Intelligent file editing and auditing system and method and storage medium

Abstract

The invention relates to the technical field of artificial intelligence, in particular to an intelligent file editing and auditing system and method and a storage medium, and the method comprises the following steps: receiving a purchase file; vectorization processing is carried out on the purchase file; newly establishing a purchase document; relevant data are recalled from the vectorized data according to target information required by the chapters, structured conversion is carried out, and the data are added into the corresponding chapters, so that compilation of the purchase document is completed; the compiled purchase document is preprocessed; the method comprises the following steps: reading review requirements from a preset document review point library, extracting related review contents from vectorized data of a purchase document, performing review judgment on the related review contents according to the review requirements by utilizing a large language model, and outputting a review judgment result, thereby completing intelligent review of the purchase document. According to the method, manual content retrieval, extraction, integration, filling and auditing work can be replaced, the workload and time are greatly saved, and meanwhile the accuracy and speed of information retrieval are obviously improved compared with manual information retrieval.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and specifically to an intelligent document editing and reviewing system, method, and storage medium. Background Technology

[0002] In the procurement process, when project managers prepare tender documents, they often need to refer to the requirements documents already provided by the purchaser, or procurement documents that have been collaboratively prepared. In electronic procurement processes, project managers often need to extract relevant chapters, paragraphs, and key elements from documents and fill them into the corresponding form fields in the system. This process is labor-intensive for procurement personnel and prone to errors; after completion, it usually requires repeated and careful verification. Furthermore, with the increasing refinement of procurement process management, more and more companies have standardized requirements for the content and format of procurement documents. Both project managers preparing documents and approvers reviewing documents against regulations and procedures, which consumes considerable time and effort. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide an intelligent document editing and review system, method and storage medium to solve the problems of large workload, error-proneness and time and energy wasted due to the need for manual input of relevant information in the existing electronic procurement process, and to solve the problem of time and energy wasted due to the need for repeated verification. It is also used to solve the problem of time and energy wasted due to the need for manual review of documents line by line.

[0004] The technical solution to achieve the above objectives is:

[0005] This invention provides an intelligent document editing and review method, comprising the following steps:

[0006] Receive uploaded procurement documents;

[0007] The received procurement documents are vectorized to obtain basic vectorized data;

[0008] Receive the selected file template;

[0009] Create a new procurement document based on the selected document template;

[0010] Retrieve relevant data from the basic vectorized data according to the target information required in the chapters of the newly created procurement document, perform structured transformation on the retrieved relevant data and add it to the corresponding chapters to complete the compilation of the procurement document;

[0011] After the procurement documents are prepared, they are preprocessed to obtain vectorized data of the procurement documents.

[0012] Upon receiving the intelligent review instruction, the system reads the review requirements from the preset document review point library, extracts relevant review content from the vectorized data of the procurement document based on the read review requirements, uses a large language model to review and judge the relevant review content according to the review requirements, and outputs the review judgment result, thereby completing the intelligent review of the procurement document.

[0013] A further improvement to the intelligent document editing and review method of the present invention lies in the following steps: Vectorizing the received procurement documents to obtain basic vectorized data.

[0014] The received procurement documents are divided into sections according to the first-level directory.

[0015] Read the chapter content page by page from the segmented file;

[0016] A large language model is used to extract summaries from the read chapter content;

[0017] The chapter content summaries are vectorized and stored in the database, and the chapter content is placed into metadata and vectorized and stored in the database.

[0018] A further improvement to the intelligent document editing and reviewing method of the present invention lies in the following steps for retrieving relevant data from the basic vectorized data:

[0019] Retrieve relevant sections from the underlying vectorized data based on the sections of the procurement documents;

[0020] The relevant chapters in the recall were ranked using a re-ranking model;

[0021] The relevant chapters are retrieved sequentially, and the retrieval process is terminated once relevant information is found. The retrieved information is then used as the relevant data for the recall.

[0022] A further improvement to the intelligent document editing and reviewing method of the present invention lies in the following steps for the structured transformation of the recalled relevant data:

[0023] The relevant recall data is structured according to the information type. If the information type is a numerical type, it is formatted according to the requirements of the numerical type to complete the structured conversion of the numerical type.

[0024] If the information type is date, then format it according to the requirements of the date type to complete the structured conversion of the date type;

[0025] If the information type is a data dictionary type, then the large language model is used to match the format and return the corresponding data dictionary information to complete the structured transformation of the data dictionary type.

[0026] A further improvement of the intelligent document editing and review method of the present invention is that, when compiling procurement documents, semantic similarity retrieval is performed in a vectorized database using keywords based on the evaluation method;

[0027] Based on the search results, retrieve chapters and fragments related to the bid evaluation method, and then piece together the retrieved chapters and fragments to form the bid evaluation content paragraphs;

[0028] Using a large language model, generate clause node names, clause names, clause descriptions, and clause scores based on the paragraphs of the evaluation content, and output them in Markdown table format;

[0029] The Markdown tables output by the large language model are parsed to form structured evaluation method data, which is then added to the review method section of the procurement document.

[0030] The present invention also provides a storage medium storing a program for an intelligent document editing and review method, wherein the program for the intelligent document editing and review method is executed by a processor to implement the steps of the intelligent document editing and review method.

[0031] This invention also provides an intelligent document editing and review system, comprising:

[0032] The receiving unit is used to receive uploaded procurement documents, selected document templates, and intelligent review instructions;

[0033] A preprocessing unit, connected to the receiving unit, is used to perform vectorization processing on the received procurement documents to obtain basic vectorized data;

[0034] The document compilation unit, connected to the receiving unit and the preprocessing unit, is used to create a new procurement document based on the selected document template, and is also used to retrieve relevant data from the basic vectorized data according to the target information required by the chapters in the newly created procurement document, perform structured transformation on the retrieved relevant data and add it to the corresponding chapters to complete the compilation of the procurement document;

[0035] After the procurement document is prepared, the preprocessing unit is also used to preprocess the prepared procurement document to obtain vectorized data of the procurement document.

[0036] The document review unit, connected to the preprocessing unit, is used to read review requirements from a preset document review point library after receiving an intelligent review instruction, extract relevant review content from the vectorized data of the procurement document based on the read review requirements, use a large language model to review and judge the relevant review content according to the review requirements, and output the review judgment result, thereby completing the intelligent review of the procurement document.

[0037] A further improvement of the intelligent document editing and review system of the present invention is that the preprocessing unit is used to split the received procurement documents according to the first-level directory.

[0038] Read the chapter content page by page from the segmented file;

[0039] A large language model is used to extract summaries from the read chapter content;

[0040] The chapter content summaries are vectorized and stored in the database, and the chapter content is placed into metadata and vectorized and stored in the database.

[0041] A further improvement of the intelligent document editing and reviewing system of the present invention is that the document editing unit is also used to recall relevant chapters from the basic vectorized data according to the chapters of the procurement document;

[0042] The relevant chapters in the recall were ranked using a re-ranking model;

[0043] The relevant chapters are retrieved sequentially, and the retrieval process is terminated once relevant information is found. The retrieved information is then used as the relevant data for the recall.

[0044] A further improvement of the intelligent document editing and reviewing system of the present invention is that the document editing unit is also used to perform semantic similarity retrieval in a vectorized database using keywords for the evaluation method;

[0045] Based on the search results, retrieve chapters and fragments related to the bid evaluation method, and then piece together the retrieved chapters and fragments to form the bid evaluation content paragraphs;

[0046] Using a large language model, generate clause node names, clause names, clause descriptions, and clause scores based on the paragraphs of the evaluation content, and output them in Markdown table format;

[0047] The Markdown tables output by the large language model are parsed to form structured evaluation method data, which is then added to the review method section of the procurement document.

[0048] The beneficial effects of the intelligent document editing and reviewing system, method, and storage medium of this invention are as follows:

[0049] The present invention relates to an intelligent document editing and review system and method, which utilizes natural language processing technologies such as general large language model, vectorization model, and reordering model. Through in-depth understanding and sorting of the content, structure, compilation, and review process of procurement documents, it constructs an intelligent document editing and review agent based on a large language model.

[0050] This invention relates to an intelligent document editing and review system and method. In the document creation stage, this invention can replace manual work in content retrieval, extraction, integration, and filling. The creator only needs to check, verify, and adjust the AI-processed results, greatly reducing the workload of document creation. Simultaneously, the accuracy and speed of information retrieval are significantly improved compared to manual work. In the document compliance review stage, this invention will understand each review point rule, extract relevant content, and provide suggested review results and modification opinions. It can effectively assist both the creator in self-checking and optimizing documents and the approver in reviewing documents, reducing the work of manually reviewing documents and comparing rules, while also reducing the risk of omissions. Attached Figure Description

[0051] Figure 1 This is a flowchart of the intelligent document editing and review method of the present invention.

[0052] Figure 2 This is a system diagram of the intelligent document editing and review system of the present invention. Detailed Implementation

[0053] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0054] See Figure 1 This invention provides an intelligent document editing and review system, method, and storage medium, aiming to improve the efficiency of procurement personnel in preparing bidding documents within the system, reduce repetitive manual labor, and facilitate the work of both editors and reviewers. This invention utilizes the powerful understanding and reasoning capabilities of large language models for natural language, solving problems such as intelligent identification and extraction of non-standard document content, online visual editing, template-based variable substitution and synchronization. It also achieves natural language understanding based on document review points and automatically extracts and reviews document content for compliance. The intelligent document editing and review system, method, and storage medium of this invention are described below with reference to the accompanying drawings.

[0055] See Figure 2 This diagram shows the system diagram of the intelligent document editing and reviewing system of the present invention. The following is in conjunction with... Figure 2 The present invention provides a description of the intelligent document editing and review system.

[0056] like Figure 2As shown, the intelligent document editing and review system of the present invention includes a receiving unit 21, a preprocessing unit 22, a document editing unit 23, and a document review unit 24. The receiving unit 21 is connected to the preprocessing unit 22, the document editing unit 23 is connected to both the receiving unit 21 and the preprocessing unit 22, and the document review unit 24 is connected to the preprocessing unit 22. The receiving unit 21 is used to receive uploaded procurement documents, selected document templates, and intelligent review instructions. The preprocessing unit 22 is used to perform vectorization processing on the received procurement documents to obtain basic vectorized data. The document editing unit 23 is used to create new procurement documents according to the selected document templates, and is also used to select target information according to the chapters required in the newly created procurement documents. The system retrieves relevant data from the basic vectorized data, performs structured transformation on the retrieved data, and adds it to the corresponding chapters to complete the preparation of the procurement document. After the procurement document is prepared, the preprocessing unit 22 is also used to preprocess the prepared procurement document to obtain the vectorized data of the procurement document. The document review unit 24 is used to read the review requirements from the preset document review point library after receiving the intelligent review instruction, extract the relevant review content from the vectorized data of the procurement document based on the read review requirements, use the large language model to review and judge the relevant review content according to the review requirements, and output the review judgment result, thereby completing the intelligent review of the procurement document.

[0057] The intelligent document editing and review system of this invention achieves intelligent, chapter-based, and visual document editing by deeply analyzing and sorting out the business links and related management system requirements of the electronic procurement process for preparing bidding documents.

[0058] The intelligent document editing and review system of this invention is connected to the electronic procurement system. After the procurement documents are reviewed, they can be directly synchronized to the electronic procurement system for bidding. This can greatly reduce the workload of the personnel who prepare and review the procurement documents, and also save time and energy.

[0059] The intelligent document editing and review system of the present invention also includes an operation interface. The operation interface allows users to easily upload relevant procurement documents and select suitable document templates. When uploading procurement documents, users can upload procurement documents of any format. The operation interface displays a corresponding upload box and upload button. After selecting the procurement document address and clicking the upload button, the uploaded procurement document is received by the receiving unit 21. Correspondingly, the operation interface also displays a button for selecting a file template. Clicking this button leads to a sub-interface for selecting a file template. After selecting the appropriate file template, the user returns to the original interface. The selected file template is received by the receiving unit 21.

[0060] In one specific embodiment of the present invention, after receiving the uploaded procurement document, the receiving unit 21 automatically identifies the content of the uploaded procurement document. First, the preprocessing unit 22 preprocesses the procurement document. Specifically, the preprocessing unit 22 is used to segment the received procurement document according to a first-level directory; read the chapter content page by page of the segmented document; extract a summary of the read chapter content using a large language model; vectorize the chapter content summary and store it in the database; and then vectorize the chapter content into the metadata database. In other words, the chapter content and its summary are stored as basic vectorized data in a vector database.

[0061] The intelligent document editing and review system of the present invention reads the document chapter directory according to rules and segments the document according to the first-level directory, which can improve the accuracy of content recognition and avoid interference between related content in different chapters during the content extraction and recall stage. Then, the content of each chapter is summarized and vectorized into the database.

[0062] Furthermore, during document preparation, information is extracted from the basic vectorized data in the vector database to complete the compilation and review of procurement documents.

[0063] The document compilation unit 23 is also used to recall relevant chapters from the basic vectorized data according to the chapters of the procurement document; sort the recalled relevant chapters using a re-ranking model; retrieve the content of the recalled relevant chapters one by one in sequence; terminate the retrieval process after retrieving relevant information; and use the retrieved relevant information as the relevant data for recall.

[0064] During the document compilation process, secondary sorting of the recalled data can further improve the quality of recall ranking. First, the recalled chapters are sorted according to their relevance. Then, a re-ranking model is used for secondary sorting. Subsequently, the target information content is retrieved one by one in the recall results in order. If relevant information is found, the retrieval process is terminated.

[0065] Furthermore, the document compilation unit 23 is also used to perform structured transformation on the recalled relevant data according to the information type. If the information type is a numerical type, it is formatted according to the requirements of the numerical type to complete the structured transformation of the numerical type; if the information type is a date type, it is formatted according to the requirements of the date type to complete the structured transformation of the date type; if the information type is a data dictionary type, it uses a large language model to match the format (when matching the format, the large model matching is the most likely option) and returns the corresponding data dictionary information to complete the structured transformation of the data dictionary type.

[0066] When retrieving and extracting content from the chapters of the procurement document, the document compilation unit 23 can retrieve the content of each chapter sequentially from beginning to end, and then perform structured transformation on the retrieved data and fill it into the corresponding chapter until the compilation of the entire procurement document is completed.

[0067] In one specific embodiment of the present invention, the document compilation unit 23 of the present invention is also used to perform semantic similarity retrieval in a vectorized database using keywords for the evaluation method;

[0068] Based on the search results, retrieve chapters and fragments related to the bid evaluation method, and then piece together the retrieved chapters and fragments to form the bid evaluation content paragraphs;

[0069] Using a large language model, generate clause node names, clause names, clause descriptions, and clause scores based on the paragraphs of the evaluation content, and output them in Markdown table format;

[0070] The Markdown tables output by the large language model are parsed to form structured evaluation method data, which is then added to the review method section of the procurement document.

[0071] Furthermore, regarding the evaluation method, users can upload tender documents containing the evaluation method and terms to the receiving unit. The intelligent document editing and review system will then asynchronously execute the subsequent complete intelligent enhancement logic for the evaluation method. Upon receiving the tender documents, the receiving unit downloads them from the server to the application server locally. The preprocessing unit 22 reads the content of the tender documents, intelligently recognizing chapter titles and paragraph content and converting them into Markdown format.

[0072] During the reading of the tender document, the preprocessing unit 22 segments the document content into fragments based on the main headings. When table elements are encountered in the document, they are converted into Markdown format table data. If the document fragment length is too long, the excessively long fragments are further segmented based on the character length and token length, and finally stored in the vector database.

[0073] Furthermore, the document compilation unit 23 performs semantic similarity retrieval using built-in keywords such as "evaluation method" and "review criteria," recalling chapter fragments related to the evaluation method content and concatenating multiple recalled fragments into a single piece of content. The recalled evaluation method-related content is then input into the large language model using prompts provided by the large model, guiding the model to generate content including evaluation clause node names, clause names, clause descriptions, and clause scores, outputting it in Markdown table format. By parsing the Markdown table generated by the large language model, structured evaluation method data is formed. Then, all review clauses are traversed, and the large model supplements other attribute information for the clauses (including whether they are scoring items, whether they are objective items, etc.), ultimately forming structured evaluation clause data, which is then output to the front-end page for display.

[0074] When compiling the content of each chapter, the document compilation unit 23 can simultaneously display the compiled content of each chapter on the operation interface for easy viewing by users.

[0075] In one specific embodiment of the present invention, after the procurement document is completed, a prompt message indicating that the document has been completed can be displayed on the operation interface to remind the user that the document has been completed.

[0076] Once the procurement document is completed, clicking the save button on the user interface generates a vectorized data entry instruction. This prompts the preprocessing unit 22 to vectorize the completed procurement document, dividing it into chapters and tables of contents. A large language model is then used to summarize the chapter content, which is then vectorized and stored in the vector database along with the metadata. This vectorized data from the procurement document is used for subsequent intelligent document review.

[0077] The user interface of the intelligent document editing and review system of this invention allows for the configuration of review rules, review methods, review point names, and corresponding chapters for the document review point library. The configured document review point library is stored in a database for subsequent retrieval and use by the document review unit.

[0078] Document review unit 24 reads the review rules from the document review point database, understands each rule, and based on the description information of the review points, understands the review target content and the specific requirements of the review item for that content. After understanding the review rules, document review unit 24 extracts relevant information about the review target content from the vectorized database of procurement documents, inputs the review target and review requirements into the large language model, and uses the natural language understanding and reasoning capabilities of the large language model to determine whether the target content meets the review point requirements, provides reasons and justifications, and offers modification suggestions. The review judgment result of the large language model includes whether the review point requirements are met; if the result is not met, the reasons for the non-compliance are given, and modification suggestions are provided.

[0079] The intelligent document editing and review system of this invention allows project managers to directly upload offline documents (i.e., procurement documents) to the system during the procurement document preparation stage. The system automatically performs full-text search on the documents, extracts and structures key information, and automatically fills it into the information system, simultaneously replacing the standardized procurement document template. The project manager only needs to verify the document extraction and transformation results to complete the structured preparation of the procurement documents. After the document is completed, the system automatically extracts relevant content and checks the rules based on the review rules maintained in the document review point library, providing review results and modification suggestions. On the one hand, document preparers can conduct self-checks based on the review results and optimize the document content; on the other hand, document approvers can quickly locate relevant content in the document based on the review results, improving approval efficiency and quality.

[0080] This invention also provides an intelligent document editing and review method, which will be described below.

[0081] like Figure 1 As shown, the intelligent document editing and review method of the present invention includes the following steps:

[0082] Execute step S11 to receive the uploaded procurement documents; then execute step S12.

[0083] Execute step S12 to vectorize the received procurement documents to obtain basic vectorized data; then execute step S13.

[0084] Execute step S13 to receive the selected file template; then execute step S14.

[0085] Execute step S14 to create a new procurement document based on the selected document template; then execute step S15.

[0086] Execute step S15, retrieve relevant data from the basic vectorized data according to the target information required in the chapters of the newly created procurement document, perform structured transformation on the retrieved relevant data and add it to the corresponding chapters to complete the compilation of the procurement document; then execute step S16.

[0087] Execute step S16: After the procurement document is prepared, preprocess the prepared procurement document to obtain vectorized data of the procurement document; then execute step S17.

[0088] In step S17, after receiving the intelligent review instruction, the review requirements are read from the preset document review point library. Based on the read review requirements, relevant review content is extracted from the vectorized data of the procurement document. The large language model is used to review and judge the relevant review content according to the review requirements, and the review judgment result is output, thereby completing the intelligent review of the procurement document.

[0089] In one specific embodiment of the present invention, the vectorization process of the received procurement documents to obtain basic vectorized data includes the following steps:

[0090] The received procurement documents are divided into sections according to the first-level directory.

[0091] Read the chapter content page by page from the segmented file;

[0092] A large language model is used to extract summaries from the read chapter content;

[0093] The chapter content summaries are vectorized and stored in the database, and the chapter content is placed into metadata and vectorized and stored in the database.

[0094] In one specific embodiment of the present invention, retrieving relevant data from basic vectorized data includes the following steps:

[0095] Retrieve relevant sections from the basic vectorized data based on the sections of the procurement documents;

[0096] The relevant chapters in the recall were ranked using a re-ranking model;

[0097] The relevant chapters are retrieved sequentially, and the retrieval process is terminated once relevant information is found. The retrieved information is then used as the relevant data for the recall.

[0098] In one specific embodiment of the present invention, the structured transformation of recall-related data includes the following steps:

[0099] The relevant recall data is structured according to the information type. If the information type is a numerical type, it is formatted according to the requirements of the numerical type to complete the structured conversion of the numerical type.

[0100] If the information type is date, then format it according to the requirements of the date type to complete the structured conversion of the date type;

[0101] If the information type is a data dictionary type, then the large language model is used to match the format and return the corresponding data dictionary information to complete the structured transformation of the data dictionary type.

[0102] In one specific embodiment of the present invention, when preparing procurement documents, semantic similarity retrieval is performed in a vectorized database using keywords related to the bid evaluation method;

[0103] Based on the search results, retrieve chapters and fragments related to the bid evaluation method, and then piece together the retrieved chapters and fragments to form the bid evaluation content paragraphs;

[0104] Using a large language model, generate clause node names, clause names, clause descriptions, and clause scores based on the paragraphs of the evaluation content, and output them in Markdown table format;

[0105] The Markdown tables output by the large language model are parsed to form structured evaluation method data, which is then added to the review method section of the procurement document.

[0106] In one specific embodiment of the present invention, the intelligent document review process includes:

[0107] After the document is compiled, the procurement document needs to be preprocessed and stored in the vector database. The document is divided into chapters and directories, and the abstract is extracted and stored in the vector database. This process is similar to the preprocessing process of document field content recognition.

[0108] Then, we will analyze each review rule in the document review point library. Based on the description information of the review point, we will understand the review target content and the specific requirements of that review item for that content.

[0109] Then, relevant information about the review target content will be extracted from the documents to be reviewed, a process similar to document content recognition.

[0110] Finally, the extracted review targets and requirements are fed into the large language model. The natural language understanding and reasoning capabilities of the large language model are used to determine whether the target content meets the review requirements, provide reasons and evidence, and offer suggestions for modification.

[0111] The present invention also provides a storage medium storing a program for an intelligent document editing and review method, wherein the steps of the intelligent document editing and review method are implemented when the program for the intelligent document editing and review method is executed by a processor.

[0112] The intelligent document editing and review system and method of this invention can be applied to the bidding document preparation stage of an electronic procurement system, primarily for use by procurement document preparers and reviewers. During document preparation, the intelligent document preparation function allows document preparers to directly upload offline procurement documents of any format. The system automatically extracts structured data from these documents and automatically populates them into the information system, following the standard document format template. After document preparation is complete, the system automatically extracts and reviews key content based on the review items configured in the document review points, providing feedback on the review results to the document preparer for reference and modification. Simultaneously, the document review results can also be provided to the document approver for reference.

[0113] This invention leverages large language model technology to automatically extract key elements required for system input from offline-edited procurement documents. Users simply upload their offline-prepared documents to the procurement system, which automatically extracts the required variable attributes for the corresponding chapter templates. The intelligently identified core chapters include the cover, announcements, bidder instructions, review methods, and contract terms. At the document compliance review level, the powerful semantic recognition and reasoning capabilities of the large language model can also retrieve content related to the review items from the document and provide review and modification suggestions for the compilers and reviewers, improving efficiency while reducing the risk of omissions.

[0114] The present invention has been described in detail above with reference to the accompanying drawings and embodiments. Those skilled in the art can make various modifications to the present invention based on the above description. Therefore, certain details in the embodiments should not be construed as limiting the present invention, and the scope of protection of the present invention shall be defined by the appended claims.

Claims

1. An intelligent document editing and review method, characterized in that, Includes the following steps: Receive uploaded procurement documents; The received procurement documents are vectorized to obtain basic vectorized data; Receive the selected file template; Create a new procurement document based on the selected document template; Retrieve relevant data from the basic vectorized data according to the target information required in the chapters of the newly created procurement document, perform structured transformation on the retrieved relevant data and add it to the corresponding chapters to complete the compilation of the procurement document; After the procurement documents are prepared, they are preprocessed to obtain vectorized data of the procurement documents. Upon receiving the intelligent review instruction, the system reads the review requirements from the preset document review point library, extracts relevant review content from the vectorized data of the procurement document based on the read review requirements, uses a large language model to review and judge the relevant review content according to the review requirements, and outputs the review judgment result, thereby completing the intelligent review of the procurement document.

2. The intelligent document editing and reviewing method as described in claim 1, characterized in that, The process of vectorizing the received procurement documents to obtain basic vectorized data includes the following steps: The received procurement documents are divided into sections according to the first-level directory. Read the chapter content page by page from the segmented file; A large language model is used to extract summaries from the read chapter content; The chapter content summaries are vectorized and stored in the database, and the chapter content is placed into metadata and vectorized and stored in the database.

3. The intelligent document editing and reviewing method as described in claim 1, characterized in that, Retrieving relevant data from basic vectorized data involves the following steps: Retrieve relevant sections from the underlying vectorized data based on the sections of the procurement documents; The relevant chapters in the recall were ranked using a re-ranking model; The relevant chapters are retrieved sequentially, and the retrieval process is terminated once relevant information is found. The retrieved information is then used as the relevant data for the recall.

4. The intelligent document editing and reviewing method as described in claim 1, characterized in that, The structured transformation of recall-related data includes the following steps: The relevant recall data is structured according to the information type. If the information type is a numerical type, it is formatted according to the requirements of the numerical type to complete the structured conversion of the numerical type. If the information type is date, then format it according to the requirements of the date type to complete the structured conversion of the date type; If the information type is a data dictionary type, then the large language model is used to match the format and return the corresponding data dictionary information to complete the structured transformation of the data dictionary type.

5. The intelligent document editing and reviewing method as described in claim 1, characterized in that, When preparing procurement documents, semantic similarity searches are performed in a vectorized database using keywords related to the bid evaluation methods. Based on the search results, retrieve chapters and fragments related to the bid evaluation method, and then piece together the retrieved chapters and fragments to form the bid evaluation content paragraphs; Using a large language model, generate clause node names, clause names, clause descriptions, and clause scores based on the paragraphs of the evaluation content, and output them in Markdown table format; The Markdown tables output by the large language model are parsed to form structured evaluation method data, which is then added to the review method section of the procurement document.

6. A storage medium, characterized in that, The storage medium stores a program for an intelligent document editing and review method. When the program for the time-based intelligent document editing and review method is executed by a processor, it implements the steps of the intelligent document editing and review method as described in any one of claims 1 to 5.

7. An intelligent document editing and review system, characterized in that, include: The receiving unit is used to receive uploaded procurement documents, selected document templates, and intelligent review instructions; A preprocessing unit, connected to the receiving unit, is used to perform vectorization processing on the received procurement documents to obtain basic vectorized data; The document compilation unit, connected to the receiving unit and the preprocessing unit, is used to create a new procurement document based on the selected document template, and is also used to retrieve relevant data from the basic vectorized data according to the target information required by the chapters in the newly created procurement document, perform structured transformation on the retrieved relevant data and add it to the corresponding chapters to complete the compilation of the procurement document; After the procurement document is prepared, the preprocessing unit is also used to preprocess the prepared procurement document to obtain vectorized data of the procurement document. The document review unit, connected to the preprocessing unit, is used to read review requirements from a preset document review point library after receiving an intelligent review instruction, extract relevant review content from the vectorized data of the procurement document based on the read review requirements, use a large language model to review and judge the relevant review content according to the review requirements, and output the review judgment result, thereby completing the intelligent review of the procurement document.

8. The intelligent document editing and reviewing system as described in claim 7, characterized in that, The preprocessing unit is used to split the received procurement documents according to the first-level directory. Read the chapter content page by page from the segmented file; A large language model is used to extract summaries from the read chapter content; The chapter content summaries are vectorized and stored in the database, and the chapter content is placed into metadata and vectorized and stored in the database.

9. The intelligent document editing and reviewing system as described in claim 7, characterized in that, The document compilation unit is also used to recall relevant chapters from the basic vectorized data based on the chapters of the procurement document; The relevant chapters in the recall were ranked using a re-ranking model; The relevant chapters are retrieved sequentially, and the retrieval process is terminated once relevant information is found. The retrieved information is then used as the relevant data for the recall.

10. The intelligent document editing and reviewing system as described in claim 7, characterized in that, The document compilation unit is also used to perform semantic similarity retrieval in a vectorized database using keywords for the evaluation method; Based on the search results, retrieve chapters and fragments related to the bid evaluation method, and then piece together the retrieved chapters and fragments to form the bid evaluation content paragraphs; Using a large language model, generate clause node names, clause names, clause descriptions, and clause scores based on the paragraphs of the evaluation content, and output them in Markdown table format; The Markdown tables output by the large language model are parsed to form structured evaluation method data, which is then added to the review method section of the procurement document.

Images