Paper archive automatic inputting and archiving method and system based on artificial intelligence

Abstract

The invention provides an artificial intelligence-based paper archive automatic inputting and archiving method and system. The method comprises the following steps of: 1, digitally converting a paper archive into a digital archive; 2, comparing the digital archive with an in-library template; 3, when the digital archives have templates in the library, extracting, inputting and indexing, and classifying the archives; 4, when the digital archives do not have templates in the library, determining the position content of the to-be-input elements of the archives according to the archives of the same category; 5, if the same class of archives do not exist in the step 4, related archive elements are input and added into a to-be-learned template library, and the like. Template generation, template recognition, annual archive catalogue rule learning and the like are combined, full-process automation of paper archive data recognition, data input and classified catalogue is achieved under the condition of no manual intervention, and the efficiency is improved. Therefore, the labor loss and the error rate of data sorting are reduced, and the working efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of information management and archive management, and in particular to an artificial intelligence-based automatic paper archive entry and filing method and system. Background Art

[0002] The advent of the digital age has brought a new and convenient experience to emerging technologies such as information technology and artificial intelligence in promoting archive management. In order to ensure the convenience, flexibility and comprehensiveness of archival resources, a large amount of archival data has been converted from paper resources to digital resources. As archival data continues to grow. In the existing technology:

[0003] Publication No. "CN116627914A" discloses a comprehensive archive management system and management method, which proposes a comprehensive archive management system and management method, and specifically discloses the technical idea of ​​using templates for archive entry. This technical solution realizes the function of intelligently optimizing archive entry management, and can determine the matching situation between the archive module and the submitted archive and make appropriate adjustments. However, it does not consider how to classify, catalog and archive the archives that meet the template through technical means, and cannot support effective archiving.

[0004] Publication No. "CN117058704A" discloses a method and device for extracting textbook content and structure based on visual and text features. It discloses a method and device for extracting textbook content and structure based on both visual and text features, and specifically discloses a technical idea for identifying image memory blocks. This technology clearly proposes a text recognition and extraction method for PDF files and the corresponding structural relationship to facilitate the establishment of a subject knowledge graph. The technology proposes a combination of text recognition and visual threshold correction recognition for PDF files, which is beneficial for the device to identify and judge the file content, but it is still far from being able to identify and judge the ten common data in archives, nor can it automatically set templates according to the recognition situation, and it is even more impossible to complete template-related data entry and automatic archiving of archives through recognition.

[0005] Publication number "CN115328854A", an artificial intelligence-based archive management system, discloses a method for realizing archival file information repair and archive classification through artificial intelligence methods, and clarifies how to realize the comparison and repair of manually entered data and file information in the case of manually entered data. This technology clearly defines a method for realizing archival entry data repair by extracting file information through archival analysis and comparing it with the data entered in the same direction. The technology gives a detailed method of how to confirm and compare the data extracted through archival analysis with the data entered in the same direction, and realizes archival data analysis and repair through artificial intelligence technology. However, this method must first be based on the situation where the archival information has already been entered, and it is impossible to realize the self-entry of archival data through technical means.

[0006] In summary, the existing archive classification management methods cannot meet the problems of archive data redundancy, large output, time-consuming and labor-intensive manual input, and low classification accuracy. Compared with electronic archive sorting, paper archives converted into digital archives are in picture format, which cannot be directly recognized by text, and it is even more difficult to extract archive information, input archive elements, and complete cataloging and archiving. This limits the convenience and accuracy of archive work. Summary of the invention

[0007] In view of this, the present invention starts with text recognition technology, relies on artificial intelligence, and through technical analysis and template learning, explores a paper archive entry method that automatically identifies, automatically enters, and automatically archives, and realizes an artificial intelligence-based paper archive automatic entry and archiving method and system that fully automates the archive collection and archiving process.

[0008] The specific technical solutions are as follows:

[0009] The method for automatic entry and filing of paper archives based on artificial intelligence is characterized by comprising the following steps:

[0010] Step 1: Convert paper archives into digital archives through digitization;

[0011] Step 2: Compare the digital file with the template in the library;

[0012] Step 3: When there is a template for the digital archive in the library, extract, enter, index, and classify the archive;

[0013] Step 4: When there is no template for the digital archive in the database, the location and content of the elements to be entered in the archive are determined based on archives of the same category;

[0014] Step 5: If there are no archives of the same category in step 4, enter the relevant archive elements and add them to the template library to be learned. At the same time, enter the completed archives. According to the archive classification catalogs of previous years entered in the library, use artificial intelligence data analysis to complete automatic classification and cataloging. If there are no classification rules for archives of the same category, the administrator will complete the archive classification and the system will learn and add them to the archive classification catalogs of previous years.

[0015] In one embodiment of the present invention, step four includes establishing templates and classification rules that meet the file entry requirements.

[0016] The paper archive automatic entry and filing system based on artificial intelligence is based on the automatic entry and filing method described in claim 1, including template generation, template recognition and annual archive catalog rule learning,

[0017] The template recognition is performed by comparing the uploaded file to be processed with the template for overlap;

[0018] The learning of the rules for the archive catalogs of previous years converts the archive catalogs of previous years into a unified format. In addition, the element data such as year, retention period, category, etc. in the field names in the text correspond to the corresponding cataloging positions.

[0019] In one embodiment of the present invention, the template generation selects a number of sample files for which the template needs to be generated, requires the scanning quality of the sample files to meet the national archive digitization standards, unifies the digitized file format, determines the number of pages of the file according to the file attributes, fills the data into the page number field, and extracts the first and last pages containing the template entry elements;

[0020] In one embodiment of the present invention, the template recognition uploads the file to be processed, and the received file is recognized through template matching. First, the text to be recognized is superimposed with the template in the library to determine whether it overlaps and the similarity of the overlap. Natural language processing technology is used to divide the text into several areas according to the differences in text fonts and paragraph structures, and the text structure, style, keywords, etc. are extracted to construct a feature vector space, compare the file features with the template features, and find similar templates.

[0021] In one embodiment of the present invention, the historical archive catalog rule learning converts the historical archive catalog into a format that is easy to extract text, such as TXT format, ET format, EXCEL format, etc., and then corresponds the element data such as year, retention period, category, etc. in the text field name to the corresponding cataloging position, and clarifies the arrangement order corresponding to each different classification.

[0022] The above technical solution has the following beneficial effects:

[0023] The present invention combines template generation, template recognition and learning of historical archive catalog rules, etc., to achieve full process automation of paper archive data recognition, data entry, and classification cataloging without human intervention, thereby reducing labor loss and data sorting error rate and improving work efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a timing diagram of automatic recording and sorting of archives in the method and system for automatic entry and filing of paper archives based on artificial intelligence of the present invention. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] Example 1: See Figure 1 As shown, the method for automatic entry and filing of paper archives based on artificial intelligence includes the following steps:

[0027] Step 1: Convert paper archives into digital archives through digitization;

[0028] Step 2: Compare the digital file with the template in the library;

[0029] Step 3: When there is a template for the digital archive in the library, extract, enter, index, and classify the archive;

[0030] Step 4: When there is no template for the digital archive in the database, the location and content of the elements to be entered in the archive are determined based on archives of the same category;

[0031] Step 5: If there are no archives of the same category in step 4, enter the relevant archive elements and add them to the template library to be learned. At the same time, enter the completed archives and complete automatic classification and cataloging according to the archive classification directory of previous years entered in the library. If there are no classification rules for archives of the same category, the administrator will complete the archive classification and the system will learn and add them to the archive classification directory of previous years.

[0032] The details are as follows: collect paper archives, count them and complete the scanning process, and the paper archives will be digitally converted. The digital archives will be stored in the pre-processing library of the archive processing system. When the archive administrator initiates the automatic arrangement of archives, the system starts the archive recognition and capture module and starts data recognition; compare the template in the library according to the recognition and results. If the archive is consistent with a template in the library, the archive elements are extracted, entered and indexed according to the template settings, and the archives are classified according to the system settings of the template. The sequential cataloging is completed according to the cataloging situation and filing rules of the archives in the library. If there is no template for the archive in the library, the archive administrator is prompted to provide several archives of the same category, and the location content of the archive elements to be entered is determined through the artificial intelligence algorithm, and a set of templates and classification rules that meet the requirements of archive entry are established to facilitate automatic entry and cataloging of archives. If the archive administrator does not have archives of the same category, the administrator enters the relevant archive elements and adds them to the template library to be learned, so that similar archives can be found for learning later. At the same time, the completed archives are entered, and automatic classification and cataloging are completed according to the archive classification catalogs of previous years entered in the library using artificial intelligence data analysis. If there are no classification rules for archives of the same category, the administrator will complete the archive classification and the system will learn and add it to the archive classification directory of previous years.

[0033] The paper archive automatic entry and filing system based on artificial intelligence is based on the above-mentioned automatic entry and filing method, including template generation, template recognition and learning of archive catalog rules of previous years.

[0034] In this embodiment, the template generation selects a number of sample archives for which templates need to be generated, requires that the scanning quality of the sample archives meet the national archive digitization standards, unifies the digitized file format (such as PDF format, OFD format, etc.), determines the number of pages of the archive according to the file attributes, fills the data into the page number field, and extracts the first and last pages containing the template entry elements;

[0035] In this embodiment, the template recognition uploads the file to be processed, and the received file is recognized through template matching. First, the text to be recognized is superimposed with the template in the library to determine whether it overlaps and the similarity of the overlap. Natural language processing technology is used to divide the text into several areas according to the differences in text fonts and paragraph structures, and the text structure, style, keywords, etc. are extracted to construct a feature vector space, compare the file features with the template features, and find similar templates.

[0036] In this embodiment, the historical archive directory rule learning converts the historical archive directory into a uniform format such as TXT format, ET format, EXCEL format, etc. that is easy to extract text, and then corresponds the element data such as year, retention period, category, etc. in the text field name to the corresponding cataloging position, and clarifies the arrangement order corresponding to each different classification.

[0037] Example 2, based on Example 1, this example selects several sample archives for which templates need to be generated, requires that the scanning quality of the sample archives meet the national archive digitization standards, unifies the digitized file format (such as PDF format, OFD format, etc.), determines the number of pages of the archive according to the file attributes, and fills this data into the page number field. Extract the first and last pages containing template input elements, and mark fixed reference points and content areas that need to be identified on the sample by dragging the selection tool. The reference points are used to decompose and determine the coordinates of the characters in the sample. Each character coordinate corresponds to the corresponding input element. Rely on visual features for information extraction. Use convolutional neural network technology to design different convolution kernels corresponding to different window sizes (n-tuples) selected by the selection to capture text features of different granularities (such as using convolution kernels of sizes 4, 5, and 6 to extract n-tuple features in the text), extract font and format features such as font size or different font formats, colors, etc., use the activation function SELU to improve model normalization and increase stability, and add a pooling layer to determine the position of the reference point in the layout and the proportional relationship between each position (for example, extracting structured information from tables, specific areas, etc.) to extract the most important features and increase model invariance. Use the fully connected layer classification, compile the model according to the neurons output by the previous calculation results, use the cross entropy loss function to calculate the difference between the output probability distribution and the true distribution, use the back propagation algorithm to calculate the partial derivative of the loss function with respect to the template parameters, and use the gradient descent method to optimize the model parameters. Map the extracted features to the final classification results to form a basic template. Repeatedly provide a number of samples of the same category, and detect the recognition accuracy and stability of the template by uploading actual sample digital files under different angles and lighting conditions. Adjust the field boundaries and optimize the image preprocessing for iterative optimization, use the convolutional neural network technology for deep learning, ensure the similarity of the extracted data content, feature extraction and pattern recognition, and continuously adjust the template configuration according to the test feedback to improve the recognition accuracy and stability. Adjust the reference point and recognition area according to the test results, set the affine transformation parameters, transmission transformation parameters or pyramid parameters to determine the transformation parameters of the proportional recognition area corresponding to the sample change, ensure that the setting of the recognition area is consistent with the layout and content of the actual document, optimize the recognition accuracy, and form a data set template. Provide a user-friendly interface that allows archivists to use supervised learning methods to adjust the parameters of the generated template as needed to further repair the deviation. After completing the test and optimization, save and form a set of formal templates. Optical character recognition (OCR) technology is used to identify and convert archive content into machine-encoded text. Each framed recognition area is configured with a corresponding entry element field name, and the recognized text is set to be filled in the entry element field name. Necessary field names include but are not limited to file number, material name, responsible person, date of writing, volume number, confidentiality level, etc., to complete a complete archive entry template.The generated archive entry template sets the corresponding archive category according to the archive scope of the archive department. If the names of different materials in the same template are classified differently, use pre-trained language models such as BERT and GPT to further enhance the model's ability to understand complex language structures and semantics, add file numbers, responsible persons and other factors to assist in analysis, and increase the classification requirements. Set sorting rules to generate cataloging rules.

[0038] In this embodiment, the archives to be processed are uploaded in template recognition, and the recognition of the received archives is realized by template matching. First, the text to be recognized is superimposed with the template in the library to determine whether it overlaps and the overlap similarity. The text is divided into several areas according to the difference in text font and paragraph structure by natural language processing technology, and the text structure, style, keywords, etc. are extracted to construct a feature vector space, and the file features are compared with the template features to find similar templates. After determining the template to be used, the content area to be entered is selected, and the elements to be entered are extracted and recognized as text, and returned to fill in the archive entry element item to complete the information element entry of this archive. According to the filing category and cataloging rules of this set of templates, the last code (assuming it is xxx) of the archives of this type that have been cataloged in the library in this year and within the current preservation period is found, and the archive is cataloged (entered into xxx+1). After the storage of archives of this category exceeds n (n value can be customized, for example, set to 100) pieces each year, the new sample in the library is called, the template generation process is repeated, and the template is reviewed and updated to ensure recognition accuracy and relevance.

[0039] In the study of the rules of the archive catalogues of the past years, the archive catalogues of the past years are uniformly converted into TXT format, ET format, EXCEL format and other formats that are easy to extract text, and the year, storage period, category and other element data in the field name of the text are matched to the corresponding catalogue position, and the arrangement order corresponding to each different classification is clarified. The recurrent neural network technology is used for feature extraction and pattern recognition, and the pre-trained language models such as BERT and GPT are used to semantically understand the file name content of the entered archives to distinguish the corresponding categories. The information is extracted, parsed, and text classified to extract the summary, and the natural language processing (NLP) technology such as sentiment analysis is completed through semantic understanding to achieve automatic classification. The extracted information is checked by rules to ensure the correctness of the format and content of the data, so as to achieve rapid classification and labeling. The archive catalogue rules are formed. The administrator correction function is provided, and the annual catalogue is completed, and the evaluation and update technical solutions are added to improve the adaptability and accuracy of the model.

[0040] Embodiment 3: Based on the illustration and description of artificial intelligence template learning, any specific value should be interpreted as merely exemplary and not as a limitation. Other computer, information means and methods may be adopted to implement the template recognition method, such as using a scanning pen or other equipment to directly identify the information elements on the paper archive to complete the data entry. The artificial intelligence data analysis method may not use convolutional neural networks, but may also use clustering algorithms, recurrent neural networks, and pooling network algorithms. However, no matter which method is used, the template method for organizing archives must first organize a set of general templates, and set the archive category and the information elements that should be entered for the archive for the template to ensure that the archive information is comprehensive and the data can be archived.

[0041] Based on the illustration and description of the learning of the artificial intelligence classification model, any specific value should be interpreted as merely exemplary and not as a limitation. The artificial intelligence data analysis method may not use convolutional neural networks, recurrent neural networks, or pooling network algorithms, but no matter which method is used, a set of usable classification models should be formed. The model should set up filing categories with clear logic and no execution conflicts and the information elements that should be entered for filing to ensure that the archive information is comprehensive and the data can be filed.

[0042] The recognition of paper archive information is not limited to optical character recognition (OCR) technology. Other technologies such as image recognition technology, deep learning technology, feature matching, multimodal fusion technology, etc. are recognized as long as they can complete the recognition of archive information. The digital archive formats after paper archive conversion include but are not limited to PDF format, OFD format, JPG format, TIFF format, and TXT format.

[0043] The basic principles and main features of the present invention are described above. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for illustrating the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which shall fall within the scope of the present invention to be protected. The scope of protection of the invention shall be defined by the attached claims and their equivalents.

Claims

1. The method for automatic paper archive entry and filing based on artificial intelligence is characterized by: The following steps are involved: Step 1: Convert paper archives into digital archives through digitization; Step 2: Compare the digital file with the template in the library; Step 3: When there is a template for the digital archive in the library, extract, enter, index, and classify the archive; Step 4: When there is no template for the digital archive in the database, the location and content of the elements to be entered in the archive are determined based on archives of the same category; Step 5: If there are no archives of the same category in step 4, enter the relevant archive elements and add them to the template library to be learned. At the same time, enter the completed archives and complete automatic classification and cataloging according to the archive classification directory of previous years entered in the library. If there are no classification rules for archives of the same category, the administrator will complete the archive classification and the system will learn and add them to the archive classification directory of previous years.

2. The method for automatically entering and filing paper archives based on artificial intelligence according to claim 1 is characterized in that: The step four includes establishing templates and classification rules that meet the file entry requirements.

3. An artificial intelligence-based paper archive automatic entry and filing system, based on the automatic entry and filing method of claim 1, including template generation, template recognition and annual archive catalog rule learning, The template recognition is performed by comparing the uploaded file to be processed with the template for overlap; The learning of the rules for the archive catalogs of previous years converts the archive catalogs of previous years into a unified format. In addition, the element data such as year, retention period, category, etc. in the field names in the text correspond to the corresponding cataloging positions.

4. The paper archive automatic entry and filing system based on artificial intelligence according to claim 3 is characterized in that: The template generation selects several sample archives for which templates need to be generated, requires that the scanning quality of the sample archives conform to the national archive digitization standards, unifies the digitized file format, determines the number of pages of the archive according to the file attributes, fills the data into the page number field, and extracts the first and last pages containing the template entry elements.

5. The paper archive automatic entry and filing system based on artificial intelligence according to claim 3 is characterized in that: The template recognition uploads the files to be processed, and the received files are recognized through template matching. First, the text to be recognized is superimposed with the template in the library to determine whether it overlaps and the similarity of the overlap. Natural language processing technology is used to divide the text into several areas according to the differences in text fonts and paragraph structures, and the text structure, style, keywords, etc. are extracted to construct a feature vector space, compare the file features with the template features, and find similar templates.

6. The paper archive automatic entry and filing system based on artificial intelligence according to claim 3 is characterized in that: The historical archive catalog rule learning converts the historical archive catalog into TXT format, ET format, EXCEL format and other formats that are easy to extract text, and then corresponds the element data such as year, retention period, category, etc. in the text field name to the corresponding cataloging position, and clarifies the arrangement order corresponding to each different classification.

Images