Official document element extraction and auditing system based on large model

Abstract

The invention relates to an official document element extraction and auditing system based on a large model. The system comprises an official document element extraction module, a text element extraction module and an element auditing module. The official document element extraction module is used for automatically identifying and extracting key elements of the official document; the text element extraction module is used for analyzing the text of the official document so as to extract ordered and structured information; and the element auditing module is used for auditing the official document elements item by item to ensure the accuracy and completeness of each element. The method has the advantages of automatically and intelligently processing the official document, and improving the efficiency and accuracy of official document element extraction and auditing.

Description

Technical Field

[0001] The present invention relates to the field of information processing technology, and in particular to a document element extraction and review system based on a large model. Background Art

[0002] In government affairs and corporate management, document processing is a basic and critical task, which involves the transmission of information, the issuance of policies, and the recording of decisions. Traditional document processing methods mainly rely on manual operations, including reading, understanding, and extracting key information from documents, as well as subsequent review processes. These processes are not only time-consuming and labor-intensive, but also prone to errors and omissions due to human factors.

[0003] With the acceleration of informatization, the number and complexity of official documents are also increasing, which brings greater challenges to document processing; the existing official document processing technology has achieved automation to a certain extent, and the application of model technology in the field of natural language processing has made significant progress, providing possible solutions for the automated processing of official documents; however, the existing official document structure and content are often complex and diverse, and there is still room for improvement in processing efficiency and accuracy. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the technical problem to be solved by this patent application is how to provide an automated and intelligent document processing system to improve the efficiency and accuracy of document element extraction and review.

[0005] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0006] A document element extraction and review system based on a large model includes a document element extraction module, a text element extraction module and an element review module; the document element extraction module is used to automatically identify and extract the key elements of the document; the text element extraction module is used to parse the document body to extract orderly and structured information; the element review module is used to review the document elements item by item to ensure the accuracy and completeness of each element.

[0007] As an optimization, the official document element extraction module is the part of the system used to automatically identify and extract key elements of official documents. The official document element extraction module includes the steps of data preprocessing, named entity recognition, element extraction and verification, output and optimization.

[0008] As an optimization, data preprocessing first collects official document texts in various styles, then unifies the format and cleans the data, and divides the entire official document text into independent sentences or paragraphs; named entity recognition first defines the entity category, then extracts text features such as word features, character features, position features, and part-of-speech features, trains the NER model, and then uses the NER model to identify each entity and its category in the official document text; feature extraction and verification extracts each element in the official document based on the output of the NER model, performs rule verification, and uses context information to verify the rationality of the extracted entities; the output and optimization step first outputs the extracted elements and their categories in the form of structured data, and then provides a user interface for users to provide feedback on the extraction results. The system adjusts the NER model based on the feedback, and then fine-tunes it based on the pre-trained large model through transfer learning, performs entity disambiguation through context information and entity relationship networks, and finally combines multiple different NER models through weighted averaging.

[0009] As an optimization, the body element extraction module parses the body of the official document to extract orderly and structured information, including the body subheadings, the content corresponding to the body subheadings, the attachment titles and the attachment content, including the steps of text segmentation and identification, extraction of the content corresponding to the body subheadings, extraction of the attachment titles and content, and special format processing.

[0010] As an optimization, text segmentation and identification first divides the main body of the document into independent paragraphs through regular expressions. The system uses contextual information to understand the content under each subheading, and then identifies the subheading of each paragraph through the large model; when extracting the corresponding content of the main text subheading, the system locates the content under each subheading, extracts the main text content corresponding to the subheading, and associates it with the subheading to form structured data; when extracting the attachment title and content, the system uses keyword search to identify the attachment or match the content of the adjacent position of the attachment, and locates the starting position of the attachment, and then extracts the attachment title. When extracting the attachment content, for text-type attachments, the system will extract all text content after the attachment title; for non-text-type attachments, the system will extract the attachment file name information; finally, special formats are processed. For tables in the main text or attachments, they are first processed into Markdown format, and then the table structure is identified through the large model, and the information in the table is extracted; for attachments containing pictures, the system will extract the descriptive text or picture file name of the picture.

[0011] As an optimization, the element review module uses a big model combined with a rule engine to review the elements of official documents item by item to ensure the accuracy and completeness of each element, including steps such as title review, content review, and other element review.

[0012] As an optimization, the title review first establishes the judgment criteria for the correct title based on the common errors in the official document titles, and then uses the online model to generate negative data for the official document titles in accordance with the standards to ensure that all possible errors are covered, check the fine-tuning data, and perform LoRA fine-tuning. After the fine-tuning is completed, the LM-Cocktail tool is used to merge the fine-tuned model and the basic model. The fine-tuned merged model is then used to review the common errors in the official document titles. Finally, based on the semantic capabilities of the large model, the theme consistency of each title and the corresponding content is reviewed. For the subheadings of the text, the semantic relationship between the subheadings is also reviewed. The content review first reviews the grammar, semantics, logic, and standard usage of the official document body and attachments to establish the correct title. Confirm the standard, and then filter or call the online model from the historical document database to generate representative positive and negative examples. The online model will mark the errors in the data set, check, correct and score the data, and then fine-tune and merge the model. Use the trained grammar review model to review the content of the document body and attachments; the review of the other elements uses regular expression matching, combined with the issuing agency information, to verify the correctness of the document number; verify the accuracy of the name of the main sending agency by matching the agency name, agency responsibilities, and agency level information in the knowledge base; verify the accuracy of the issuing agency name by matching the agency name in the knowledge base; use the date parsing library to check whether the issuing time is in logical order and whether the time format is correct.

[0013] The present invention has the following advantages:

[0014] (1) Improving the efficiency of document processing: The method and system can quickly extract key elements from documents and review them, which can shorten the time of document processing and improve office efficiency.

[0015] (2) Ensure the accuracy of official documents: Through the deep understanding and intelligent review of the big model, this method and system effectively ensure the accuracy of official document elements and reduce the probability of errors in official documents.

[0016] (3) Reduce manual burden: Automated extraction and review of official document elements reduces manual input in tedious review work, allowing staff to devote more energy to other tasks.

[0017] (4) Improving the level of automation in document processing: The implementation of this patent is conducive to the development of document processing towards automation and intelligence, and is also conducive to improving the modernization level of the entire document processing process.

[0018] (5) Strong applicability: The system can conduct differentiated reviews on different types of official document elements, adapting to diverse official document processing needs and having wide applicability.

[0019] (6) Promote standardized management: Through unified review standards and methods, this patent helps to achieve standardized management of official document processing and improve the quality and standardization of official documents. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a framework diagram of a document element extraction and review system based on a large model described in the present invention. DETAILED DESCRIPTION

[0021] The present invention is further described in detail below in conjunction with the accompanying drawings. In the description of the present invention, it should be understood that the directions or positional relationships indicated by directional words such as "upper, lower" and "top, bottom" are usually based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description. Unless otherwise stated, these directional words do not indicate or imply that the devices or components referred to must have a specific direction or be constructed and operated in a specific direction, and therefore cannot be understood as limiting the scope of protection of the present invention; the directional words "inside and outside" refer to the inside and outside relative to the contours of each component itself.

[0022] like Figure 1 As shown, a document element extraction and review system based on a large model includes a document element extraction module, a text element extraction module and an element review module; the document element extraction module is used to automatically identify and extract the key elements of the document; the text element extraction module is used to parse the document body to extract orderly and structured information; the element review module is used to review the document elements item by item to ensure the accuracy and completeness of each element.

[0023] In this embodiment, the official document element extraction module is the part of the system used to automatically identify and extract key elements of official documents. The official document element extraction module includes the steps of data preprocessing, named entity recognition, element extraction and verification, output and optimization.

[0024] In this embodiment, data preprocessing first collects official document texts of various styles, then unifies the format and cleans the data, and divides the entire official document text into independent sentences or paragraphs; named entity recognition first defines the entity category, then extracts text features such as word features, character features, position features, and part-of-speech features, trains the NER model, and then identifies each entity and its category in the official document text through the NER model; element extraction and verification extracts each element in the official document according to the output of the NER model, performs rule verification, and uses context information to verify the rationality of the extracted entities; the output and optimization step first outputs the extracted elements and their categories in the form of structured data, and then provides a user interface for users to provide feedback on the extraction results. The system adjusts the NER model based on the feedback, and then fine-tunes it based on the pre-trained large model through transfer learning, performs entity disambiguation through context information and entity relationship networks, and finally combines multiple different NER models through weighted averaging.

[0025] Specifically, during data preprocessing, official document collection is to collect official document texts in different formats and different types of documents, with as diverse styles as possible; format unification is to convert official document texts in different formats into a unified text format for subsequent processing; official document text cleaning is to remove special characters, blanks, and redundant spaces in the official document text to ensure the neatness of the document content; sentence processing is to divide the entire official document text into independent sentences or paragraphs to provide basic units for NER.

[0026] Specifically, in named entity recognition, the entity category definition is to define the entity category according to the characteristics of the official document, that is, the various elements of the official document, including the title, document number, main addressee, issuing agency, issuing time, and attachment title; feature engineering is to extract the features of the text, including word features, character features, position features, and part-of-speech features to provide input for the NER model; model training is to use annotated official document datasets to train the NER model, and BiLSTM-CRF and BERT models can be used; entity recognition is to input the preprocessed official document text into the trained NER model to identify the various entities and their categories in the text.

[0027] Specifically, in feature extraction and verification, feature extraction is to extract the various elements in the official document based on the output of the NER model; rule verification is to perform rule verification on the extracted elements. For example, the document number usually contains a specific format, which can be verified by regular expressions; context verification is to use context information to verify the rationality of the extracted entities.

[0028] Specifically, in output and optimization, structured output is to output the extracted elements and their categories in the form of structured data, such as JSON or XML format; error feedback is to provide a user interface to allow users to provide feedback on the extraction results, and the system adjusts the NER model based on the feedback to continuously improve the recognition accuracy; model optimization is to use transfer learning technology to fine-tune the pre-trained large model to adapt to the specific language characteristics of the official document field; entity disambiguation is to disambiguate potentially ambiguous entities through contextual information and entity relationship networks; multi-model fusion is to combine multiple NER models with different structures to improve the accuracy of entity recognition through weighted averaging.

[0029] In this embodiment, the body element extraction module parses the body of the official document to extract ordered and structured information, including body subheadings, content corresponding to the body subheadings, attachment titles, and attachment content, including the steps of text segmentation and identification, extraction of content corresponding to the body subheadings, extraction of attachment titles and content, and special format processing.

[0030] In this embodiment, text segmentation and identification first divides the main body of the document into independent paragraphs through regular expressions, and the system uses context information to understand the content under each subheading, and then identifies the subheading of each paragraph through the large model; when extracting the corresponding content of the main text subheading, the system locates the content under each subheading, extracts the main text content corresponding to the subheading, and associates it with the subheading to form structured data; when extracting the attachment title and content, the system identifies the attachment or matches the content of the adjacent position of the attachment through keyword search, and locates the starting position of the attachment, and then extracts the attachment title. When extracting the attachment content, for text-type attachments, the system will extract all text content after the attachment title; for non-text-type attachments, the system will extract the file name information of the attachment; finally, special formats are processed. For tables in the main text or attachments, they are first processed into Markdown format, and then the table structure is identified through the large model, and the information in the table is extracted; for attachments containing pictures, the system will extract the descriptive text or picture file name of the picture.

[0031] In this embodiment, the element review module reviews the elements of the official document item by item through a large model combined with a rule engine to ensure the accuracy and completeness of each element, including the steps of title review, content review, and other element review.

[0032] In this embodiment, the title review first establishes the judgment standard of the correct title according to the common errors in the official document titles, and then uses the online model to generate negative data of the official document titles according to the standard to ensure that all possible errors are covered, check the fine-tuning data, and perform LoRA fine-tuning. After the fine-tuning is completed, the LM-Cocktail tool is used to merge the fine-tuned model and the basic model, and then the common errors in the official document titles are reviewed through the fine-tuned merged model. Finally, according to the semantic capabilities of the large model, the theme consistency of each title and the corresponding content is reviewed. For the subheadings of the text, the semantic relationship between the subheadings is also reviewed; the content review first reviews the grammar, semantics, logic, and standard usage of the official document body and attachments to establish the correct title. Confirm the standard, and then filter or call the online model from the historical document database to generate representative positive and negative examples. The online model will mark the errors in the data set, check, correct and score the data, and then fine-tune and merge the model. Use the trained grammar review model to review the content of the document body and attachments; the review of the other elements uses regular expression matching, combined with the issuing agency information, to verify the correctness of the document number; verify the accuracy of the name of the main sending agency by matching the agency name, agency responsibilities, and agency level information in the knowledge base; verify the accuracy of the issuing agency name by matching the agency name in the knowledge base; use the date parsing library to check whether the issuing time is in logical order and whether the time format is correct.

[0033] This patent aims to deeply understand and accurately extract key information in official documents, including titles, issuing units, etc., as well as subheadings and content of the text, so as to reduce manual operations and reduce error rates. In addition, the purpose of this patent is to ensure the integrity and accuracy of extracted elements through the intelligent review function of large models, reduce the workload of manual review, and improve the level of automation of official document processing. In short, this patent aims to provide an efficient and reliable official document element extraction and review system to improve the office efficiency of government affairs and enterprises, reduce human resource costs, and improve the quality of official document processing and user satisfaction, and promote the modernization and intelligence of official document processing processes.

[0034] Finally, it should be noted that those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A document element extraction and review system based on a large model, characterized in that: It includes official document element extraction module, text element extraction module and element review module; The document element extraction module is used to automatically identify and extract the key elements of the document; The text element extraction module is used to parse the text of the official document to extract orderly and structured information; The element review module is used to review the elements of the official document item by item to ensure the accuracy and completeness of each element.

2. According to the large model-based official document element extraction and review system of claim 1, it is characterized in that: The document element extraction module is the part of the system used to automatically identify and extract key elements of documents. The document element extraction module includes the steps of data preprocessing, named entity recognition, element extraction and verification, output and optimization.

3. According to the large model-based official document element extraction and review system of claim 2, it is characterized in that: Data preprocessing first collects official documents in various styles, then unifies the format and cleans the data, and divides the entire official document text into independent sentences or paragraphs; Named entity recognition first defines the entity category, then extracts the text features of word features, character features, position features, and part-of-speech features, trains the NER model, and then uses the NER model to identify the entities and their categories in the official document text; Element extraction and verification: Based on the output of the NER model, extract the elements in the official document, perform rule verification, and use context information to verify the rationality of the extracted entities; The output and optimization step first outputs the extracted elements and their categories in the form of structured data, and then provides a user interface for users to provide feedback on the extraction results. The system adjusts the NER model based on the feedback, and then fine-tunes the pre-trained large model through transfer learning, performs entity disambiguation through contextual information and entity relationship networks, and finally combines multiple different NER models through weighted averaging.

4. According to the large model-based official document element extraction and review system of claim 3, it is characterized in that: The text element extraction module parses the text of the official document to extract orderly and structured information, including the text subheadings, the content corresponding to the text subheadings, the attachment titles and the attachment content, including the steps of text segmentation and identification, extraction of the content corresponding to the text subheadings, extraction of the attachment titles and content, and special format processing.

5. According to the large model-based official document element extraction and review system of claim 4, it is characterized in that: Text segmentation and identification: Regular expressions are used to segment the main body of the document into independent paragraphs. The system uses context information to understand the content under each subheading, and then uses the large model to identify the subheading of each paragraph. When extracting the content corresponding to the subheadings of the text, the system locates the content under each subheading, extracts the text content corresponding to the subheading, and associates it with the subheading to form structured data; When extracting attachment titles and content, the system uses keyword searches to identify attachments or match the content adjacent to the attachment, locates the beginning of the attachment, and then extracts the attachment title. When extracting attachment content, for text-type attachments, the system will extract all text content after the attachment title; for non-text-type attachments, the system will extract the attachment file name information. Finally, special formats are processed. For tables in the text or attachments, they are first processed into Markdown format, and then the table structure is recognized through the large model, and the information in the table is extracted; for attachments containing pictures, the system will extract the descriptive text or picture file name of the picture.

6. A document element extraction and review system based on a large model according to claim 5, characterized in that: The element review module uses a large model combined with a rule engine to review the elements of official documents item by item to ensure the accuracy and completeness of each element, including the steps of title review, content review, and other element review.

7. A document element extraction and review system based on a large model according to claim 6, characterized in that: The title review first establishes the judgment standard of the correct title based on the common errors in the official document title, and then uses the online model to generate negative data of the official document title according to the standard to ensure that all possible errors are covered, check the fine-tuning data, and perform LoRA fine-tuning. After the fine-tuning is completed, the LM-Cocktail tool is used to merge the fine-tuned model and the basic model, and then the common errors in the official document title are reviewed through the fine-tuned merged model. Finally, based on the semantic capabilities of the large model, the theme consistency of each title and the corresponding content is reviewed. For the sub-titles of the text, the semantic relationship between the sub-titles is also reviewed; The content review first reviews the grammar, semantics, logic, and standard usage of the official document body and attachments to establish correct standards, and then screens or calls online models from the historical document database to generate representative positive and negative examples. The online model marks the errors in the data set, verifies, corrects, and scores the data, and then fine-tunes and merges the model. The trained grammar review model is used to review the official document body and attachments. The review of the other elements mentioned above utilizes regular expression matching, combined with the issuing agency information, to verify the correctness of the document number; verifies the accuracy of the name of the main sending agency by matching the agency name, agency responsibilities, and agency level information in the knowledge base; verifies the accuracy of the issuing agency name by matching the agency name in the knowledge base; and uses a date parsing library to check whether the issuing time is in logical order and whether the time format is correct.

Images