type is added to the attachment detection information to obtain the attachment summary information, which can be specifically used as text summary information, and finally modified and displayed in the display stage.

[0102] Specifically, the GPT prompt word for images is limited to: let the model recognize / summarize the title of the entire image of the target attachment, and wrap it with "****", and recognize the amount information and wrap it with "$$$$"; the content of the GPT prompt word for images can be:

[0103] “

[0104] The above is the content of the attachment. Please organize it into a text output that is easy to read, structured, and beautiful. Determine the type of attachment and output the following information based on the result of the determination:

[0105] You only need to include the following key information:

[0106] Output the title, wrap it around ****; output the amount, wrap it around $$$$; if there is no matching attachment type, do not output it.

[0107] Other information and user reminders shall not be output. "

[0108] For text content algorithm processing, after obtaining the summary content of the large model, that is, obtaining the attachment detection information of the target attachment of the image type, it is necessary to extract the attachment name and link; the method is as follows: regular match and intercept attachmentid to obtain an array dataArray, which contains the attachment id and the attachment summary content, that is, dataArray includes the attachment summary information of the target attachment, specifically including the attachment detection information and the attachment identifier; construct a Map object splitMap, traverse the array dataArray, and pass / ^\d+$ / .test(element ) determines whether it is an attachment id. If it is an attachment id, take the attachment id as the key of splitMap and store it in an empty array; if the regular match is false, use the regular expression element.split( / (\*\*\*\*.*\*\*\*\*) / ).filter(Boolean) to intercept the attachment title (in the case of multiple bills for one picture), and extract the / (\${3,}.*\${3,}) / content in the process and replace it with the amount style; finally, fill the last empty array of splitMap with content to obtain the attachment style information for the target attachment.

[0109] In a specific application, such as Fig.10 As shown in the figure, the text summary is the attachment summary information, which is obtained by adding the attachment ID (attachment identifier) ​​to the attachment detection information. The attachment information includes the attachment link information and the attachment ID. Based on the text summary, instructions are extracted according to the attachment information, such as ocr.split( / (attachmentid!:: ]?[d]+(?: <vbr>)?) / ).filter(Boolean) instruction, determine the attachment information according to the attachment id, such as the specific "attachmentid:123" to extract the number of attachment information in the text, thereby obtaining an array, and replace the attachmentid in the array to obtain the attachment style information including title and amount; then combine the attachment information + regular expression to replace the attachment style information such as title and amount for display.

[0110] The document attachment risk detection method provided in this embodiment uses OCR+big model to identify and extract the specified content on the attachment, and then uses the big model to compare the information on other attachments or fields on the document. The big model responsibilities are specified through GPT prompt words, configuration audit elements, and risk rule details, and a specific format is output for later processing. In view of the current complex attachment inspection rules, natural language expression is used to simplify the configuration workload and reduce the difficulty of implementation. At the same time, the intention recognition and semantic understanding capabilities of the big model can be used to realize complex rule inspections of unstructured attachments / texts, thereby improving the risk detection efficiency of document attachments.

[0111] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0112] Based on the same inventive concept, the embodiment of the present application also provides a document attachment risk detection device for implementing the document attachment risk detection method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more document attachment risk detection device embodiments provided below can refer to the limitations of the document attachment risk detection method above, and will not be repeated here.

[0113] In an exemplary embodiment, Fig.11 As shown, a document attachment risk detection device 1100 is provided, comprising: an attachment acquisition module 1102, a content type processing module 1104, a detection information extraction module 1106 and a risk detection module 1108, wherein:

[0114] The attachment acquisition module 1102 is used to acquire the target attachment associated with the target document and determine the attachment content type of the target attachment;

[0115] Content type processing module 1104, used to determine information detection elements and risk detection rules that match the attachment content type;

[0116] A detection information extraction module 1106 is used to extract attachment detection information from the target attachment according to the information detection elements through the large language model, where the attachment detection information includes information in the target attachment corresponding to the information detection elements;

[0117] The risk detection module 1108 is used to perform information risk detection based on the attachment detection information, risk detection rules and document information of the target document through a large language model to obtain a risk detection result of the target attachment.

[0118] In some embodiments, the detection information extraction module 1106 is also used to extract the original attachment information from the target attachment in an information extraction method that is compatible with the attachment format type of the target attachment; extract prompt words based on the attachment format type and information detection element configuration information; input the information extraction prompt words and the original attachment information into the large language model to extract detection information, and obtain the attachment detection information output by the large language model.

[0119] In some embodiments, the detection information extraction module 1106 is also used to determine the attachment format type of the target attachment; when the attachment format type includes a picture type, the original attachment information is extracted from the target attachment based on character recognition technology; when the attachment format type includes a non-picture type, the original attachment information is extracted from the target attachment based on file vectorization technology.

[0120] In some embodiments, the risk detection module 1108 is also used to obtain document information of the target document; configure risk detection prompt words based on attachment detection information, risk detection rules and document information; input the risk detection prompt words into the large language model for information risk detection, and obtain the risk detection results for the target attachment output by the large language model.

[0121] In some embodiments, the target attachments include at least two, and the risk detection module 1108 is also used to configure risk detection prompt words based on the attachment detection information and risk detection rules of each target attachment; the risk detection prompt words are input into the large language model for information risk detection, and the risk detection results for each target attachment output by the large language model are obtained.

[0122] In some embodiments, an output display module is also included, which is used to add the attachment identifier of the target attachment to the attachment detection information when the attachment format type of the target attachment includes a picture type, to obtain attachment summary information for the target attachment; extract the attachment identifier from the attachment summary information, and intercept the attachment style information from the target attachment according to the attachment identifier; and display the attachment style information and the attachment detection information in the model output display area of ​​the large language model.

[0123] In some embodiments, an output display module is further included, which is used to display risk detection rules and risk detection results in a first output display area of ​​the large language model; and to display attachment detection information in a second output display area of ​​the large language model.

[0124] In some embodiments, the output display module is also used to display statistical results for the attachment detection information in response to an information statistics instruction for the attachment detection information in the second output display area; wherein the statistical results are obtained by the large language model performing information statistics on the attachment detection information in accordance with the information statistics instruction.

[0125] Each module in the document attachment risk detection device can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.

[0126] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Fig.12 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store various data involved in the document attachment risk detection method. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a document attachment risk detection method is implemented.

[0127] In an exemplary embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as shown in FIG. Fig.13 As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, near field communication (NFC) or other technologies. When the computer program is executed by the processor, a document attachment risk detection method is implemented. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0128] Those skilled in the art will understand that Fig.12 The structure shown in or 13 is merely a block diagram of a partial structure related to the present application scheme, and does not constitute a limitation on the computer device to which the present application scheme is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0129] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0130] In one embodiment, a computer-readable storage medium is provided, storing a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0131] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0132] It should be noted that the 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, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0133] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0134] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0135] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be construed as limiting the scope of the present application. It should be noted that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.< / vbr>

Claims

1. A document attachment risk detection method, characterized in that: The method comprises: Acquire a target attachment associated with a target document, and determine the attachment content type of the target attachment; Determining information detection elements and risk detection rules that match the type of content in the attachment; Extracting attachment detection information from the target attachment according to the information detection element through a large language model, wherein the attachment detection information includes information in the target attachment corresponding to the information detection element; By using the large language model, information risk detection is performed based on the attachment detection information, the risk detection rules and the document information of the target document to obtain a risk detection result of the target attachment.

2. The method according to claim 1, characterized in that The step of extracting attachment detection information from the target attachment according to the information detection element by using the large language model includes: Extracting original attachment information from the target attachment in accordance with an information extraction method adapted to the attachment format type of the target attachment; Extracting prompt words based on the attachment format type and the information detection element configuration information; The information extraction prompt word and the original attachment information are input into a large language model to extract detection information, and the attachment detection information output by the large language model is obtained.

3. The method according to claim 2, characterized in that The extracting original attachment information from the target attachment in accordance with an information extraction method adapted to the attachment format type of the target attachment includes: Determine the attachment format type of the target attachment; When the attachment format type includes a picture type, extracting original attachment information from the target attachment based on character recognition technology; When the attachment format type includes a non-image type, the original attachment information is extracted from the target attachment based on file vectorization technology.

4. The method according to claim 1, characterized in that: The step of performing information risk detection based on the attachment detection information, the risk detection rule, and the document information of the target document through the large language model to obtain a risk detection result of the target attachment includes: Acquiring document information of the target document; Configure risk detection prompt words based on the attachment detection information, the risk detection rules and the document information; The risk detection prompt word is input into the large language model to perform information risk detection, and a risk detection result for the target attachment output by the large language model is obtained.

5. The method according to claim 1, characterized in that The target attachments include at least two, and the method further includes: configuring risk detection prompt words based on the attachment detection information of each of the target attachments and the risk detection rule; The risk detection prompt words are input into the large language model to perform information risk detection, and the risk detection results for each of the target attachments output by the large language model are obtained.

6. The method according to claim 1, characterized in that The method further comprises: When the attachment format type of the target attachment includes a picture type, adding an attachment identifier of the target attachment to the attachment detection information to obtain attachment summary information for the target attachment; Extracting the attachment identifier from the attachment summary information, and intercepting attachment style information from the target attachment according to the attachment identifier; In the model output display area of ​​the large language model, the attachment style information and the attachment detection information are displayed.

7. The method according to any one of claims 1 to 6, characterized in that: The method further comprises: In the first output display area of ​​the large language model, display the risk detection rule and the risk detection result; The attachment detection information is displayed in a second output display area of ​​the large language model.

8. The method according to claim 7, characterized in that The method further comprises: In response to an information statistics instruction for the attachment detection information in the second output display area, displaying a statistical result for the attachment detection information; The statistical result is obtained by the large language model performing information statistics on the attachment detection information according to the information statistics instruction.

9. A document attachment risk detection device, characterized in that: The device comprises: An attachment acquisition module is used to acquire a target attachment associated with a target document and determine the attachment content type of the target attachment; A content type processing module, used to determine information detection elements and risk detection rules that match the attachment content type; A detection information extraction module, configured to extract attachment detection information from the target attachment according to the information detection element through a large language model, wherein the attachment detection information includes information in the target attachment corresponding to the information detection element; The risk detection module is used to perform information risk detection based on the attachment detection information, the risk detection rules and the document information of the target document through the large language model to obtain the risk detection result of the target attachment.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

12. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images