A document auditing method, device, equipment and storage medium

Abstract

The application discloses a kind of document auditing method, device, equipment and storage medium, applied in the field of deep learning, including extracting the subject of the document to be audited;The subject is linked with the preset knowledge graph;If link is successful, the document to be audited is audited according to the attribute of the current node in the preset knowledge graph;If link fails, the document to be audited is a normal document.The application uses entity linking technology and knowledge graph technology to realize the reasoning and detection of document content, making the rule base more maintainable and expandable, reducing the operation and maintenance cost of rules, improving efficiency, and suitable for large-scale data transmission scenarios in cross-border fields.

Description

Technical Field

[0001] This invention relates to the field of deep learning, and in particular to a document review method, apparatus, device, and storage medium. Background Technology

[0002] Existing methods for manually reviewing documents require significant manpower and are inefficient. Developing pre-defined rules based on specific sentence structures and then reviewing documents according to these rules requires manual setting and maintenance, resulting in high labor costs. Model-based intelligent document review methods can only review specific content, such as tables. Therefore, existing document review methods suffer from high costs and limitations. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide a document review method, apparatus, device and storage medium, which solves the problems of high cost and limitations in document review in the prior art.

[0004] To address the aforementioned technical problems, this invention provides a document review method, comprising:

[0005] Extract the subject of the document to be reviewed;

[0006] Link the topics to a pre-defined knowledge graph;

[0007] If the link is successful, the document to be reviewed will be reviewed according to the attributes of the current node in the preset knowledge graph.

[0008] If the link fails, the document to be reviewed is a normal document.

[0009] Optionally, the extraction of the subject of the document to be reviewed includes:

[0010] Obtain the document to be reviewed;

[0011] The layout of the document to be reviewed is parsed using the Layout model to obtain the illustrations, tables, headers, footers and paragraphs in the document to be reviewed;

[0012] The illustrations, tables, and paragraphs are processed to obtain the subject matter.

[0013] Optionally, obtaining the document to be reviewed includes:

[0014] The Zeek tool was used to capture and parse data packets in network transmission to obtain the path of the document to be reviewed;

[0015] The document to be reviewed is determined using the Kafka tool and its path.

[0016] Optionally, the processing of the illustrations, tables, and paragraphs to obtain the topic includes:

[0017] The illustrations are classified using a ResNet-based image classification model to determine their subject matter;

[0018] The content in the table is parsed to obtain the table title, which is then used as the table's subject.

[0019] The topic of the paragraph is obtained by using the LDA topic model to predict the topic of the paragraph.

[0020] Optionally, the construction process of the preset knowledge graph includes:

[0021] The document review request sent by the receiving terminal is standardized into an RDF-based triple format, and the preset knowledge graph is generated.

[0022] Optionally, when the document to be reviewed is abnormal, it includes:

[0023] Input the abnormal information of the abnormal document into the preset report template to generate a detection report;

[0024] The test report is sent to the terminal, prompting the reviewer to confirm it.

[0025] Optionally, the preset knowledge graph further includes:

[0026] The entity description is used to eliminate ambiguity when the topic is linked to the preset knowledge graph.

[0027] The present invention also provides a document review device, comprising:

[0028] The topic extraction module is used to extract the topic of documents to be reviewed;

[0029] The linking module is used to link the topic with a preset knowledge graph;

[0030] The first link result module is used to review the document to be reviewed based on the attributes of the current node in the preset knowledge graph if the link is successful.

[0031] The second link result module is used to determine if the document to be reviewed is a normal document if the link fails.

[0032] The present invention also provides a document review device, comprising:

[0033] Memory, used to store computer programs;

[0034] A processor is used to implement the steps of the document review method described above when executing the computer program.

[0035] The present invention also provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described document review method.

[0036] As can be seen, this invention extracts the topic of the document to be reviewed; links the topic to a preset knowledge graph; if the link is successful, the document to be reviewed is reviewed according to the attributes of the current node in the preset knowledge graph; if the link fails, the document to be reviewed is a normal document. This invention uses entity linking technology and knowledge graph technology to achieve reasoning and detection of document content, making the rule base easier to maintain and expand, reducing the operation and maintenance costs of rules, improving efficiency, and is suitable for large-scale data transmission scenarios in cross-border fields.

[0037] In addition, the present invention also provides a document review device, equipment and storage medium, which also have the above-mentioned beneficial effects. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0039] Figure 1 A flowchart of a document review method provided in an embodiment of the present invention;

[0040] Figure 2 A flowchart illustrating a document review method provided in an embodiment of the present invention;

[0041] Figure 3 This is a schematic diagram of the structure of a document review device provided in an embodiment of the present invention;

[0042] Figure 4 This is a structural schematic diagram of a document review device provided in an embodiment of the present invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] Please refer to Figure 1 , Figure 1 A flowchart illustrating a document review method provided in an embodiment of the present invention. The method may include:

[0045] S101: Extract the subject of the document to be reviewed.

[0046] The execution entity in this embodiment is a processor. The document review method provided in this application can be applied to large-scale data transmission scenarios in cross-border fields.

[0047] Furthermore, in order to accurately review the content of each part of the document, the above-mentioned extraction of the subject of the document to be reviewed may include the following steps:

[0048] Step 21: Obtain the document to be reviewed.

[0049] This embodiment does not limit the method of obtaining the document to be reviewed. For example, the user can upload the document to the cloud and download the document to be reviewed from the cloud; or the user can monitor the traffic data packets in the local area network in real time, and when the traffic data packet is detected to be a document, the document can be captured and parsed to obtain the document to be reviewed.

[0050] Furthermore, to broaden the scope of application, the above-mentioned process of obtaining documents to be reviewed may include the following steps:

[0051] Step 31: Use the Zeek tool to capture and parse the data packets in network transmission to obtain the path of the document to be reviewed;

[0052] Step 32: Use the Kafka tool to determine the document to be reviewed by its path.

[0053] In this embodiment, Zeek is an open-source, passive network traffic analysis software, and Kafka is a high-throughput, persistent, distributed publish-subscribe message queue system. By using Zeek and Kafka to monitor traffic data packets within the local area network in real time and parse the path of the document to be reviewed, the document to be reviewed can be determined. This method is more intelligent than the document upload and download method and is applicable to a wider range of scenarios.

[0054] Step 22: Use the Layout model to parse the layout of the document to be reviewed, and obtain the illustrations, tables, headers, footers and paragraphs in the document to be reviewed.

[0055] This embodiment does not limit the specific layout model (multimodal intelligent document understanding algorithm). For example, it can use the LayoutLM model to parse the layout of the document to be reviewed; or it can use the LayoutLMv2 model to parse the layout of the document to be reviewed; or it can use the LayoutXLM model to parse the layout of the document to be reviewed; or it can use the LayoutLMv3 model to parse the layout of the document to be reviewed. In order to better parse the document to be reviewed and improve efficiency, this embodiment can use the LayoutLMv3 model to parse the layout of the document to be reviewed, and obtain the illustrations, tables, paragraphs, headers and footers of the document to be reviewed. The purpose of obtaining the headers and footers from the layout is to avoid the impact of page spans on the extraction of illustrations, tables and paragraphs, and to improve the accuracy of subsequent processing of illustrations, tables and paragraphs to obtain the theme. In this embodiment, the parsing process may include: converting the document into an image and annotating the image; using an OCR (optical character recognition) service to recognize the text and coordinates in the image; wherein the OCR service uses the PSENet model (a text detection method based on semantic segmentation) for text detection and the CRNN model (Convolutional Recurrent Neural Network) for text recognition. The CRNN model is used to solve image-based sequence recognition problems, especially scene text recognition problems.

[0056] Step 23: Process the illustrations, tables, and paragraphs to obtain the theme.

[0057] This embodiment does not limit the method for extracting topics from illustrations, tables, and paragraphs. For example, the VGG image classification model can be used to classify illustrations; or the ResNet image classification model can be used to classify illustrations; or the LDA topic model (Latent Dirichlet Allocation) can be used to predict the topic of a paragraph.

[0058] Furthermore, to improve processing speed, the above-mentioned processing of illustrations, tables, and paragraphs to obtain the main topic may include the following steps:

[0059] Step 41: Classify the illustrations using a ResNet-based image classification model to determine the subject of the illustrations;

[0060] Step 42: Parse the content in the table to obtain the table title, which will then serve as the table's subject.

[0061] Step 43: Use the LDA topic model to predict the topic of the paragraph.

[0062] This embodiment employs different methods for different types of content to extract the themes from paragraphs, illustrations, and tables, enabling the identification and detection of various content types within the document. For example, manufacturing companies with cross-border operations are particularly sensitive to product parameter structure diagrams. If a product parameter structure diagram is detected in the current document and linked to a node in the diagram, the document's compliance is determined based on the node's attributes. For instance, in documents from manufacturing companies with cross-border operations, if the LDA topic model detects that a paragraph describes product structure and parameters, the reviewers should be alerted to pay close attention to that document.

[0063] S102: Link the topic to the preset knowledge graph.

[0064] This embodiment links topics to a pre-defined knowledge graph and determines whether a document contains anomalies based on the linking results. This embodiment does not limit the construction process of the pre-defined knowledge graph. For example, document review requirements can be standardized into an RDF-based triple format, and a pre-defined knowledge graph can be generated.

[0065] Furthermore, to improve detection efficiency, the construction process of the aforementioned pre-defined knowledge graph may include the following steps:

[0066] The system receives document review requests from the terminal, standardizes these requests into an RDF-based triple format, and generates a pre-defined knowledge graph.

[0067] In this embodiment, reviewers can configure HTTP requests for document review requirements via a terminal. Upon receiving the request, the server sends a page configuration response to the terminal and renders the configuration page. Reviewers input their review requirements on the configuration page and send them to the server. The server receives the review requirements and standardizes them into a RDF (Resource Description Framework, a markup language for describing web resources) triple format, generating a preset knowledge graph. For example, (Company A, Prohibited, Flowchart) describes that Company A prohibits the appearance of flowcharts in documents.

[0068] Furthermore, in order to eliminate the ambiguity of different characters representing the same thing, the aforementioned pre-defined knowledge graph may also include the following steps:

[0069] The entity description is used to eliminate ambiguity when different texts represent the same thing when a topic is linked to a pre-defined knowledge graph.

[0070] This embodiment considers that OCR recognition may have certain errors, and different companies may use different names for the same thing (e.g., Apple computers may be called "Mac" or "Apple laptop"). To address these issues, this invention can perform link disambiguation between the topic of the parsed document to be reviewed and the preset knowledge graph. This embodiment does not limit the specific model used for link disambiguation. For example, it can be a knowledge graph-based representation learning algorithm—the Trans series: it could use the TransH algorithm model; or it could use the TransR algorithm model; or it could use the TransE algorithm model. To better achieve the disambiguation problem, this embodiment can use the TransH algorithm model to perform link disambiguation between the topic of the parsed document to be reviewed and the preset knowledge graph.

[0071] S103: If the link is successful, the document to be reviewed will be reviewed according to the attributes of the current node in the preset knowledge graph.

[0072] In this embodiment, a successful link indicates that the document to be reviewed contains content that needs to be reviewed. Therefore, the document to be reviewed is reviewed according to the attributes of the current node in the preset knowledge graph to determine whether the document is compliant.

[0073] Furthermore, to facilitate auditors in reviewing abnormal documents, the following steps may be included when a document to be reviewed is abnormal:

[0074] Step 61: Input the abnormal information of the abnormal document into the preset report template to generate a detection report;

[0075] Step 62: Send the test report to the terminal and prompt the reviewer for confirmation.

[0076] S104: If the link fails, the document to be reviewed is a normal document.

[0077] The link failure in this example indicates that the document to be reviewed does not contain any content that needs to be reviewed.

[0078] The document review method provided in this invention extracts the topic of the document to be reviewed; links the topic to a preset knowledge graph; if the link is successful, the document to be reviewed is reviewed according to the attributes of the current node in the preset knowledge graph; if the link fails, the document to be reviewed is considered normal. This invention uses entity linking technology and knowledge graph technology to achieve reasoning and detection of document content, making the rule base easier to maintain and expand, reducing the operational costs of rules, improving efficiency, and applicable to large-scale data transmission scenarios in cross-border fields. Furthermore, abnormal documents generate detection reports, facilitating further review by reviewers; the preset knowledge graph includes descriptions of entities, enabling disambiguation by linking topics to the preset knowledge graph; converting document review requirements into triples to form the preset knowledge graph improves detection efficiency; different methods are used for different types of topics, improving processing efficiency; real-time monitoring of traffic data packets within the local area network is more intelligent than document upload and download methods, and is applicable to a wider range of scenarios; and parsing the layout of the document to be reviewed allows for accurate review of each part of the document's content.

[0079] For a clearer understanding of this invention, please refer to the following details. Figure 2 , Figure 2 A flowchart illustrating a document review process provided in this embodiment of the invention may specifically include:

[0080] Network traffic is monitored, traffic packets are captured and parsed to obtain the content of the document to be reviewed, i.e., document data. The layout of the document to be reviewed is analyzed, mainly divided into illustration data, paragraph data, and table data. Illustration data is classified and illustration themes are determined; text data is analyzed to obtain paragraph themes; and table data is parsed to obtain table themes. Each theme is linked with a preset knowledge graph, which includes entity descriptions, to achieve the purpose of link disambiguation. Based on the link results, it is determined whether the content of the document to be reviewed contains any illegal content, thereby determining whether the document to be reviewed is abnormal.

[0081] The document review device provided in the embodiments of the present invention will be described below. The document review device described below and the document review method described above can be referred to in correspondence.

[0082] Please refer to the details. Figure 3 , Figure 3 A schematic diagram of a document review device provided in an embodiment of the present invention may include:

[0083] The topic extraction module 100 is used to extract the topic of documents to be reviewed.

[0084] Linking module 200 is used to link the topic with a preset knowledge graph;

[0085] The first link result module 300 is used to review the document to be reviewed based on the attributes of the current node in the preset knowledge graph if the link is successful.

[0086] The second link result module 400 is used to determine if the document to be reviewed is a normal document if the link fails.

[0087] Based on the above embodiments, the topic extraction module 100 may include:

[0088] The acquisition unit is used to acquire the document to be reviewed;

[0089] The parsing unit is used to parse the layout of the document to be reviewed using the layout model, and to obtain the illustrations, tables, headers, footers and paragraphs in the document to be reviewed;

[0090] The processing unit is used to process the illustrations, tables, and paragraphs to obtain the subject matter.

[0091] Based on the above embodiments, the acquisition unit may include:

[0092] The first subunit is used to capture and parse data packets in network transmission using the Zeek tool to obtain the path of the document to be reviewed.

[0093] The second subunit is used to determine the document to be reviewed by using the Kafka tool and the path of the document to be reviewed.

[0094] Based on the above embodiments, the processing unit may include:

[0095] The first processing subunit is used to classify the illustration using a ResNet-based image classification model to determine the subject of the illustration;

[0096] The second processing subunit is used to parse the content in the table to obtain the title of the table, which is then used as the subject of the table.

[0097] The third processing subunit is used to predict the topic of the paragraph using the LDA topic model to obtain the topic of the paragraph.

[0098] Based on the above embodiments, the construction process of the preset knowledge graph in the link module 200 may include:

[0099] A preset knowledge graph construction unit is used to receive document review requirements sent by the terminal, standardize the document review requirements into an RDF-based triple format, and generate the preset knowledge graph.

[0100] Based on the above embodiments, the first link result module 300 may include the following when the document to be reviewed is abnormal:

[0101] The input unit is used to input the abnormal information of the abnormal document into the preset report template to form a detection report;

[0102] The sending unit is used to send the test report to the terminal and prompt the reviewer for confirmation.

[0103] Based on any of the above embodiments, the preset knowledge graph may further include descriptions of entities to eliminate ambiguity when the topic is linked to the preset knowledge graph.

[0104] It should be noted that the order of the modules and units in the above-mentioned document review device can be changed without affecting the logic.

[0105] The document review device provided in this embodiment of the invention comprises a topic extraction module 100 for extracting the topic of the document to be reviewed; a linking module 200 for linking the topic with a preset knowledge graph; a first linking result module 300 for reviewing the document according to the attributes of the current node in the preset knowledge graph if the link is successful; and a second linking result module 400 for reviewing the document as a normal document if the link fails. This invention uses entity linking technology and knowledge graph technology to achieve reasoning and detection of document content, making the rule base easier to maintain and expand, reducing the operational costs of rules, improving efficiency, and is suitable for large-scale data transmission scenarios in cross-border fields. Furthermore, the system generates detection reports for documents with anomalies, facilitating further review by auditors. The pre-defined knowledge graph includes entity descriptions, enabling disambiguation by linking topics to the pre-defined knowledge graph. Converting document review requirements into triples to form the pre-defined knowledge graph improves detection efficiency. Different methods are used for different types of topics, enhancing processing efficiency. Real-time monitoring of traffic data packets within the local area network is more intelligent and applicable to a wider range of scenarios compared to document uploads and downloads. Finally, the system analyzes the layout of documents to be reviewed, enabling accurate review of each part of the document's content.

[0106] The document review device provided in the embodiments of the present invention will be described below. The document review device described below and the document review method described above can be referred to in correspondence.

[0107] Please refer to Figure 4 , Figure 4 A schematic diagram of the document review device provided in this embodiment of the invention may include:

[0108] Memory 10 is used to store computer programs;

[0109] Processor 20 is used to execute computer programs to implement the document review method described above.

[0110] The system includes a memory 10, a processor 20, a communication interface 31, and a communication bus 32. The memory 10, processor 20, and communication interface 31 communicate with each other through the communication bus 32.

[0111] In this embodiment of the invention, the memory 10 is used to store one or more programs. The programs may include program code, which includes computer operation instructions. In this embodiment, the memory 10 may store programs for implementing the following functions:

[0112] Extract the subject of the document to be reviewed;

[0113] Link the topic to a pre-defined knowledge graph;

[0114] If the link is successful, the document to be reviewed will be reviewed according to the attributes of the current node in the preset knowledge graph.

[0115] If the link fails, the document awaiting review is a normal document.

[0116] In one possible implementation, the memory 10 may include a program storage area and a data storage area, wherein the program storage area may store the operating system and applications required for at least one function; and the data storage area may store data created during use.

[0117] Furthermore, memory 10 may include read-only memory and random access memory, providing instructions and data to the processor. A portion of the memory may also include NVRAM. The memory stores operating systems and operating instructions, executable modules, or data structures, or subsets thereof, or extended sets thereof, wherein the operating instructions may include various operating instructions for implementing various operations. The operating system may include various system programs for implementing various basic tasks and handling hardware-based tasks.

[0118] Processor 20 can be a central processing unit (CPU), an application-specific integrated circuit, a digital signal processor, a field-programmable gate array, or other programmable logic device. Processor 20 can be a microprocessor or any conventional processor. Processor 20 can call programs stored in memory 10.

[0119] The communication interface 31 can be the interface of the communication module, used to connect with other devices or systems.

[0120] Of course, it should be noted that, Figure 4 The structure shown does not constitute a limitation on the document review device in the embodiments of this application. In practical applications, the document review device may include devices that are more advanced than those described above. Figure 4 More or fewer components as shown, or combinations of certain components.

[0121] The storage medium provided in the embodiments of the present invention is described below. The storage medium described below can be referred to in correspondence with the document review method described above.

[0122] The present invention also provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the document review method described above.

[0123] The storage medium can include various media that can store program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0124] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0125] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0126] Finally, it should be noted that in this document, relationships such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations 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.

[0127] The above provides a detailed description of the document review method, apparatus, device, and storage medium provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A document review method, characterized in that, include: Extract the subject of the document to be reviewed; Link the topics to a pre-defined knowledge graph; If the link is successful, the document to be reviewed will be reviewed according to the attributes of the current node in the preset knowledge graph. If the link fails, the document to be reviewed is a normal document; The extraction of the subject of the document to be reviewed includes: Obtain the document to be reviewed; The layout of the document to be reviewed is parsed using the Layout model to obtain the illustrations, tables, headers, footers and paragraphs in the document to be reviewed; The illustrations, tables, and paragraphs are processed to obtain the subject matter; The process of processing the illustrations, tables, and paragraphs to obtain the topic includes: The illustrations are classified using a ResNet-based image classification model to determine their subject matter; The content in the table is parsed to obtain the table title, which is then used as the table's subject. The topic of the paragraph is obtained by using the LDA topic model for prediction. The construction process of the preset knowledge graph includes: The document review request sent by the receiving terminal is standardized into an RDF-based triple format, and the preset knowledge graph is generated. The preset knowledge graph also includes: The entity description is used to eliminate ambiguity when the topic is linked with the preset knowledge graph; the TransH algorithm model is used to link and disambiguate the topic of the parsed document to be reviewed with the preset knowledge graph.

2. The document review method according to claim 1, characterized in that, The process of obtaining the document to be reviewed includes: The Zeek tool was used to capture and parse data packets in network transmission to obtain the path of the document to be reviewed; The document to be reviewed is determined using the Kafka tool and its path.

3. The document review method according to claim 1, characterized in that, When the document to be reviewed is abnormal, it includes: Input the abnormal information of the abnormal document into the preset report template to generate a detection report; The test report is sent to the terminal, prompting the reviewer to confirm it.

4. A document review device, characterized in that, include: The topic extraction module is used to extract the topic of documents to be reviewed; The linking module is used to link the topic with a preset knowledge graph; The first link result module is used to review the document to be reviewed based on the attributes of the current node in the preset knowledge graph if the link is successful. The second link result module is used to determine if the document to be reviewed is a normal document if the link fails. The topic extraction module includes: The acquisition unit is used to acquire the document to be reviewed; The parsing unit is used to parse the layout of the document to be reviewed using the layout model, and to obtain the illustrations, tables, headers, footers and paragraphs in the document to be reviewed; A processing unit is used to process the illustrations, tables, and paragraphs to obtain the theme; The processing unit includes: The first processing subunit is used to classify the illustration using a ResNet-based image classification model to determine the subject of the illustration; The second processing subunit is used to parse the content in the table to obtain the title of the table, which is then used as the subject of the table. The third processing subunit is used to predict the topic of the paragraph using the LDA topic model to obtain the topic of the paragraph. The construction process of the pre-defined knowledge graph in the linking module includes: A preset knowledge graph construction unit is used to receive document review requirements sent by the terminal, standardize the document review requirements into an RDF-based triple format, and generate the preset knowledge graph. The preset knowledge graph also includes entity descriptions to eliminate ambiguity when the topic is linked to the preset knowledge graph; the TransH algorithm model is used to link and disambiguate the topic of the parsed document to be reviewed with the preset knowledge graph.

5. A document review device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the document review method as described in any one of claims 1 to 3 when executing the computer program.

6. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the steps of the document review method as described in any one of claims 1 to 3.

Images