Legal compliance document auditing method and system

Abstract

The invention relates to the technical field of new energy project management, discloses a legal compliance file auditing method and system, and aims to convert a legal compliance file into a readable structured file through an optical character recognition method and solve the problem of high cost of traditional manual data annotation. According to the method, the file is preliminarily audited by using the preset multi-dimensional semantic recognition rule, so that the consistency of audit standards of different legal compliance files and different stages is ensured, and errors caused by artificial experience differences are avoided. Meanwhile, through multi-dimensional auditing, the accuracy of auditing is improved; and the preset intention classification model is used for carrying out deep semantic analysis on the file passing the preliminary auditing and carrying out secondary auditing, so that the auditing accuracy is improved. Therefore, by implementing the method and the device, a plurality of legal compliance files can be audited at one time, the manual audit workload of legal officers is reduced, the labor cost is reduced, the audit efficiency and accuracy are improved, and the compliance of legal compliance file management can be enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of new energy project management, and particularly relates to a legal compliance file auditing method and system. BACKGROUND

[0002] In the development and construction process of a new energy project, legal compliance file auditing at each stage is a time-consuming and complicated work. Under the traditional file auditing mode, a legal personnel needs to open and upload files one by one for manual auditing. Taking a single project as an example, the number of legal compliance files to be handled is between 70 and 100, and if the average auditing time for each file is 1 minute, the auditing work of a single project will consume more than 70 minutes of the legal personnel.

[0003] With the continuous expansion of the business scale of an enterprise and the continuous increase of the number of projects, the number of legal compliance files for manual auditing is increasing year by year. Under this condition, the disadvantages of manual auditing are increasingly prominent, not only including time-consuming, low efficiency, auditing lag and other problems, but also being prone to errors, thus not only making the enterprise face a severe challenge in labor, but also having an adverse effect on the management of legal compliance procedures. SUMMARY

[0004] Therefore, the present application provides a legal compliance file auditing method and system to solve the problems of time-consuming, low efficiency, auditing lag, being prone to errors and other problems of the traditional file auditing mode under the condition of the increasing number of legal compliance files year by year, thus not only making the enterprise face a severe challenge in labor, but also having an adverse effect on the management of legal compliance files.

[0005] In a first aspect, the present application provides a legal compliance file auditing method for an intelligent auditing system, which comprises the following steps:

[0006] obtaining a plurality of to-be-audited legal compliance files of a target new energy project; based on the plurality of to-be-audited legal compliance files, obtaining a plurality of initial legal compliance files through an optical character recognition method; auditing the plurality of initial legal compliance files by using a preset multi-dimensional semantic recognition rule to obtain a plurality of target legal compliance files that pass the auditing; and auditing the plurality of target legal compliance files by using a preset intent classification model to obtain a plurality of file auditing results of the plurality of target legal compliance files.

[0007] The legal compliance file auditing method provided by the application converts legal compliance files into readable structured files through an optical character recognition method, solving the high cost problem of traditional manual data labeling. Further, the preliminary audit of the files is performed using preset multi-dimensional semantic recognition rules, ensuring the consistency of the audit standards of different project files and different stages, and avoiding errors caused by differences in manual experience. At the same time, through multi-dimensional auditing, the accuracy of the audit is improved. Further, the files that pass the preliminary audit are subjected to deep semantic analysis and secondary audit using a preset intent classification model, improving the accuracy of the audit. Therefore, by implementing the application, multiple legal compliance files can be audited at one time, replacing the manual processing mode, reducing the workload of the legal personnel, reducing the labor cost, improving the efficiency and accuracy of the audit, and thus helping to strengthen the compliance of the project legal file compliance management and the risk control ability of the enterprise.

[0008] In an optional implementation, based on a plurality of legal compliance files to be audited, a plurality of initial legal compliance files are obtained through an optical character recognition method, comprising:

[0009] The plurality of legal compliance files to be audited are subjected to image conversion processing to obtain a plurality of target file pictures; the plurality of target file pictures are processed using a text detection algorithm based on deep learning to determine a plurality of text region pictures; the plurality of text region pictures are subjected to character recognition using a convolutional neural network model to obtain a plurality of first legal compliance files; and the plurality of first legal compliance files are subjected to structured processing to obtain the plurality of initial legal compliance files.

[0010] The legal compliance file auditing method provided by the application solves the complexity problem of traditional manual processing of multiple format files by uniformly converting the unstructured plurality of legal compliance files from WORD, PDF, and picture format to document format, improving the compatibility of the audit system. Further, the text region pictures in the pictures are automatically detected using a text detection algorithm based on deep learning, excluding non-text region pictures in the pictures, improving the accuracy of text positioning. Further, character recognition is performed using a convolutional neural network model, reducing the need for manual word-by-word proofreading, improving the recognition efficiency and accuracy. Finally, the recognized files are converted into structured text, avoiding the semantic logic break caused by pure text. Therefore, by implementing the application, through accurate recognition and structured processing, the audit risks caused by file format or character recognition can be filtered in advance, improving the accuracy of the audit.

[0011] In an optional implementation, the plurality of initial legal compliance files are audited using preset multi-dimensional semantic recognition rules to obtain a plurality of target legal compliance files that pass the audit, comprising:

[0012] According to the preset multi-dimensional semantic recognition rule, the first dimension verification rule, the second dimension verification rule and the third dimension verification rule are determined; the first dimension verification rule is used to audit the project name of the plurality of initial legal compliance files, and a plurality of second legal compliance files that pass the audit are obtained; the second dimension verification rule and the preset element data dictionary are used to audit the key element data of the plurality of second legal compliance files, and a plurality of third legal compliance files that pass the audit are obtained; and the third dimension verification rule is used to perform semantic auditing on the plurality of third legal compliance files, and a plurality of target legal compliance files that pass the audit are obtained.

[0013] The legal compliance file auditing method provided by the application can quickly locate the project to which the file belongs by using the first dimension verification rule to audit the file project name, avoids cross-project misjudgment, and improves the pertinence of the audit. Further, the second dimension verification rule is used in combination with the preset element data dictionary to accurately verify the core element data. Further, the third dimension verification rule is used to deeply mine semantic connotations and perform auditing, thereby avoiding surface auditing risks. Therefore, by implementing the application, the consistency of the auditing standards of different projects and different stages is ensured, and errors caused by differences in manual experience are avoided. Further, the accuracy of the audit is improved by multi-dimensional auditing.

[0014] In an optional implementation, the first dimension verification rule is used to audit the project name of the plurality of initial legal compliance files, and a plurality of second legal compliance files that pass the audit are obtained, including:

[0015] It is judged whether the plurality of initial legal compliance files has a project name, and a plurality of fourth legal compliance files with the project name and a plurality of fifth legal compliance files without the project name are obtained; the first dimension verification rule is used to match and verify the project name of the plurality of fourth legal compliance files, and a plurality of sixth legal compliance files that pass the matching and verification are obtained; and the plurality of second legal compliance files are determined according to the plurality of fifth legal compliance files and the plurality of sixth legal compliance files.

[0016] The legal compliance file auditing method provided by the application separately processes the files without the project name and directly performs key element verification in the subsequent process, thereby avoiding auditing blockage caused by rigid auditing processes and improving the adaptability of the intelligent auditing system to complex documents. At the same time, by distinguishing between the files with and without the project name, it is ensured that all types of files can be effectively audited, and auditing loopholes are reduced.

[0017] In an optional implementation, the third dimension verification rule is used to audit the plurality of third legal compliance files, and a plurality of target legal compliance files that pass the audit are obtained, including:

[0018] determine whether the plurality of third legal compliance files need to be semantically checked, and obtain a plurality of seventh legal compliance files that need to be semantically checked and a plurality of eighth legal compliance files that do not need to be semantically checked; perform semantic checking on the plurality of seventh legal compliance files by using a third dimension checking rule to obtain a plurality of ninth legal compliance files that pass the checking; and determine the plurality of target legal compliance files according to the plurality of eighth legal compliance files and the plurality of ninth legal compliance files.

[0019] The legal compliance file auditing method provided by the application directly passes the files that do not need semantic judgment, thereby reducing the auditing time. Further, the files that need semantic judgment are subjected to deep semantic checking, thereby avoiding excessive auditing and balancing the efficiency and accuracy of the auditing.

[0020] In an optional implementation, the method further includes:

[0021] obtaining a plurality of historical legal compliance files to be audited; obtaining a plurality of historical target legal compliance files by processing the plurality of historical legal compliance files to be audited through an optical character recognition method; performing vectorization processing on the plurality of historical target legal compliance files to obtain a plurality of historical text vectors; performing intent recognition on the plurality of historical legal compliance files to be audited based on the plurality of historical text vectors by using a preset neural network model to obtain a plurality of expression intents, wherein the expression intents are used to reflect whether the historical legal files to be audited are qualified; and constructing a preset intent classification model by taking the plurality of historical text vectors as input and taking the plurality of expression intents as output.

[0022] The legal compliance file auditing method provided by the application converts the historical legal compliance files into readable structured files through the optical character recognition method, thereby solving the high cost problem of traditional manual data labeling. Further, by converting the files into text vectors, the semantic association between words can be captured, and then the auditing system can understand the abstract semantics. At the same time, through the vectorization processing, the computational complexity is reduced, while the semantic information is preserved, and the model training efficiency is improved. Further, the intent recognition is performed by using the preset neural network model, thereby replacing the manual labeling of labels and reducing the labeling cost. Finally, the preset intent classification model is constructed by taking the plurality of historical text vectors as input and taking the plurality of expression intents as output, thereby providing support for subsequent secondary auditing.

[0023] In an optional implementation, the plurality of target legal compliance files are audited by using the preset intent classification model to obtain a plurality of file auditing results of the plurality of target legal compliance files, including:

[0024] performing vectorization processing on the plurality of target legal compliance files to obtain a plurality of target text vectors; and inputting the plurality of target text vectors into the preset intent classification model to obtain the plurality of file auditing results of the plurality of target legal compliance files.

[0025] The legal compliance file auditing method provided by the application can capture semantic association between words by converting files into text vectors, so that the auditing system can understand abstract semantics. Further, the text vectors are input into a preset intention classification model trained, so that the file can be automatically audited again, and the efficiency and accuracy of the auditing are improved.

[0026] In an optional embodiment, the method further comprises: when the plurality of encrypted legal compliance files of the target new energy project are acquired, performing decryption processing on the plurality of encrypted legal compliance files to obtain a plurality of to-be-audited legal compliance files.

[0027] The legal compliance file auditing method provided by the application performs decryption processing on encrypted files, ensures the security of confidential files during transmission and auditing, and meets the file confidentiality auditing requirements.

[0028] In a first aspect, the application provides a legal compliance file auditing method, which is used in a file information system and connected with a user terminal and an intelligent auditing system respectively; the method comprises:

[0029] receiving a plurality of to-be-audited legal compliance files sent by the user terminal; judging whether there is a confidential file in the plurality of to-be-audited legal compliance files; when there is no confidential file, sending the plurality of to-be-audited legal compliance files to the intelligent auditing system, wherein the intelligent auditing system executes the legal compliance file auditing method of the first aspect or any of the corresponding embodiments thereof; and receiving a plurality of file auditing results sent by the intelligent auditing system.

[0030] The legal compliance file auditing method provided by the application can improve the auditing efficiency and accuracy by sending non-confidential legal compliance files to the intelligent auditing system for batch auditing.

[0031] In an optional embodiment, the file information system is further connected with a management end; the method further comprises:

[0032] sending the plurality of file auditing results to the user terminal; when receiving auditing result complaint information sent by the user terminal, sending the auditing result complaint information to the management end, so that the management end rechecks the file auditing result corresponding to the auditing result complaint information; when receiving a complaint information auditing pass instruction sent by the management end, updating the auditing result of the to-be-audited legal compliance file corresponding to the auditing result complaint information, and sending the updated auditing result to the user terminal; and when receiving a complaint information auditing fail instruction sent by the management end, sending the complaint information auditing result to the user terminal.

[0033] The legal compliance file auditing method provided by the application can enable the user terminal to appeal to the auditing result, and further, the management end can review the appeal information and update the auditing result when the review is passed, thereby improving the credibility of the auditing.

[0034] In a third aspect, the application provides a legal compliance file auditing system, which comprises a file information system, an intelligent auditing system and a rule management system.

[0035] The rule management system is configured to obtain preset multi-dimensional semantic recognition rules and send the preset multi-dimensional semantic recognition rules to the intelligent auditing system.

[0036] The legal compliance file auditing system provided by the application can audit multiple legal compliance files at a time by executing the corresponding legal compliance file auditing method in the intelligent auditing system and the file information system, thereby replacing the manual processing mode, reducing the artificial auditing workload of legal personnel, reducing the labor cost, improving the auditing efficiency and accuracy, and further helping to strengthen the compliance of project procedure management and the risk control ability of enterprises. BRIEF DESCRIPTION OF DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the specific embodiments or the prior art, the following will briefly introduce the drawings needed to be used in the specific embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0038] Figure 1 FIG. 1 is a structural block diagram of a legal compliance file auditing system according to an embodiment of the application;

[0039] Figure 2 FIG. 2 is a flowchart of a legal compliance file auditing method according to an embodiment of the application;

[0040] Figure 3 FIG. 3 is a flowchart of another legal compliance file auditing method according to an embodiment of the application;

[0041] Figure 4 FIG. 4 is a flowchart of still another legal compliance file auditing method according to an embodiment of the application;

[0042] Figure 5 FIG. 5 is a hardware structure schematic diagram of a computer device according to an embodiment of the application. DETAILED DESCRIPTION

[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, 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] Unlike large-scale infrastructure and hydropower projects, the development and construction of new energy projects is a complex and systematic project. Vertically, it involves multiple stages such as project planning, feasibility studies, investment decisions, engineering construction, and operation and commissioning. Horizontally, it covers multiple sectors such as law, economics, policy, engineering, management, and technology. Without systematic thinking and scientific methods, it is difficult to effectively coordinate the work.

[0045] Legal compliance documents serve as the logical entry point for management throughout the entire project development and construction process. They are intricately linked to all the aforementioned stages and are also an important subject of project compliance and risk management. Therefore, legal compliance documents are the best analytical object for clarifying the compliance management and risk prevention of new energy projects.

[0046] Legal compliance document review involves multiple areas, spans a long period of time, and involves various and ever-changing procedures. Tens of thousands of legal compliance documents need to be reviewed annually, posing a significant challenge to compliance management.

[0047] According to an embodiment of the present invention, a method for reviewing legal compliance documents is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0048] This embodiment provides a method for reviewing legal compliance documents, which can be used for, for example Figure 1 The intelligent auditing system 12 shown is... Figure 2 This is a flowchart of a legal compliance document review method according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps:

[0049] Step S201: Obtain multiple legal and compliance documents pending review for the target new energy project.

[0050] Among them, the legal compliance documents to be reviewed refer to various legal documents that need to be reviewed for compliance, completeness and validity during the development, construction and operation of new energy projects. These documents may include the project name, the name of the legal review procedure node, the name of the review procedure document, and the supplementary documents for the review procedure.

[0051] In some alternative implementations, if multiple encrypted legal compliance documents of the target new energy project are obtained, the corresponding private key can be used to decrypt the symmetric key, and then the encrypted legal compliance documents can be decrypted using the symmetric key to obtain multiple decrypted legal compliance documents to be reviewed.

[0052] Step S202: Based on multiple legal compliance documents to be reviewed, multiple initial legal compliance documents are obtained through optical character recognition (OCR) methods.

[0053] Optical Character Recognition (OCR) is a method that uses optical devices (such as scanners and cameras) to capture character images from paper documents or electronic pictures, and then uses image processing and machine learning techniques to convert them into editable and searchable structured text data.

[0054] Specifically, optical character recognition (OCR) can convert multiple legal compliance documents awaiting review into editable and readable structured documents, solving the high cost problem of traditional manual data annotation.

[0055] Step S203: Using preset multi-dimensional semantic recognition rules, multiple initial legal compliance documents are reviewed to obtain multiple target legal compliance documents that have passed the review.

[0056] Among them, the preset multi-dimensional semantic recognition rules are a multi-level, multi-dimensional semantic verification rule system pre-generated based on intelligent audit model technology and clear business processes.

[0057] Specifically, pre-defined multi-dimensional semantic recognition rules can be used to conduct preliminary reviews of initial legal compliance documents from different dimensions, ensuring consistency of review standards for different project documents and different stages, and avoiding errors caused by differences in human experience.

[0058] Step S204: Use a preset intent classification model to review multiple target legal compliance documents to obtain multiple document review results for multiple target legal compliance documents.

[0059] Among them, the pre-defined intent classification model refers to a semantic understanding model pre-trained using deep learning technology, which is used to automatically determine whether the expressive intent of legal compliance documents meets the review requirements.

[0060] The legal compliance document review method provided in this embodiment converts legal compliance documents into readable structured documents using optical character recognition (OCR), solving the high cost problem of traditional manual data annotation. Furthermore, it utilizes preset multi-dimensional semantic recognition rules for preliminary document review, ensuring consistency in review standards across different projects and stages, avoiding errors caused by differences in human experience. Simultaneously, multi-dimensional review improves the accuracy of the review. Furthermore, it employs a preset intent classification model to perform deep semantic analysis and secondary review on documents that pass the preliminary review, further enhancing the accuracy rate. Therefore, by implementing this invention, multiple legal compliance documents can be reviewed at once, replacing the manual processing model, reducing the workload of legal personnel, lowering labor costs, and improving review efficiency and accuracy. This, in turn, helps strengthen the compliance of project procedures and the enterprise's risk management capabilities.

[0061] This embodiment provides a method for reviewing legal compliance documents, which can be used for, for example Figure 1 The intelligent auditing system 12 shown is... Figure 3 This is a flowchart of a legal compliance document review method according to an embodiment of the present invention, such as... Figure 3 As shown, the process includes the following steps:

[0062] Step S301: Obtain multiple pending legal and compliance documents for the target new energy project. For details, please refer to [link / reference needed]. Figure 2 Step S201 of the illustrated embodiment will not be described again here.

[0063] Step S302: Based on multiple legal compliance documents to be reviewed, multiple initial legal compliance documents are obtained through optical character recognition (OCR) methods.

[0064] Specifically, step S302 includes:

[0065] Step S3021: Perform image conversion processing on multiple legal compliance documents to be reviewed to obtain multiple target document images.

[0066] Specifically, the format of the legal compliance documents to be reviewed can be determined, including formats such as PDF, WORD, JPG / JPEG / PNG, and ZIP. If it is in ZIP format, an uncompression operation is performed first to extract the internal legal compliance documents.

[0067] Furthermore, the content and form of different legal compliance documents awaiting review vary.

[0068] Furthermore, if the legal compliance documents to be reviewed are in PDF or WORD format, image conversion tools (such as PDF to image library or WORD to image functions) can be used to convert the document content into image format page by page.

[0069] Furthermore, if the legal compliance document to be reviewed is in JPG / JPEG / PNG image format, it can be directly used as the final target file image.

[0070] Furthermore, the resulting multi-target document images can showcase the different contents of various legal compliance documents awaiting review.

[0071] Step S3022: Use a deep learning-based text detection algorithm to process multiple target file images and identify multiple text region images.

[0072] Among them, deep learning-based text detection algorithms refer to algorithms that use deep learning models (such as convolutional neural networks) to automatically find the region where text is located in an image and determine the position and range of the text.

[0073] Specifically, a deep learning-based text detection algorithm (such as the EAST algorithm) is obtained and the corresponding algorithm model is generated using the algorithm.

[0074] Furthermore, multiple converted target image files can be input into the algorithm model. Then, the algorithm model can use structures such as convolutional neural networks (CNNs) to perform multi-scale feature extraction on the images, capturing features such as the edges, textures, and contours of the text in the images, and distinguishing the text from the background.

[0075] Furthermore, based on the extracted features, the algorithm model can predict the bounding boxes (coordinate information) of text region images and determine which regions in the image contain text. It then marks the location and extent of each text region image, achieving automatic localization of text region images within an image.

[0076] Furthermore, the coordinates and range of the text region images in each target file image are output to clarify which regions need to be recognized by characters, thus identifying multiple text region images.

[0077] Step S3023: Using a convolutional neural network model, character recognition is performed on multiple text region images to obtain multiple first legal compliance documents.

[0078] Among them, the Convolutional Neural Network (CNN) model represents a deep learning model used in fields such as image recognition, object detection, and image generation.

[0079] Specifically, preprocessing can be performed on images of multiple defined text regions. For example, individual text region images can be cropped and subjected to operations such as grayscale conversion, normalization, and noise reduction to optimize character recognition conditions.

[0080] Furthermore, the preprocessed text region image is input into a convolutional neural network model. This model can then extract character features and match them with a trained character feature library through multi-layer convolution, pooling, and fully connected operations, recognizing each character and converting the text content in the image into computer-editable text data.

[0081] Furthermore, the recognition results of all text regions in each target file image can be integrated, and the text can be spliced ​​together according to the position order of the text regions in the image (such as from top to bottom, from left to right) to restore the text content in the image and form the initial draft of the corresponding legal compliance document, thus obtaining the first legal compliance document.

[0082] Step S3024: The multiple first legal compliance documents are structured to obtain multiple initial legal compliance documents.

[0083] Specifically, the first legal compliance document identified can be structured to restore the original document's paragraphs, tables, headings, and other layout information as much as possible, which will facilitate subsequent semantic analysis and review.

[0084] In some alternative implementations, a combination of deep learning models (to identify layout patterns) and rule engines (based on common layout rules for legal compliance documents) can be used to identify layout elements such as paragraphs, tables, and headings in the text.

[0085] Then, the plain text content of the first legal compliance document is analyzed to extract features such as line breaks, indentation, font style, and character spacing. Combined with the position information of the text areas determined in step S3022 (such as which text areas belong to tables and which belong to paragraphs), the layout structure of the text is determined.

[0086] Furthermore, the paragraphs, tables, heading levels, etc. contained in the document can be structurally restored to obtain multiple initial legal compliance documents that have been structurally processed and restored to their original layout information, including paragraphs, tables, and headings.

[0087] Furthermore, the structured text can be validated by comparing it with the layout of the original target file image, adjusting and restoring inaccurate parts, ensuring that the layout information such as paragraphs, tables, and headings is as close as possible to the original document, and forming an initial legal compliance document with a clear structure and conforming to the original document format.

[0088] Step S303: Using preset multi-dimensional semantic recognition rules, multiple initial legal compliance documents are reviewed to obtain multiple target legal compliance documents that have passed the review.

[0089] Specifically, step S303 includes:

[0090] Step S3031: Determine the first-dimensional verification rule, the second-dimensional verification rule, and the third-dimensional verification rule according to the preset multi-dimensional semantic recognition rules.

[0091] Specifically, the preset multi-dimensional semantic recognition rules include three different dimensions of verification rules, which are used to review documents from different dimensions.

[0092] Furthermore, the first-dimensional verification rules are divided into two categories: the first category explicitly states that projects without project names in the legal compliance documents are not verified; the second category explicitly states that projects with project names in the legal compliance documents need to be verified.

[0093] Furthermore, the third-dimensional verification rules are divided into two categories: the first category explicitly states that no verification is required if semantic judgment is not needed; the second category explicitly states that a preliminary judgment on the basic semantics of the obtained initial legal compliance documents is made based on the determined documents that need to be verified.

[0094] Step S3032: Using the first-dimensional verification rules, the project names of multiple initial legal compliance documents are reviewed to obtain multiple approved second legal compliance documents.

[0095] In some optional implementations, step S3032 above includes:

[0096] Step a1: Determine whether the multiple initial legal compliance documents contain a project name, and obtain multiple fourth legal compliance documents containing a project name and multiple fifth legal compliance documents not containing a project name.

[0097] Specifically, text extraction technology can be used to extract the "project name" field from each initial legal compliance document.

[0098] Furthermore, if a specific project name can be extracted, the corresponding initial legal compliance document is marked as the fourth legal compliance document; if the project name cannot be extracted, the corresponding initial legal compliance document is marked as the fifth legal compliance document.

[0099] Step a2: Using the first-dimensional verification rules, the project names of multiple fourth-level legal compliance documents are matched and verified to obtain multiple sixth-level legal compliance documents that pass the matching verification.

[0100] Specifically, for each fourth legal compliance document, extract its project name.

[0101] Furthermore, following the first dimension verification rules, the operation of "removing commonly used non-core words" is performed to obtain the corresponding simplified project name.

[0102] Furthermore, a preset fuzzy matching algorithm can be used to perform a fuzzy match between the simplified project name and the pre-filled project name in the write-in storage. The preset fuzzy matching algorithm can be an edit distance algorithm, cosine similarity, etc.

[0103] Furthermore, if the match is successful, the corresponding fourth legal compliance document is marked as the sixth legal compliance document and the next verification step is initiated; if the match fails, an error is reported.

[0104] Step a3: Based on multiple fifth and multiple sixth legal compliance documents, identify multiple second legal compliance documents.

[0105] Specifically, the obtained fifth and sixth legal compliance documents are integrated to form corresponding second legal compliance documents.

[0106] Step S3033: Using the second-dimensional verification rules and the preset element data dictionary, the key element data of multiple second legal compliance documents are reviewed to obtain multiple third legal compliance documents that have passed the review.

[0107] Specifically, the business type and audit procedure name of each second legal compliance document are identified, and the list of key elements of the corresponding business and audit procedure is retrieved from the preset element data dictionary.

[0108] Furthermore, text extraction techniques such as regular expressions and named entity recognition can be used to extract key elements required by the rules from the second legal compliance document.

[0109] Furthermore, the extracted elements are compared one by one with the "element dictionary rules". If all key elements are complete and comply with the rules, the verification is successful and the corresponding second legal compliance document is marked as the third legal compliance document.

[0110] Furthermore, if any key element is missing or violates the rules, an error is flagged and the issue is processed manually.

[0111] Furthermore, when generating the preset element data dictionary, in addition to considering the business nature of the review procedures, it can also fully reference various laws and regulations that are updated in a timely manner in the relevant legal database. Through flexible configuration, legal procedures and review element content can be added, deleted, or modified at any time, ensuring that the intelligent review system always adopts the latest and most applicable rules and fully meets the corresponding legal and compliance requirements.

[0112] Through the above review process, the semantic meaning of the document can be analyzed in detail and the core elements in the text can be selected, making it easier to achieve intelligent review in a faster, more efficient and more accurate manner.

[0113] Step S3034: Using the third-dimensional verification rules, perform semantic review on multiple third-party legal compliance documents to obtain multiple target legal compliance documents that have passed the review.

[0114] In some optional implementations, step S3034 above includes:

[0115] Step b1: Determine whether multiple third-party legal compliance documents need to undergo semantic verification, and obtain multiple seventh-party legal compliance documents that need to undergo semantic verification and multiple eighth-party legal compliance documents that do not need to undergo semantic verification.

[0116] Specifically, for each third-party legal compliance document, it can be determined whether semantic verification is required based on the "scenarios without semantic verification" in the third-dimensional rules.

[0117] Furthermore, if the third legal compliance document falls under a scenario requiring semantic verification, it is marked as the seventh legal compliance document; if the third legal compliance document falls under a scenario not requiring semantic verification, it is marked as the eighth legal compliance document.

[0118] Step b2: Using the third-dimensional verification rules, perform semantic verification on multiple seventh-level legal compliance documents to obtain multiple ninth-level legal compliance documents that pass the verification.

[0119] Specifically, for each seventh legal compliance document, semantic intent can be extracted using natural language processing technologies such as keyword matching and sentiment analysis models.

[0120] Furthermore, if the semantic intent expresses positive content such as support, agreement, or compliance (e.g., containing keywords such as "agree" or "grant"), the verification is considered successful and the corresponding seventh legal compliance document is marked as the ninth legal compliance document; if the semantic intent expresses negative content such as rejection or disapproval, or is semantically ambiguous (e.g., "agree in principle, supplementary materials required"), an error is reported.

[0121] Step b3: Identify multiple target legal compliance documents based on multiple eighth-level legal compliance documents and multiple ninth-level legal compliance documents.

[0122] Specifically, the obtained eighth and ninth legal compliance documents are integrated to form a number of corresponding target legal compliance documents.

[0123] Step S304: Use a preset intent classification model to review multiple target legal compliance documents to obtain multiple document review results for multiple target legal compliance documents.

[0124] The preset intent classification model can be constructed through the following steps:

[0125] Step c1: Obtain multiple historical legal compliance documents pending review.

[0126] The specific process can be referred to in step S201 above, and will not be repeated here.

[0127] Step c2 involves processing multiple historical legal compliance documents pending review using optical character recognition (OCR) to obtain multiple historical target legal compliance documents.

[0128] The specific process can be referred to in step S302 above, and will not be repeated here.

[0129] Step c3 involves vectorizing multiple historical target legal compliance documents to obtain multiple historical text vectors.

[0130] First, perform basic language analysis on historical target legal compliance documents, which may include lexical analysis, syntactic analysis, and semantic role labeling.

[0131] Lexical analysis decomposes document content into basic lexical units and performs operations such as part-of-speech tagging and named entity recognition, providing basic lexical information for subsequent semantic understanding; syntactic analysis constructs a grammatical structure tree of a sentence by studying the grammatical relationships between words, which helps to understand the overall structure and semantic logic of the sentence; semantic role labeling, based on syntactic analysis, further identifies the semantic roles played by each word in the sentence, enabling a more accurate understanding of the sentence's semantic meaning.

[0132] Specifically, lexical analysis is performed on the text content of each historical target legal compliance document, breaking it down into lexical units and labeling them with parts of speech and named entities (such as "project name" and "approval document number").

[0133] Furthermore, syntactic analysis is performed to construct the sentence's grammatical structure tree and identify components such as subject, verb, object, attributive, adverbial, and complement (e.g., in "A agrees to B's construction project," "A" is the subject and "agrees" is the verb).

[0134] Furthermore, semantic role labeling is performed to clarify the semantic role of words (such as "agent", "patient", "action"). For example, in "B's reply", "B" is the agent and "reply" is the action.

[0135] Secondly, based on the above analysis results, the document content is transformed into a form that computers can understand and process using the Vector Space Model (VSM), Word2Vec model, and Pre-trained Language Model (BERT), that is, text vectorization processing is achieved.

[0136] Among them, the vector space model represents words in a document as vectors and measures the semantic similarity between words by calculating the similarity between vectors; the word embedding model maps words to a low-dimensional vector space, so that semantically similar words are closer in the vector space; and the pre-trained language model uses a large-scale corpus for pre-training, learns rich language knowledge and semantic information, and can better capture the semantic features of the document.

[0137] Specifically, the TF-IDF values ​​of words are calculated using the vector space model (SVM), and the text is converted into word frequency vectors to highlight key terms.

[0138] Furthermore, words are mapped to a low-dimensional vector space (such as 200-dimensional), so that words with similar meanings (such as "photovoltaics" and "solar energy") are close in distance in the vector space.

[0139] Furthermore, the input text is used to generate context-sensitive vector representations through the BERT model, and the meaning of words is dynamically adjusted.

[0140] Furthermore, through the above process, each historical file is converted into a fixed-length vector, i.e., multiple historical text vectors.

[0141] Step c4: Based on multiple historical text vectors, use a preset neural network model to identify the intent of multiple historical legal compliance documents to be reviewed, and obtain multiple expressive intents.

[0142] The expression of intent is used to reflect whether the legal history documents to be reviewed are qualified.

[0143] Specifically, due to the complex and diverse document styles and contents of the legal history documents to be reviewed, various deep learning model architectures such as Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN) and their variants Long Short-Term Memory Networks (LSTM), Gated Recurrent Units (GRU), and Transformers can be adopted.

[0144] Convolutional Neural Networks (CNNs) are primarily used to process data with a grid structure, such as images and text. In document review, the document content can be viewed as a two-dimensional grid structure, where each row represents a sentence and each column represents a word. CNNs extract local features from the document through convolutional layers, then perform dimensionality reduction and filtering of these features through pooling layers, and finally perform classification or regression tasks through fully connected layers. For example, when determining whether a document's intended meaning meets requirements, CNNs can learn local feature patterns related to specific intentions, thus enabling accurate classification.

[0145] Furthermore, Recurrent Neural Networks (RNNs) are well-suited for processing sequential data, such as text. In document review, an RNN can process each word in a document sequentially, updating the hidden state based on preceding context information to capture the document's semantic information. However, traditional RNNs suffer from vanishing and exploding gradients, making them difficult to handle long sequences of data. To address these issues, variants of RNNs such as LSTM and GRU have emerged. LSTM, by introducing memory units and gating mechanisms, effectively solves the vanishing and exploding gradient problems, better capturing semantic information in long sequences. GRU is a simplification of LSTM, reducing the number of gating mechanisms while maintaining good performance. In practical applications, the appropriate RNN variant can be selected based on specific needs and data characteristics.

[0146] Furthermore, the Transformer is a deep learning model based on a self-attention mechanism. Specifically, the Transformer can directly model long-distance dependencies between words through self-attention, avoiding the problems of traditional RNNs when dealing with long sequences. In this embodiment, the Transformer can encode the document content into a fixed-length vector representation, and then use a classifier to determine whether the expressed intent meets the requirements.

[0147] Furthermore, the obtained historical text vectors can be input into a pre-defined neural network model constructed based on the aforementioned network. This pre-defined neural network model then extracts semantic features through a multi-layer neural network and outputs the intent classification probability.

[0148] Furthermore, mapping the classification results to expressive intent labels creates multiple corresponding expressive intents.

[0149] Step c5: Using multiple historical text vectors as input and multiple expressive intentions as output, construct a pre-defined intent classification model.

[0150] Specifically, the model is trained using multiple historical text vectors as input data and multiple expressive intentions as output data until a preset intention classification model that meets the requirements is obtained.

[0151] In some alternative implementations, the historical text vectors and the expressive intent labels are first divided into a training set and a validation set in an 8:2 ratio.

[0152] Then, the cross-entropy loss function is used to measure the difference between the predicted intent and the true label.

[0153] Furthermore, the Adam algorithm is used to iteratively update the model parameters and dynamically adjust the learning rate to accelerate convergence.

[0154] Furthermore, by conducting multiple rounds of training and monitoring the accuracy of the validation set, overfitting can be avoided.

[0155] Finally, by calculating accuracy, recall, and F1 score, we ensure that the model performance meets the standards and obtain the corresponding pre-trained intent classification model.

[0156] Specifically, step S304 includes:

[0157] Step S3041: Vectorize multiple target legal compliance documents to obtain multiple target text vectors.

[0158] The specific process can be found in the description of step c3 above, and will not be repeated here.

[0159] Step S3042: Input multiple target text vectors into a preset intent classification model to obtain multiple document review results for multiple target legal compliance documents.

[0160] Specifically, according to the description of step c5, the multiple target text vectors obtained are input into the trained preset intent classification model, which can output the expressive intent of the target legal compliance document corresponding to each target text vector.

[0161] Furthermore, the expressed intent is used to reflect whether the legal history documents to be reviewed are compliant. Therefore, based on the multiple expressed intents output, the compliantness of each target legal compliance document can be determined, resulting in multiple document review results for multiple target legal compliance documents.

[0162] The legal compliance document review method provided in this embodiment solves the complexity of traditional manual processing of multi-format files by uniformly converting multiple unstructured legal compliance documents into image format, thus improving the compatibility of the review system. Furthermore, it automatically detects text regions in images using a deep learning-based text detection algorithm, excluding non-text regions and improving text location accuracy. Furthermore, it utilizes a convolutional neural network model for character recognition, reducing the need for manual word-by-word proofreading and improving recognition efficiency and accuracy. Finally, it converts the recognized documents into structured text, avoiding semantic logic breaks caused by pure text. Furthermore, it processes files without project names separately and performs key element verification directly afterward, avoiding review blockages caused by rigid review processes and improving the adaptability of the intelligent review system to complex documents. Simultaneously, by distinguishing between files with and without project names, it ensures that all types of files can be effectively reviewed, reducing review loopholes. Furthermore, it directly approves files that do not require semantic judgment, reducing review time. Furthermore, it performs deep semantic verification on files requiring semantic judgment, avoiding over-review and balancing review efficiency and accuracy.

[0163] This embodiment provides a method for reviewing legal compliance documents, which can be used for, for example Figure 1The document information system 11 shown is connected to the intelligent auditing system 12, the user terminal 2, and the management terminal 3. Figure 4 This is a flowchart of a legal compliance document review method according to an embodiment of the present invention, such as... Figure 4 As shown, the process includes the following steps:

[0164] Step S401: Receive multiple legal and compliance documents to be reviewed from the user terminal.

[0165] Specifically, users can upload the corresponding legal compliance documents to be reviewed through user terminal 2.

[0166] Furthermore, user terminal 2 sends the received multiple legal compliance documents pending review to the corresponding document information system 11.

[0167] Step S402: Determine whether any confidential documents exist among the multiple pending legal compliance documents.

[0168] Specifically, when reviewing confidential documents, it is necessary to ensure their security. Therefore, before sending the legal compliance documents to be reviewed to the intelligent review system 12, it is necessary to first determine whether any of the received legal compliance documents are confidential.

[0169] Furthermore, the document information system 11 can determine whether the received legal compliance documents to be reviewed are confidential documents by means of document tagging or content recognition.

[0170] Step S403: If no confidential documents exist, send multiple legal compliance documents to be reviewed to the intelligent review system.

[0171] The intelligent auditing system executes the legal compliance document auditing method provided in the above embodiments of the present invention.

[0172] Specifically, if none of the received legal compliance documents to be reviewed are confidential, the received legal compliance documents to be reviewed can be sent directly to the intelligent review system 12.

[0173] Furthermore, the intelligent review system 12 performs intelligent review on the received multiple legal compliance documents to be reviewed by executing the legal compliance document review method provided in the above embodiments of the present invention.

[0174] Step S404: Receive the review results of multiple documents sent by the intelligent review system.

[0175] Specifically, after the intelligent review system 12 completes the review, it sends the review results of multiple documents to the document information system 11.

[0176] The legal compliance document review method provided in this embodiment improves review efficiency and accuracy by sending non-confidential legal compliance documents to an intelligent review system for batch review.

[0177] In some optional implementations, the above method further includes: when there are confidential documents, using a preset multi-layer protection mechanism to review multiple legal compliance documents that are to be reviewed and contain confidential documents.

[0178] Among them, the preset protection mechanism refers to the multi-level and three-dimensional security control strategy and operation process that the document information system 11 pre-sets to ensure the security and compliance of confidential documents in the review process.

[0179] First, account permission isolation is implemented through the first level.

[0180] Specifically, the file information system 11 reads the account permissions of the user terminal 2 that uploaded the currently confidential file and determines whether it has the permission to "view / process confidential files".

[0181] Furthermore, if the conditions are not met, the confidential document will be automatically hidden and will not be able to proceed to the next review process; if the conditions are met, it will proceed to the second level of manual review.

[0182] Specifically, after entering the second-level manual review process, the document information system 11 marks the relevant classified procedures in the rule table. If the classified document is identified as belonging to the corresponding category, it will automatically enter the review process.

[0183] Furthermore, once the review process begins, user terminal 2 cannot directly upload the legal compliance documents to be reviewed to the document information system 11. In this case, the legal compliance documents to be reviewed can be transmitted to legal personnel for manual review through secure means (such as encrypted USB flash drive, confidential dedicated line, offline confidential channel).

[0184] Furthermore, legal personnel can upload the results of manual review to the document information system 11.

[0185] Furthermore, when the document information system 11 receives multiple legal compliance documents pending review from the user terminal 2, the review process of the intelligent review system 12 is automatically triggered. First, the symmetric key (AES-256) is encrypted using a high-strength asymmetric encryption algorithm (RSA-2048). Simultaneously, the symmetric key is used to encrypt the multiple existing confidential legal compliance documents pending review, resulting in multiple encrypted legal compliance documents.

[0186] Furthermore, the encrypted symmetric key and multiple encrypted legal compliance documents are sent to the intelligent audit system 12.

[0187] Furthermore, the intelligent review system 12 performs intelligent review on the received multiple legal compliance documents to be reviewed by executing the legal compliance document review method provided in the above embodiments of the present invention.

[0188] In some optional implementations, the document information system 11 is also connected to the management terminal 3; the above method further includes:

[0189] Step d1: Send the review results of multiple documents to the user terminal.

[0190] Step d2: When the user terminal sends the review result appeal information, the review result appeal information is sent to the management terminal so that the management terminal can review the review result of the document corresponding to the review result appeal information.

[0191] Step d3: When the appeal information is approved by the management terminal, update the review result of the pending legal compliance documents corresponding to the appeal information, and send the updated review result to the user terminal.

[0192] Step d4: When the appeal information review is not approved by the management terminal, the appeal information review result is sent to the user terminal.

[0193] Specifically, the document information system 11 can send the received review results of multiple documents to the user terminal 2 so that the user can view the review results through the user terminal 2.

[0194] Furthermore, if a user disagrees with the review result, they can upload the corresponding review result appeal information through user terminal 2.

[0195] Furthermore, the user terminal 2 sends the received review result appeal information to the management terminal 3 through the document information system 11 so that the corresponding reviewer can view the appeal information.

[0196] Furthermore, the reviewers can review the received appeal information and send the review results to the user terminal 2 through the document information system 11, so that the user can view the review results through the user terminal 2.

[0197] Specifically, if the review is successful, the reviewer can send the appeal information review approval instruction to the document information system 11 through the management terminal 3.

[0198] Furthermore, upon receiving the instruction that the appeal information has been approved, the document information system 11 can update the review result of the legal compliance document to be reviewed corresponding to the appeal information and send the updated review result to the user terminal 2 so that the user can view it through the user terminal 2.

[0199] Furthermore, after receiving the instruction to approve the appeal information, the document information system 11 can also determine whether the review rules in the intelligent review system 12 need to be adjusted. If so, it sends an adjustment instruction to the intelligent review system 12 to optimize the review rules.

[0200] Furthermore, upon receiving an instruction that the appeal information review has failed, the document information system 11 can send the corresponding appeal information review result to the user terminal 2, so that the user can view the appeal information review result through the user terminal 2.

[0201] Furthermore, users can modify the pending legal compliance documents based on the appeal information review results and upload the modified pending legal compliance documents to the document information system 11, and then execute the above-mentioned legal compliance document review method again.

[0202] The legal compliance document review method provided in this embodiment allows user terminals to appeal the review results. Furthermore, the management terminal can review the appeal information and update the review results when the review is approved, thereby improving the credibility of the review.

[0203] This embodiment provides a legal compliance document review system, such as Figure 1 As shown, the legal compliance document review system 1 includes a document information system 11, an intelligent review system 12, and a rule management system 13.

[0204] The document information system 11 is connected to the corresponding user terminal 2 and management terminal 3.

[0205] Specifically, the rule management system 13 is used to obtain preset multi-dimensional semantic recognition rules and send the preset multi-dimensional semantic recognition rules to the intelligent review system. The preset multi-dimensional semantic recognition rules are described in step S203 above and will not be repeated here.

[0206] Furthermore, the intelligent auditing system 12 and the document information system 11 respectively execute the legal compliance document auditing method provided in the above embodiments.

[0207] The legal compliance document review system provided in this embodiment can review multiple legal compliance documents at once by executing the corresponding legal compliance document review methods on the intelligent review system and the document information system respectively. This replaces the manual processing mode of one document at a time, reduces the workload of legal personnel in manual review, lowers labor costs, and improves the efficiency and accuracy of review. In turn, it helps to strengthen the compliance of project procedure management and the risk control capabilities of enterprises.

[0208] This invention also provides a computer device for performing the above-described... Figures 2 to 4 The method for reviewing legal compliance documents is shown.

[0209] Please see Figure 5 , Figure 5 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 5 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 5 Take a processor 10 as an example.

[0210] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0211] The memory 20 stores instructions executable by at least one processor 10 to cause at least one processor 10 to perform the method shown in the above embodiments.

[0212] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0213] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0214] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0215] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0216] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0217] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for reviewing legal compliance documents, characterized in that, For use in intelligent auditing systems; the method includes: Obtain multiple pending legal and compliance documents for the target new energy project; Based on the multiple legal compliance documents pending review, multiple initial legal compliance documents are obtained through optical character recognition (OCR) methods. By using preset multi-dimensional semantic recognition rules, the multiple initial legal compliance documents are reviewed to obtain multiple target legal compliance documents that have passed the review; The multiple target legal compliance documents are reviewed using a pre-defined intent classification model to obtain multiple document review results for the multiple target legal compliance documents.

2. The method according to claim 1, characterized in that, Based on the aforementioned multiple legal compliance documents awaiting review, several initial legal compliance documents were obtained through optical character recognition (OCR) processing, including: The multiple legal compliance documents pending review are processed by image conversion to obtain multiple target document images; The multiple target file images are processed using a deep learning-based text detection algorithm to identify multiple text region images; Using a convolutional neural network model, character recognition is performed on the multiple text region images to obtain multiple first legal compliance documents; The plurality of first legal compliance documents are structured to obtain the plurality of initial legal compliance documents.

3. The method according to claim 1, characterized in that, Using preset multi-dimensional semantic recognition rules, the multiple initial legal compliance documents are reviewed to obtain multiple target legal compliance documents that have passed the review, including: Based on the preset multi-dimensional semantic recognition rules, the first-dimensional verification rule, the second-dimensional verification rule, and the third-dimensional verification rule are determined. Using the first dimension verification rules, the project names of the multiple initial legal compliance documents are reviewed to obtain multiple approved second legal compliance documents; Using the second dimension verification rules and a preset element data dictionary, the key element data of the multiple second legal compliance documents are reviewed to obtain multiple third legal compliance documents that have passed the review; Using the aforementioned third-dimensional verification rules, semantic audits are performed on the multiple third-party legal compliance documents to obtain the multiple target legal compliance documents that have passed the audit.

4. The method according to claim 3, characterized in that, Using the first dimension verification rules, the project names of the multiple initial legal compliance documents are reviewed to obtain multiple approved second legal compliance documents, including: Determine whether the multiple initial legal compliance documents contain a project name, and obtain multiple fourth legal compliance documents containing a project name and multiple fifth legal compliance documents not containing a project name; Using the first dimension verification rules, the project names of the multiple fourth legal compliance documents are matched and verified to obtain multiple sixth legal compliance documents that pass the matching verification. The plurality of second legal compliance documents are determined based on the plurality of fifth legal compliance documents and the plurality of sixth legal compliance documents.

5. The method according to claim 3, characterized in that, Using the aforementioned third-dimensional verification rules, semantic auditing is performed on the multiple third-party legal compliance documents to obtain the multiple target legal compliance documents that have passed the audit, including: Determine whether the multiple third-party legal compliance documents need to undergo semantic verification, and obtain multiple seventh-party legal compliance documents that need to undergo semantic verification and multiple eighth-party legal compliance documents that do not need to undergo semantic verification; Using the aforementioned third-dimensional verification rules, semantic verification is performed on the multiple seventh legal compliance documents to obtain multiple ninth legal compliance documents that pass the verification. The plurality of target legal compliance documents are determined based on the plurality of eighth legal compliance documents and the plurality of ninth legal compliance documents.

6. The method according to claim 1, characterized in that, The method further includes: Obtain multiple pending historical legal compliance documents; Based on the multiple historical legal compliance documents pending review, and processed by the optical character recognition method, multiple historical target legal compliance documents are obtained; The aforementioned historical target legal compliance documents are vectorized to obtain multiple historical text vectors; Based on the multiple historical text vectors, a preset neural network model is used to identify the intent of the multiple historical legal compliance documents to be reviewed, resulting in multiple expressive intents. The expressive intents are used to reflect whether the historical legal documents to be reviewed are qualified. Using the multiple historical text vectors as input and the multiple expressive intentions as output, the preset intention classification model is constructed.

7. The method according to claim 6, characterized in that, The multiple target legal compliance documents are reviewed using a pre-defined intent classification model to obtain multiple document review results for the multiple target legal compliance documents, including: The multiple target legal compliance documents are vectorized to obtain multiple target text vectors; The multiple target text vectors are input into the preset intent classification model to obtain multiple document review results for the multiple target legal compliance documents.

8. The method according to claim 1, characterized in that, The method further includes: Once multiple encrypted legal compliance documents for the target new energy project are obtained, these documents are decrypted to obtain the multiple legal compliance documents pending review.

9. A method for reviewing legal compliance documents, characterized in that, Used in a document information system, connected to both a user terminal and an intelligent review system; the method includes: Receive multiple pending legal and compliance documents sent by the user terminal; Determine whether any confidential documents exist among the multiple pending legal compliance documents; When no confidential documents exist, the plurality of legal compliance documents to be reviewed are sent to the intelligent review system, wherein the intelligent review system performs the legal compliance document review method according to any one of claims 1 to 8; Receive multiple file review results sent by the intelligent review system.

10. The method according to claim 9, characterized in that, The document information system is also connected to a management terminal; the method further includes: The review results of the multiple documents are sent to the user terminal; When the user terminal sends an appeal message for the review result, the appeal message is sent to the management terminal so that the management terminal can review the review result of the file corresponding to the appeal message. When the management terminal receives the appeal information approval instruction, it updates the review result of the pending legal compliance document corresponding to the appeal information and sends the updated review result to the user terminal. When the management terminal receives an instruction indicating that the appeal information review has failed, it sends the appeal information review result to the user terminal.

11. A legal compliance document review system, characterized in that, The system includes: a document information system, an intelligent auditing system, and a rule management system; The rule management system is used to acquire preset multi-dimensional semantic recognition rules and send the preset multi-dimensional semantic recognition rules to the intelligent review system; The intelligent review system is used to perform the legal compliance document review method according to any one of claims 1 to 8; The document information system is used to perform the legal compliance document review method as described in claim 9 or 10.

Images