Archives intelligent cataloging method, device and storage medium

Abstract

The application discloses an archive intelligent cataloging method and device and a storage medium, which are applied to an edge node. The method comprises the following steps: performing hierarchical feature extraction and fusion on pre-collected archive information through a pre-deployed lightweight double model, and generating a cataloging item set with confidence annotation; uploading low-confidence cataloging items in the cataloging item set to a cloud server; acquiring cataloging correction data delivered by the cloud server, wherein the cataloging correction data is generated based on multi-modal comparative analysis of the cloud server on the low-confidence cataloging items; integrating cataloging items which are not marked as low-confidence cataloging items with the cataloging correction data, and generating a standard cataloging file according to a standard cataloging rule. The high-quality generation of the standard cataloging file in a complex scene is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of archival informatization, in particular to an archival intelligent registration method and device and a storage medium. BACKGROUND

[0002] With the acceleration of the process of archiving digitization, intelligent registration technology has gradually become a core means to improve the efficiency of archiving management. The current industry generally uses cloud large models combined with optical character (OCR, Optical Character Recognition) recognition technology to realize automatic extraction of archival information, and relies on preset rule templates to complete the structured processing of registration items. However, in the existing technical architecture, massive archival data needs to be processed by public or enterprise private cloud servers in a centralized manner, which leads to real-time response capability being restricted by network delay, and sensitive archival information has data leakage risks in the transmission process

[0003] At the same time, the traditional OCR recognition technology relies on the accurate matching of rule templates for printed text recognition, and when facing handwritten scrawny traces, blurred scanned image or non-standard format archives, it is easy to miss key fields or misjudge semantics, which seriously restricts the registration accuracy and system generalization ability in complex scenarios, and it is difficult to meet the actual needs of high security, high real-time and multi-modal archival processing. SUMMARY

[0004] The main purpose of the present application is to provide an archival intelligent registration method and device and a storage medium, aiming to solve the technical problems of response delay, privacy leakage risk and poor generalization ability of archival intelligent registration services based on public clouds.

[0005] To achieve the above purpose, the present application provides an archival intelligent registration method applied to an edge node, which comprises:

[0006] The pre-deployed lightweight double model is used to extract and fuse the hierarchical features of the pre-collected archival information, and a registration item set with confidence annotation is generated;

[0007] The low-confidence registration items in the registration item set are uploaded to a cloud server;

[0008] Obtain registration correction data delivered by the cloud server, which is generated based on multi-modal comparative analysis of the cloud server on the low-confidence registration items;

[0009] Integrate the registration items that are not marked as low-confidence registration items with the registration correction data, and generate a standard registration file according to the standard registration rules.

[0010] In an embodiment, the step of performing hierarchical feature extraction and fusion on the pre-collected archive information by the pre-deployed lightweight dual model to generate a set of cataloging items with confidence annotations comprises:

[0011] The lightweight dual model comprises a visual understanding model and a text understanding model;

[0012] Performing three-level feature analysis on the archive information by the visual understanding model to generate a structured visual feature vector;

[0013] Obtaining a text recognition result of the archive information, inputting the text recognition result and the structured visual feature vector into the text understanding model, and generating a set of cataloging items with confidence annotations by the text understanding model.

[0014] In an embodiment, the step of inputting the text recognition result and the structured visual feature vector into the text understanding model to generate a set of cataloging items with confidence annotations by the text understanding model comprises:

[0015] Performing text alignment on the text recognition result and the structured visual feature vector to generate a visually enhanced text, and inputting the visually enhanced text into the text understanding model;

[0016] Performing entity extraction, semantic completion, and confidence calculation on the visually enhanced text by the text understanding model to generate a set of cataloging items with confidence annotations.

[0017] In an embodiment, the step of performing entity extraction, semantic completion, and confidence calculation on the visually enhanced text by the text understanding model to generate a set of cataloging items with confidence annotations comprises:

[0018] Based on the spatial coordinate data in the structured visual feature vector, performing a spatial constraint operation on the visually enhanced text, defining an entity extraction region according to archive format features, and generating a regionalized entity set containing a region positioning accuracy parameter;

[0019] According to the document type identifier of the archive information, loading corresponding cataloging rules from a pre-set archive knowledge graph, performing semantic completion and format standardization operations on the regionalized entity set, and generating standardized cataloging items containing a semantic rationality parameter;

[0020] Based on the region positioning accuracy parameter and the semantic rationality parameter, constructing a multi-modal scoring matrix, performing confidence fusion calculation on each standardized cataloging item, and generating a set of cataloging items with confidence annotations.

[0021] In an embodiment, the archive intelligent cataloging method further comprises:

[0022] uploading the to-be-processed archive information to the cloud server when receiving a task takeover instruction from the cloud server.

[0023] An archive intelligent cataloging method applied to a cloud server, the archive intelligent cataloging method comprising:

[0024] receiving a low-confidence cataloging item delivered by an edge node, the low-confidence cataloging item being generated based on the edge node performing hierarchical feature extraction and fusion on pre-collected archive information by a pre-deployed lightweight double model;

[0025] performing multi-modal comparative analysis on the low-confidence cataloging item to generate cataloging correction data, and delivering the cataloging correction data to the edge node;

[0026] integrating, by the edge node, cataloging items that are not marked as low-confidence cataloging items with the cataloging correction data, and generating a standard cataloging file according to a standard cataloging rule.

[0027] In an embodiment, the step of performing multi-modal comparative analysis on the low-confidence cataloging item to generate cataloging correction data, and delivering the cataloging correction data to the edge node comprises:

[0028] calling a similar archive template library to perform multi-modal data alignment on the low-confidence cataloging item, and extracting a key field difference feature vector;

[0029] performing cataloging rule conflict analysis based on the key field difference feature vector to generate a conflict analysis report containing a difference type identifier;

[0030] performing data correction operations according to the difference type identifier by calling a pre-set cataloging correction strategy library to generate cataloging correction data;

[0031] binding the cataloging correction data with a corresponding edge node identifier, and delivering the cataloging correction data to the edge node through an encrypted channel.

[0032] In an embodiment, the archive intelligent cataloging method further comprises:

[0033] sending a task takeover instruction to the edge node when detecting that any of the edge nodes is invalid;

[0034] obtaining to-be-processed archive information of an edge node;

[0035] performing hierarchical feature extraction and multi-modal comparative analysis on the archive information to generate a complete cataloging item set;

[0036] performing format conversion and data encapsulation on the complete cataloging item set to generate a standard cataloging file.

[0037] In addition, to achieve the above object, the application further provides an archive intelligent cataloging device, which comprises:

[0038] A first generation module 10 is configured to perform hierarchical feature extraction and fusion on the pre-collected archive information by using a pre-deployed lightweight double model, so as to generate a cataloging item set with confidence annotation;

[0039] A transmission module 20 is configured to upload low-confidence cataloging items in the cataloging item set to a cloud server;

[0040] An acquisition module 30 is configured to acquire cataloging correction data delivered by the cloud server, wherein the cataloging correction data is generated based on multi-modal comparative analysis of the cloud server on the low-confidence cataloging items;

[0041] A second generation module 40 is configured to integrate cataloging items that are not marked as low-confidence cataloging items with the cataloging correction data, and generate a standard cataloging file according to a standard cataloging rule.

[0042] In addition, to achieve the above object, the application further provides a terminal device, which comprises a memory, a processor, and an archive intelligent cataloging program stored in the memory and executable on the processor, and the archive intelligent cataloging program realizes the steps of the archive intelligent cataloging method when executed by the processor.

[0043] In addition, to achieve the above object, the application further provides a computer readable storage medium, which stores an archive intelligent cataloging program, and the archive intelligent cataloging program realizes the steps of the archive intelligent cataloging method when executed by a processor.

[0044] The one or more technical solutions provided by the application have at least the following technical effects:

[0045] The application effectively improves the comprehensive performance of the archive intelligent cataloging system through the cooperative processing mechanism of the edge node and the cloud server.

[0046] Specifically, the edge node utilizes a lightweight double model to perform hierarchical feature extraction and fusion on pre-collected archive information, generates a catalog item set with confidence annotation, completes the analysis of most catalog items locally, reduces the amount of data to be uploaded to the cloud, thereby reducing network transmission delay and avoiding global exposure of sensitive archive data; for the edge-end low-confidence catalog items generated in a complex scenario, the cloud server performs multi-modal comparative analysis on the low-confidence catalog items, generates catalog correction data, and issues the catalog correction data to the corresponding edge node; the edge node integrates the catalog items that are not marked as low-confidence catalog items with the catalog correction data, and generates a standard catalog file according to the standard catalog rule.

[0047] The present application dynamically integrates edge-end catalog items and cloud-end correction data, maintains the advantage of local processing efficiency, uses cloud computing power to specifically reinforce the generalization ability of the edge model, balances processing speed, data security and recognition accuracy, and realizes high-quality generation of standard catalog files in complex scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 A flowchart of a first exemplary embodiment of the archive intelligent cataloging method of the present application;

[0049] Figure 2 A flowchart of a second exemplary embodiment of the archive intelligent cataloging method of the present application;

[0050] Figure 3 A flowchart of a third exemplary embodiment of the archive intelligent cataloging method of the present application;

[0051] Figure 4 A flowchart of a fourth exemplary embodiment of the archive intelligent cataloging method of the present application;

[0052] Figure 5 A module structure diagram of the archive intelligent cataloging device of the present application embodiment;

[0053] Figure 6 A device structure diagram of the hardware running environment involved in the archive intelligent cataloging method in the present application embodiment.

[0054] The implementation, functional features and advantages of the present application will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0055] It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0056] The main technical solution of the present application is that the archive intelligent cataloging method of the present application can be applied to an intelligent cataloging system, and the intelligent cataloging system comprises a cloud server and a plurality of edge nodes;

[0057] The edge node performs hierarchical feature extraction and fusion on the pre-collected archive information through a pre-deployed lightweight dual model to generate a catalog item set with confidence annotations; uploads low-confidence catalog items in the catalog item set to a cloud server; obtains catalog correction data delivered by the cloud server, integrates catalog items that are not marked as low-confidence catalog items with the catalog correction data, and generates a standard catalog file according to standard catalog rules.

[0058] The cloud server receives low-confidence catalog items delivered by the edge node, which are generated based on hierarchical feature extraction and fusion of pre-collected archive information by the edge node through a pre-deployed lightweight dual model; performs multi-modal comparative analysis on the low-confidence catalog items to generate catalog correction data, and distributes the catalog correction data to the edge node; and the edge node integrates catalog items that are not marked as low-confidence catalog items with the catalog correction data, and generates a standard catalog file according to standard catalog rules.

[0059] Specifically, the first exemplary embodiment of the archive intelligent cataloging method provided by the present application is an archive intelligent cataloging method applied to an edge node, and the detailed steps of the first exemplary embodiment of the archive intelligent cataloging method provided by the present application are as follows:

[0060] Referring to Figure 1 , Figure 1 The flowchart of the first exemplary embodiment of the archive intelligent cataloging method provided by the present application is shown in FIG. 1. In this embodiment, the archive intelligent cataloging method includes steps S10-S40:

[0061] Step S10, hierarchical feature extraction and fusion of pre-collected archive information through a pre-deployed lightweight dual model to generate a catalog item set with confidence annotations;

[0062] It should be noted that the edge node refers to a local computing node in a distributed system, which is usually located at the edge of the network and is used to process local data and reduce dependence on the central server. The lightweight dual model refers to an optimized model architecture that can include two sub-models working together to efficiently process data. Different large models can be integrated according to the edge node deployment scenario and the type of archives. The archive information refers to the file content that needs to be cataloged, including text, images and other forms. Hierarchical feature extraction refers to feature analysis of data at different levels to obtain more rich information. Fusion refers to integrating features from different sources or different levels to enhance the model's understanding of data. The catalog item set refers to a set of catalog information obtained after processing, and each catalog information includes corresponding confidence annotations. The confidence annotations are used to measure the credibility of the catalog items.

[0063] Specifically, the edge node receives the pre-acquired profile information and then inputs it into the lightweight dual model. The lightweight dual model can include a visual understanding model and a textual understanding model, which can extract, fuse, and analyze the visual features and textual features of the profile information.

[0064] The visual understanding model performs three-level feature analysis on the image data, first extracting the large area features in the image, then extracting the detailed features such as lines and shapes within the area, and finally extracting more detailed features such as textures and color transitions, to generate a structured visual feature vector. The textual understanding model receives the text recognition results and the structured visual feature vector, performs text alignment and other operations, generates visual enhanced text, and then performs entity extraction, semantic completion, and confidence calculation on the visual enhanced text, finally generating a set of cataloging items with confidence annotations.

[0065] Illustratively, there is an edge node that is an intelligent profile processing terminal deployed in the archive room, which receives a scanned contract profile image and the text content identified by OCR technology. The visual understanding model in its lightweight dual model performs three-level feature analysis on the image, the first-level feature is the division of the overall layout area of the contract, the second-level feature is the shape outline of each clause paragraph of the contract, and the third-level feature is the font, font size, and other details of the text in the paragraph. The textual understanding model combines the text content with these visual features to generate visual enhanced text. Then extract the cataloging items such as the names of the two parties to the contract and the signing date from the visual enhanced text, complete the missing contract amount unit according to the context, and calculate the confidence of each cataloging item, generating a set of contract-related cataloging items with confidence annotations, such as "Signing Date: 2024-05-20 (Confidence 0.98)".

[0066] In addition, it should be noted that advanced face recognition algorithms can be integrated into the visual understanding model of the edge node. When pre-acquiring profiles containing personnel image information, the edge node can use the algorithm to extract face feature vectors. When processing personnel profile photo pages and other profile information containing face images, the face can be quickly recognized and associated with the personnel identity information in the profile, providing accurate personnel-related data for cataloging items. In addition, for profiles containing rich image elements, the edge node identifies various objects in the image through the corresponding model, extracts object categories, positions, and other information, and better constructs the structured description of the profile content.

[0067] For example, in the process of recording the personnel files of an enterprise, when processing the onboarding files of employees, the edge node quickly associates the employee photos with the identity information such as the name and ID number in the files through face recognition algorithms, to provide accurate face identity data for the recording items. Meanwhile, the edge node uses the object recognition model to identify the file types in the files, to extract key information such as the certificate category and issuing agency, and enrich the content of the recording item set. In addition, if the edge node integrates the enterprise human resource knowledge large model, it can perform deep semantic analysis on the text information such as the work experience and skill description of the employees, to generate more standardized and detailed semantic recording items, and improve the completeness and professionalism of the personnel file recording.

[0068] Step S20, uploading the low-confidence recording items in the recording item set to the cloud server;

[0069] It should be noted that in the embodiments of the present application, the low-confidence recording items refer to the part of the recording items with confidence lower than the preset threshold in the recording items with confidence annotation generated after the lightweight double model processing. For example, when the confidence is lower than 0.8, it is determined as a low-confidence recording item. The confidence threshold can be flexibly set according to the actual application scenario and the requirement for data accuracy.

[0070] Specifically, after obtaining the recording item set with confidence annotation, the edge node judges the confidence of each recording item. The judgment method can be to compare the confidence of each recording item with the preset threshold one by one. If it is found that the confidence of a certain recording item is lower than the threshold, the low-confidence recording item is sent to the cloud server through the network communication module. The relevant identification information of the edge node needs to be carried during the sending process, so that the cloud server can identify the data from which edge node.

[0071] For example, if the confidence of the recording item “Total contract amount: 100,000 yuan (confidence 0.7)” in the recording item set generated by the edge node is lower than the preset threshold of 0.8. The edge node sends this low-confidence recording item and its own device ID: edge_node_001 to the cloud server through the built-in Wi-Fi module. After receiving, the cloud server can know that this low-confidence recording item is the processing result of a certain contract file from the edge node numbered “edge_node_001”.

[0072] Step S30, obtaining the recording correction data delivered by the cloud server, the recording correction data being generated based on the multi-modal comparative analysis of the cloud server on the low-confidence recording item;

[0073] It should be noted that in the embodiments of the present application, the multi-modal contrast analysis refers to comprehensive analysis of low-confidence catalog items in combination with multiple data types to find potential errors or inaccuracies. The catalog correction data refers to data generated for correcting catalog items according to the results of the contrast analysis.

[0074] Specifically, after the edge node uploads the low-confidence catalog items in the catalog item set to the cloud server, the edge node obtains catalog correction data delivered by the cloud server, the catalog correction data is generated based on multi-modal contrast analysis of the cloud server on the low-confidence catalog items. Specifically, the cloud server will retrieve the corresponding original archive information according to the edge node identifier, and the original archive information can be an archive backup stored in the cloud or associated archive data obtained by other means. Then, the multi-modal contrast analysis method is used to determine the error type and correction direction of the low-confidence catalog item, and the catalog correction data is generated.

[0075] Step S40, integrating the catalog items that are not marked as low-confidence catalog items with the catalog correction data, and generating a standard catalog file according to the standard catalog rules.

[0076] It should be noted that in the embodiments of the present application, the catalog items that are not marked as low-confidence catalog items refer to the part of the catalog items in the catalog item set generated by the edge node that have a confidence higher than a preset threshold. The standard catalog rules are a set of rules for data format conversion and coding organization of catalog content according to specific archive industry standards or enterprise self-defined specifications, which specifies the storage structure, field definition, data type, etc. of catalog data to ensure that the generated catalog file can be compatible, exchanged and long-term saved between different archive management systems.

[0077] Specifically, after the edge node receives the catalog correction data issued by the cloud server, the edge node performs data merging operation on the catalog correction data and the local saved trusted catalog items that are not marked as low-confidence catalog items. When merging, the two can be integrated into a complete catalog data set through field identifier matching. Then, the edge node converts the integrated complete catalog data set according to the standard catalog rules in the pre-set archive knowledge graph, and encapsulates the data to generate a catalog file that meets the standard requirements.

[0078] Exemplarily, in a library ancient book archive management system, after the edge node processes an ancient book, the bibliographic items not marked as low confidence are "book name: A Poems (confidence 0.92)" and "author: B (confidence 0.89)", and the bibliographic correction data received from the cloud server is "publishing dynasty: Ming Dynasty (corrected confidence 0.98, original confidence 0.7)". The edge node integrates the corrected data and the original reliable bibliographic items through the "publishing dynasty" field identifier to form a complete bibliographic data set. Then, according to the standard bibliographic rules of the library ancient book archive, the data is converted into a bibliographic file format conforming to the MARC standard, including setting the labels, indicators and the like of each field, such as the book name corresponding to the 245 field, the author corresponding to the 100 field, and the publishing dynasty corresponding to the 260 field, and finally generating a standard MARC format bibliographic file, facilitating the unified management and utilization of the ancient book archive by the library system.

[0079] In a feasible implementation, the lightweight double model includes a visual understanding model and a text understanding model, and step S10 can include steps S11-S12:

[0080] In step S11, the visual understanding model is used to perform three-level feature analysis on the archive information to generate a structured visual feature vector.

[0081] It should be noted that in the embodiments of the present application, the visual understanding model is a deep learning-based image processing model used to extract a structured visual feature vector from an archive image. Three-level feature analysis refers to analysis of an image by the model according to different levels to extract feature information from coarse to fine. First-level features usually involve the overall layout and regional division of an image; second-level features focus on local shapes and textures; and third-level features focus on more detailed visual elements such as colors and edges.

[0082] Specifically, the pre-collected archive information is input into the visual understanding model deployed on the edge node. After preprocessing the archive information by the visual understanding model, the archive information is analyzed layer by layer through the multi-layer structure of the convolutional neural network. In the first-level feature analysis, a larger convolution kernel can be used to extract the overall layout features of the image, such as identifying the title area, text area and chart area of the archive page. In the second-level feature analysis, a medium-sized convolution kernel can be used to extract local features, such as the shape of a text line and the outline of a chart. In the third-level feature analysis, a smaller convolution kernel can be used to extract detailed features, such as the font, color and edge of a text. These features are integrated into a structured visual feature vector for subsequent processing.

[0083] Step S12, obtaining the text recognition result of the archive information, inputting the text recognition result and the structured visual feature vector into the text understanding model, and generating a set of cataloging items with confidence annotations through the text understanding model.

[0084] It should be noted that in the embodiments of the present application, the text recognition result can be the text content extracted from the archive image through optical character recognition (OCR) technology. The text understanding model can be a natural language processing (NLP) model for semantic understanding and analysis of the text content, and a set of cataloging items with confidence annotations can be generated in combination with visual features.

[0085] Specifically, on the edge node, the OCR recognition result and the structured visual feature vector generated by the visual understanding model are input into the text understanding model. After preprocessing the text recognition result, the OCR recognition result is fused with the visual feature vector, the context information is provided by the visual feature, and the accuracy of text understanding is enhanced. Further, the text understanding model performs semantic analysis on the text content through natural language processing techniques such as word embedding, long short-term memory network or Transformer architecture, extracts key entities and semantic relationships, and calculates the confidence of each cataloging item. The calculation of the confidence is based on the semantic understanding accuracy of the model on the text content and the support degree of the visual feature.

[0086] In a possible implementation, step S12 can include steps S121-S122:

[0087] Step S121, text alignment is performed on the text recognition result and the structured visual feature vector to generate a visual enhanced text, and the visual enhanced text is input into the text understanding model.

[0088] It should be noted that text alignment refers to matching the text recognition result and the structured visual feature vector in space and semantics to generate a visual enhanced text. The visual enhanced text is an intermediate representation form that fuses visual features and text content, which can better reflect the position and context relationship of the text in the image, and provide more rich information for subsequent text understanding.

[0089] Specifically, on the edge node, the text recognition result and the structured visual feature vector generated by the visual understanding model are input into the text understanding model. The text understanding model parses the spatial coordinate data in the visual feature vector and determines the specific position of the text in the image. Then, the text recognition result is rearranged according to these spatial coordinates to make it consistent with the layout of the text in the image. For example, if the visual feature vector records that a certain piece of text is located in the upper left corner of the image, the model will place the corresponding text content in the corresponding position. In addition, the text understanding model formats the text according to the font, font size, etc. in the visual feature to generate visual enhancement text. For example, if the visual feature indicates that a certain piece of text is a title, the model will format it as bold, centered, etc. to enhance its visual effect. In this way, the visual enhancement text not only contains the text content, but also retains the visual features and spatial relationships of the text in the image.

[0090] Step S122, performing entity extraction, semantic completion and confidence calculation on the visual enhancement text through the text understanding model to generate a set of cataloging items with confidence annotations.

[0091] It should be noted that entity extraction refers to identifying entities with specific semantics from the visual enhancement text, such as names, dates, locations, etc. Semantic completion refers to supplementing and correcting incomplete or ambiguous text content based on context information to improve the completeness and accuracy of the text. Confidence calculation refers to confidence evaluation of each extracted entity or generated cataloging item to reflect the degree of confidence of the model in its accuracy.

[0092] Specifically, the text understanding model is used to analyze the visual enhancement text. First, key entities can be extracted using natural language processing techniques. Then, the text is semantically completed based on context information. For example, if a piece of text mentions "Meeting time: 2024-05", but lacks the second half of the date, the text understanding model can infer that the missing part is "-20" based on the complete date contained in the meeting record, and make the supplement. Further, confidence calculation is performed on each entity. The calculation of confidence is based on multiple factors, including the accuracy of entity extraction, the reasonableness of semantic completion, and the degree of support of visual features.

[0093] For example, suppose the visually enhanced text is “[Text content: Sales revenue in the first quarter of 2024 was 1 million yuan, location coordinates: (100, 200), color: black, font: SimSun]”. First, entity extraction is performed, identifying “first quarter of 2024” as a time entity and “1 million yuan” as a monetary entity. Then, semantic completion is performed, completing the unit of “sales revenue” to “ten thousand yuan” based on the information in the text and grammatical rules. Finally, confidence is calculated. Assuming the entity extraction accuracy is 0.95, the semantic completion accuracy is 0.98, and the visual feature support rate is 0.99, with weights of 0.4, 0.3, and 0.3 respectively, the confidence is 0.95×0.4+0.98×0.3+0.99×0.3=0.97. A set of bibliographical entries with confidence labels is generated, such as “[Time: First quarter of 2024 (confidence 0.97), Amount: 1 million yuan (confidence 0.97)]”.

[0094] In one feasible implementation, step S122 may include steps S1221 to S1223:

[0095] Step S1221: Based on the spatial coordinate data in the structured visual feature vector, perform spatial constraint operation on the visually enhanced text, delineate the entity extraction region according to the document layout features, and generate a regional entity set containing regional positioning accuracy parameters.

[0096] It should be noted that the aforementioned spatial coordinate data refers to the positional information of each component of the archival information recorded in the visual feature vector within the image. This information is typically represented in coordinate form and is used to determine the specific location of elements such as text and charts on the archival page. A regionalized entity set refers to a set formed by dividing entities in the visually enhanced text into specific regions according to their spatial location within the archival image, based on the archival layout characteristics. Each regionalized entity includes the entity itself and its corresponding regional positioning accuracy parameter, which measures the degree to which the entity is accurately located with the support of spatial coordinate data.

[0097] Specifically, at the edge nodes, after receiving visually enhanced text, the text understanding model first parses the spatial coordinate data within it. This coordinate data can be the boundary coordinates of text blocks, the coordinates of chart areas, etc., enabling precise location of each entity in the text within the archival image. Then, based on the archival layout features and the spatial coordinate data, entity extraction regions are defined. For each defined region, entities located within that region are extracted from the visually enhanced text, and the region localization accuracy parameter is calculated based on the precision of the spatial coordinate data.

[0098] Exemplarily, assuming that there is an academic paper archive image, the visual enhancement text contains multiple entities and their coordinates. After the coordinate analysis by the literal understanding model, according to the usual layout of the academic paper, the entity located at the top is determined as the title and author information area, the entity located in the middle part is determined as the main text area, and the entity located at the bottom is determined as the reference area. For an entity “experimental method” in the main text area, the coordinates are highly consistent with the boundary of the main text area, and the coordinate accuracy reaches the pixel level, so a higher area positioning accuracy parameter is calculated; and for an entity “page number” located near the margin, the coordinates are slightly blurred, so a lower area positioning accuracy parameter is calculated.

[0099] In step S1222, according to the document type identifier of the archive information, the corresponding cataloging rule is loaded from the preset archive knowledge graph, the semantic completion and format standardization operation is performed on the regionalized entity set, and the standardized cataloging item containing the semantic rationality parameter is generated.

[0100] It should be noted that the document type identifier refers to a specific identifier used to distinguish the document category to which the archive information belongs, such as contract document, academic paper, technical report, etc., which can help the system quickly identify the type characteristics and corresponding cataloging rules of the document. The archive knowledge graph is a structured knowledge base that stores different types of document cataloging specifications, semantic associations, etc.

[0101] Specifically, when processing archive information, the edge node identifies and obtains the document type identifier by analyzing the file extension, document template features, specific keywords, etc. of the archive file. Then, according to the document type identifier, the corresponding cataloging rule is loaded from the preset archive knowledge graph. The archive knowledge graph is a structured knowledge base that stores different types of document cataloging specifications, semantic associations, etc. For the extracted regionalized entity set, the semantic completion and format standardization operation is performed according to the loaded cataloging rule. Through these operations, the standardized cataloging item containing the semantic rationality parameter is generated, and the semantic rationality parameter is used to measure the semantic and format compliance of the entity after semantic completion and format standardization.

[0102] In step S1223, based on the area positioning accuracy parameter and the semantic rationality parameter, a multi-modal scoring matrix is constructed, the confidence fusion calculation is performed on each standardized cataloging item, and the cataloging item set with confidence annotation is generated.

[0103] It should be noted that in the embodiments of the present application, the multi-modal scoring matrix is a matrix structure for comprehensive evaluation of the confidence of standardized bibliographic items, the dimensions of which include a regional positioning accuracy parameter and a semantic reasonableness parameter, and the fusion processing of different parameters is realized through matrix operation to obtain the final confidence value of each standardized bibliographic item. The regional positioning accuracy parameter reflects the positioning accuracy of the entity in the spatial position, and the semantic reasonableness parameter embodies the standard reasonableness of the entity in the semantics and format, and both of them jointly determine the credibility of the bibliographic item.

[0104] Specifically, for each standardized bibliographic item, a multi-modal scoring matrix is constructed using the regional positioning accuracy parameter and the semantic reasonableness parameter. First, the rows and columns of the matrix are determined to correspond to different parameter types, for example, the rows represent different level divisions of the regional positioning accuracy parameter, and the columns represent different level divisions of the semantic reasonableness parameter. Then, according to a pre-set weight distribution strategy, a weight is assigned to each parameter type. The weight distribution can be based on the judgment of the importance of spatial positioning and semantic specification in confidence evaluation, for example, it is considered that semantic reasonableness has a greater impact on confidence, and a higher weight can be assigned to it. Then, the regional positioning accuracy parameter and the semantic reasonableness parameter value corresponding to each standardized bibliographic item are substituted into the corresponding position of the matrix, and the parameters are fused and calculated through matrix operation, such as weighted summation, to obtain the final confidence value of each standardized bibliographic item.

[0105] In a feasible implementation, when constructing the multi-modal scoring matrix, first, the regional positioning accuracy parameter and the semantic reasonableness parameter are taken as the scoring basis of the spatial and semantic dimensions to form a two-dimensional joint scoring space. For each entity in the bibliographic item set, let its regional positioning accuracy parameter be a∈[0,1], the higher the value, the more accurate the spatial positioning, and the semantic reasonableness parameter be s∈[0,1], the higher the value, the stronger the semantic and format specification. Through the construction of the mapping relationship M:(a,s)→c, the two are fused into a comprehensive confidence parameter c.

[0106] Specifically, the multi-modal scoring matrix is defined as C=α*A+β*S, where A and S are vectors composed of the a and s values of all entities, and α and β are weight coefficients dynamically adjusted based on the document type (satisfying α+β=1).

[0107] For example, in a strictly formatted contract document, α=0.7 and β=0.3 can be set to strengthen the spatial constraint, and in a semantically complex academic literature, α=0.4 and β=0.6 can be used to focus on semantic logic. The finally generated matrix realizes cross-modal feature quantization through linear weighting, providing uniform and comparable confidence evaluation for each bibliographic item.

[0108] In the embodiment of the present application, the hierarchical feature extraction and fusion of archive information are realized by deploying a lightweight double model in the edge node, and a set of catalog items with confidence annotation is generated, which effectively improves the preliminary processing efficiency and accuracy of archive catalog. According to the three-level feature analysis of the visual understanding model and the entity extraction, semantic completion and confidence calculation of the text understanding model, the extraction of key archive information is more accurate, and the confidence annotation can intuitively reflect the reliability of the catalog items, which is convenient for subsequent targeted correction. While improving the efficiency of archive catalog, the accuracy, completeness and standardization of the catalog result are effectively ensured, which provides strong support for digital management of archives.

[0109] Further, based on Figure 2 , Figure 2 The flowchart of the second exemplary embodiment of the archive intelligent catalog method of the present application is shown in the embodiment, and step S40 can include steps S41-S42:

[0110] Step S41, performing a data merging operation on the trusted catalog items that are not marked as low-confidence catalog items and the catalog correction data, and establishing a complete catalog data set through field identifier matching;

[0111] Specifically, after receiving the catalog correction data issued by the cloud server, the edge node will first scan the trusted catalog items stored by itself and each field in the catalog correction data, and extract the respective field identifiers. These field identifiers can include archive number, file type, date and other information used to describe the properties of the archive. Then, the edge node uses a preset matching algorithm to compare these identifiers one by one. Simple string exact matching can be used, and matching based on hash algorithm can also be used to determine which field identifiers are the same in the trusted catalog items and the catalog correction data, thereby establishing the association between them. Once the matching is successful, the edge node will combine the corresponding trusted catalog items and catalog correction data according to certain data integration rules to form a complete catalog data set. This process needs to consider the integrity, consistency and accuracy of the data to ensure that the integrated data can truly and completely reflect the catalog information of the archive.

[0112] Exemplarily, assume that a trusted entry of a file contains "File Number: 123456", "File Type: Contract", and "Date: 2024-05-20", and the cloud server issues a record correction data containing "File Number: 123456", "File Type: Supplementary Contract", and "Date: 2024-05-20". Through field identifier matching, the edge node identifies that the three field identifiers "File Number", "File Type", and "Date" are the same, and then integrates the two to form a complete record data set, in which "File Type" is based on the record correction data, because it is more accurate data obtained from the cloud server after multi-modal comparison and analysis.

[0113] Step S42, based on the standard record rules in the pre-set archive knowledge graph, the complete record data set is format-converted and data-encapsulated to generate a standard record file.

[0114] It should be noted that the standard record rules in the pre-set archive knowledge graph refer to a series of criteria pre-set for standardizing the format and organization of archive data. These rules define the coding format of different fields of the archive, such as the coding method of the file number, the file type coding, the date format, and the data organization structure, such as the arrangement order of the fields, the hierarchical relationship of the data, etc. These rules ensure that the final generated standard record file conforms to the unified specification, facilitating storage, retrieval, and sharing.

[0115] Specifically, after integrating the complete record data set, the edge node calls the standard record rules in the pre-set archive knowledge graph. According to these rules, the format of each field in the complete record data set is converted. At the same time, according to the data organization structure requirements in the standard record rules, the converted data is data-encapsulated. This may involve nesting data according to certain hierarchical relationships, or adding necessary metadata information such as archive source identification, generation timestamp, etc. In the encapsulation process, the edge node will follow specific data encapsulation protocols such as XML or JSON format to ensure that the generated standard record file has good readability and compatibility. In the whole process, the edge node will conduct strict format verification to ensure that all data meet the requirements of the standard record rules, and finally generate a standard record file that meets the specifications.

[0116] The embodiment of the application realizes the integration of the trusted recording items not marked as low confidence and the recording correction data issued by the cloud. The field identifier matching technology is used to efficiently associate the data of different sources at the edge node, ensuring the integrity and accuracy of data integration. At the same time, the standard recording rules in the preset archive knowledge graph are used for format conversion and data encapsulation, to generate the standard recording files conforming to the standard, improve the standardization and compatibility of the data, and provide convenience for subsequent storage, retrieval and sharing, and improve the efficiency and quality of the entire archive intelligent recording process.

[0117] Based on the first embodiment of the application, the third exemplary embodiment of the application is proposed. In the third exemplary embodiment of the application, the same or similar contents as the above-mentioned first embodiment can be referred to the above description, and will not be described in detail hereinafter.

[0118] Based on this, the application further provides an archive intelligent recording method applied to a cloud server, which refers to Figure 3 , Figure 3 The flowchart of the third exemplary embodiment of the archive intelligent recording method is shown in the figure. In the embodiment, the message authentication method includes steps A10-A30.

[0119] Step A10, receiving the low-confidence recording items transported by the edge node, the low-confidence recording items being generated based on the edge node through pre-deployed lightweight double models for hierarchical feature extraction and fusion of pre-collected archive information;

[0120] Specifically, the cloud server receives the low-confidence recording item data packet transmitted from the edge node through a secure communication interface, and the data packet contains the to-be-corrected recording items and their associated metadata generated by the edge node lightweight double model. After decrypting and decompressing the received data, the cloud server performs integrity check to verify whether the data packet contains complete low-confidence recording items, corresponding original archive image segments and structured visual feature vectors generated by the edge node. Then, the cloud server stores the data passing the check in a distributed cache queue, and allocates a unique tracking identifier for each processing task, and establishes a mapping relationship with the source edge node. In this process, the cloud server synchronously records the audit information such as the receiving timestamp, data source node ID and data volume, to ensure the traceability of the subsequent correction process. This step establishes a standardized data input channel for intelligent correction on the cloud side, and guarantees the data reliability of the edge and cloud collaborative recording.

[0121] Step A20, performing multi-modal comparative analysis on the low-confidence recording items, generating recording correction data, and issuing the recording correction data to the edge node;

[0122] In a feasible implementation manner, step A20 can include steps A21-A24:

[0123] Step A21, calling the similar archive template library to perform multi-modal data alignment on the low-confidence entry, and extracting a key field difference feature vector;

[0124] It should be noted that the similar archive template library is a database that stores a plurality of types of archive templates, which contain standard structures, key fields and feature information of different types of archives, etc., for comparison and analysis with the to-be-processed archive. Multi-modal data alignment refers to matching and corresponding different forms of data, so that they can be compared and analyzed in the same semantic space, thereby extracting a key field difference feature vector, which is used to represent the difference features of the to-be-processed archive and the template archive in the key field.

[0125] Specifically, the cloud server filters out templates similar to the to-be-processed archive type from the similar archive template library stored by itself by analyzing the document type identifier, file extension, and pre-set key field matching of the archive. Then, the low-confidence entry is aligned with the key fields of these templates in multi-modal data. Specifically, text features and image features in the entry can be extracted and compared with corresponding features in the template. Text feature comparison can be achieved by calculating the word vector similarity of the text and the string matching degree; image feature comparison can use convolutional neural network to extract image feature vectors and then calculate the similarity. By comparison, the differences between the entry and the template in the key field are found out, and a key field difference feature vector is extracted.

[0126] Step A22, performing catalog rule conflict analysis based on the key field difference feature vector, and generating a conflict analysis report containing a difference type identifier;

[0127] It should be noted that catalog rule conflict analysis refers to evaluating the key field difference feature vector according to the pre-defined catalog rule, analyzing whether the low-confidence entry conflicts with the standard catalog rule, and generating a conflict analysis report containing a difference type identifier and a correction suggestion, wherein the difference type identifier is used to clearly indicate the specific type of conflict between the entry and the rule, and the correction suggestion provides guidance to solve the conflict.

[0128] Specifically, the pre-loaded catalog rule defines the legal value range, format specification, and semantic constraints of each key field of the archive. Then, the key field difference feature vector is compared with the catalog rule to determine whether the difference feature violates the catalog rule. If it violates, the type of conflict is determined, such as format conflict, semantic conflict, and value range conflict, and corresponding correction suggestions are generated according to the conflict type. When generating the conflict analysis report, the difference type identifier and the correction suggestion of each entry with conflict are listed in detail, so that subsequent precise data correction operations can be performed.

[0129] Exemplarily, in a low-confidence bibliographic item of an academic paper, the key field difference feature vector shows that the "reference" format is different from the standard format in the similar archive template library. The cloud server loads the bibliographic rules of the academic paper, and after comparison, it is found that the reference format of the bibliographic item does not mark the DOI (Digital Object Identifier) number of the literature according to the rules, which belongs to format conflict. Therefore, in the conflict analysis report, the difference type is identified as "reference format conflict", and the correction suggestion is "supplement the DOI number of the reference according to the standard format".

[0130] Step A23, according to the difference type identification, calling the pre-set bibliographic correction strategy library for data correction operation, generating bibliographic correction data;

[0131] It should be noted that the bibliographic correction strategy library is a resource library that stores a variety of bibliographic correction strategies. These strategies provide correction methods and steps for different types of bibliographic errors, including data format adjustment, semantic correction, missing data supplementation, etc. Data correction operation refers to selecting the corresponding correction strategy from the bibliographic correction strategy library according to the difference type identification in the conflict analysis report, and modifying and optimizing the low-confidence bibliographic item to make it comply with the standard bibliographic rules, generating bibliographic correction data, which contains the corrected bibliographic content and related correction information.

[0132] Specifically, after the cloud server generates the conflict analysis report, it matches the corresponding correction strategy from the bibliographic correction strategy library according to the difference type identification in the report. For example, for the case where the difference type identification is "date format conflict", the selected correction strategy may be to uniformly convert the date format to the standard "YY-MM-DD" format. For the bibliographic item with "semantic ambiguity", the correction strategy may include calling a semantic analysis model to further clarify the semantics, or replacing the semantics according to the context and related standard vocabulary. In the data correction operation, different technical means can be used, such as using regular expressions to correct the text format, using natural language processing models to optimize the semantics, or supplementing missing data items by searching standard data dictionaries, etc. The corrected bibliographic item forms the bibliographic correction data, which contains the corrected specific content, the correction basis strategy identification, and the association information with the original low-confidence bibliographic item, for subsequent integration and tracing.

[0133] Step A24, binding the bibliographic correction data with the corresponding edge node identification, and delivering it to the edge node through an encrypted channel.

[0134] Specifically, after completing the data correction operation, it is necessary to bind the catalog correction data with the corresponding edge node identifier. This can be achieved by adding an edge node identifier field in the data structure of the catalog correction data, or by establishing a mapping table to associate the unique identifier of the catalog correction data with the edge node identifier. Then, the cloud server sends the bound catalog correction data to the corresponding edge node through an encrypted channel. The establishment of the encrypted channel can use various encryption protocols and technologies, such as the SSL / TLS protocol to encrypt the data in the transmission layer, or a combination of symmetric and asymmetric encryption algorithms to encrypt the data. During data transmission, the encrypted channel verifies the identity of the recipient to ensure that the data is only received and decrypted by the legitimate edge node. In this way, after receiving the data, the edge node can ensure through the decryption and identity verification process that the received data is from a trusted cloud server and that the data has not been tampered with during transmission.

[0135] The above implementation process effectively solves the accuracy problem of low-confidence catalog items generated by edge nodes by performing multi-modal comparison analysis and data correction operations on the cloud server side. First, by calling the similar archive template library for multi-modal data alignment, the key field difference feature vector between the low-confidence catalog item and the standard template is accurately extracted, providing a basis for subsequent accurate correction. Then, based on the key field difference feature vector, in-depth catalog rule conflict analysis is performed, which can accurately identify the conflict type between the catalog item and the pre-set rules, and generate a conflict analysis report with targeted correction suggestions, ensuring the accuracy and reasonableness of the correction direction. Then, using the pre-set catalog correction strategy library for data correction operation, the appropriate correction strategy is selected according to the difference type identifier in the conflict analysis report to accurately correct the low-confidence catalog item and generate reliable catalog correction data. Finally, the catalog correction data is bound with the corresponding edge node identifier and delivered to the edge node through an encrypted channel, not only ensuring the security of data transmission, but also enabling the edge node to obtain accurate correction data in a timely manner, thereby realizing the efficient, accurate and secure operation of the archive intelligent cataloging system, greatly improving the overall quality and efficiency of archive cataloging.

[0136] Step A30, integrating the catalog items that are not marked as low-confidence catalog items with the catalog correction data through the edge node, and generating a standard catalog file according to the standard catalog rules.

[0137] It should be noted that in one embodiment, the catalog items that are not marked as low-confidence catalog items are integrated with the catalog correction data, and the above integration process can be performed on the edge node, and the standard catalog file generated according to the standard catalog rules is stored in the edge node or the cloud storage system.

[0138] The above embodiment adopts an edge computing priority strategy, and the edge node is taken as an execution subject to complete data integration. In this mode, the edge node first establishes a mapping relationship between local high-confidence catalog items and catalog correction data issued by the cloud, and ensures the time sequence consistency and logical integrity of the two types of data through a data verification mechanism. Then, a built-in standardization coding engine is called, structured conversion is performed according to the preset archive catalog rules, and a standard catalog file conforming to the specification is generated. Finally, according to the storage strategy configuration, the file can be selectively persisted in the local edge storage module or uploaded to the cloud object storage service through an encrypted channel, while updating the distributed index to support cross-node retrieval.

[0139] In another embodiment, the above integration process can also be executed on the cloud server, and the standard catalog file generated according to the standard catalog rules is stored in the storage system of the edge node or the cloud.

[0140] The above embodiment adopts a cloud centralized processing strategy, and the cloud server undertakes the core integration function. After receiving the original high-confidence data package uploaded by the edge node, the cloud service performs multi-version merging on the data package and the correction data generated by itself, and processes possible field-level differences using a conflict resolution algorithm. After completing data fusion, the cloud server uses an extensible standardization processing cluster to dynamically load the coding template matched with the archive type, and generates a standard catalog file containing complete metadata tags. The system supports returning the final file to the source edge node for local storage, or directly storing it in the cloud distributed archive database, and providing query interfaces to each edge node through a unified metadata service.

[0141] Further, a fourth exemplary embodiment of the archive intelligent cataloging method is provided, which refers to Figure 4 , Figure 4 The fourth exemplary embodiment of the archive intelligent cataloging method is a flowchart, and the embodiment can be applied to a cloud server. At this time, the archive intelligent cataloging method can include steps B10-B40.

[0142] Step B10, when detecting that any edge node is invalid, sending a task takeover instruction to the edge node;

[0143] Specifically, the cloud server continuously monitors the heartbeat signals and task processing states of each edge node. When it detects that a certain edge node has exceeded the preset number of consecutive heartbeat timeouts or that the task queue backlog has exceeded the preset capacity, it automatically triggers the node failure determination mechanism. The cloud server first sends three reconnection requests to the node for final confirmation. If there is still no response, it sends an encrypted task takeover instruction to the node through the message middleware. The instruction includes a takeover authorization token, a data resume start position marker, and temporary encryption channel configuration parameters. At the same time, the server marks the node as "to be taken over" in the distributed task scheduling center and freezes related task allocation.

[0144] Step B20, obtaining the to-be-processed archive information of the edge node;

[0145] The cloud server can obtain the to-be-processed data through two ways:

[0146] For data that has been partially uploaded before the edge node fails, the data is recovered from the breakpoint resume buffer. Specifically, the to-be-processed archive information and intermediate results in the local cache of the edge node are pulled through the secure SCP protocol.

[0147] For the original archive that has not been uploaded, the backup transmission channel is activated. When the edge node is completely offline, the latest synchronized copy data is obtained from the hot standby node paired with the edge node.

[0148] Step B30, performing hierarchical feature extraction and multi-modal comparison analysis on the archive information to generate a complete bibliographic item set;

[0149] It should be noted that the process of hierarchical feature extraction and multi-modal comparison analysis of the cloud server on the archive information has significant performance enhancement and functional expansion compared to the edge node. The cloud server can deploy an enhanced visual understanding model using its powerful computing resources. This model can add deep attention mechanisms and adaptive image repair modules based on the lightweight architecture of the edge node, enabling it to handle severely blurred or damaged archive images that the edge node cannot handle. In terms of text understanding, the cloud server uses a dynamic loading multi-expert model system that automatically activates corresponding professional sub-models based on the archive type, providing more fine-grained domain adaptation capabilities than the edge node.

[0150] Step B40, performing format conversion and data encapsulation on the complete bibliographic item set to generate a standard bibliographic file.

[0151] It should be noted that the cloud server can use a dynamic XSD template engine to support multiple industry bibliographic standards simultaneously and automatically detect and convert metadata fields of different standards. This multi-standard adaptive capability far exceeds the fixed encoding rule set of the edge node. In particular, the cloud processing also has automatic version management and cross-agency data sharing capabilities.

[0152] Further, the fourth exemplary embodiment of the intelligent filing method of the present application can also be applied to the edge node, at this time, the intelligent filing method can include step B50:

[0153] Step B50: When receiving the task takeover instruction from the cloud server, upload the to-be-processed archive information to the cloud server.

[0154] Specifically, the behavior of the edge node after receiving the takeover instruction can include:

[0155] Pause the local processing pipeline, serialize the to-be-processed data in memory into a checkpoint file;

[0156] Start the emergency communication mode, and preferentially upload three types of key data: unprocessed raw images, intermediate results that have been parsed but not verified, and local knowledge base incremental update logs;

[0157] Adopt block incremental transmission technology, send HMAC-SHA256 check code to the cloud end after completing 2MB data upload. Maintain dual-channel backup during transmission, the main channel uses QUIC protocol to ensure weak network transmission, and the standby channel synchronizes metadata through LoRaWAN. After completing the transmission, the node enters standby state and periodically requests state synchronization.

[0158] The embodiment of the present application builds a high-reliability and high-elasticity distributed filing system through the cooperative processing mechanism of the cloud server and the edge node. On the cloud server side, the edge node failure detection and task takeover capability is realized, and through the enhanced processing model and global knowledge base support, the processing accuracy and standardization level of complex archives are improved. The edge node side designs an efficient emergency response mechanism to ensure the safe transfer of key data in abnormal situations. In particular, the cloud service shows differentiated advantages over the edge node: in terms of processing capacity, it uses deep enhancement models to achieve accurate parsing of low-quality documents; in terms of standardization support, it meets the needs of multi-industry standards through a dynamic template engine. This hierarchical cooperative architecture not only takes advantage of the real-time nature of edge computing, but also integrates the global processing capabilities of the cloud, forming a complete intelligent filing solution that effectively guarantees the service continuity and processing quality of the system in various operating environments.

[0159] In addition, the present application also proposes an intelligent filing device for archives, please refer to Figure 5 , Figure 5 The module structure diagram of the intelligent filing device for archives of the embodiment of the present application, the intelligent filing device for archives comprises:

[0160] The first generation module 10 is configured to perform hierarchical feature extraction and fusion on the pre-collected archive information by using a pre-deployed lightweight double model, and generate a catalog item set with confidence annotation.

[0161] The transmission module 20 is configured to upload low-confidence catalog items in the catalog item set to a cloud server.

[0162] The acquisition module 30 is configured to acquire catalog correction data delivered by the cloud server, wherein the catalog correction data is generated based on multi-modal comparative analysis of the cloud server on the low-confidence catalog items.

[0163] The second generation module 40 is configured to integrate catalog items that are not marked as low-confidence catalog items with the catalog correction data, and generate a standard catalog file according to standard catalog rules.

[0164] The archive intelligent cataloging device provided by the present application adopts the archive intelligent cataloging method in the above embodiments, and aims to solve the technical problems of response delay, privacy leakage risk and poor generalization ability of the archive intelligent cataloging service based on the public cloud. Compared with the prior art, the archive intelligent cataloging device provided by the present application has the same beneficial effects as the archive intelligent cataloging method provided by the above embodiments, and other technical features in the archive intelligent cataloging device are the same as the features disclosed in the above embodiments, which will not be repeated here.

[0165] The present application provides an archive intelligent cataloging device, which comprises at least one processor and a memory in communication connection with the at least one processor. The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the archive intelligent cataloging method in the above embodiment one.

[0166] The archive intelligent cataloging device in the present application can include but is not limited to mobile terminals such as mobile phones, notebook computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), and the like, and fixed terminals such as digital TVs, desktop computers, and the like. Figure 6 The archive intelligent cataloging device shown is only an example, and should not limit the functions and use range of the present application.

[0167] As Figure 6As shown, the archive intelligent cataloging device can include a processing apparatus 1001 (such as a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to programs stored in a read-only memory 1002 or loaded from a storage apparatus 1003 into a random access memory 1004. In the random access memory 1004, various programs and data required for the operation of the archive intelligent cataloging device are also stored. The processing apparatus 1001, the read-only memory 1002, and the random access memory 1004 are connected to each other through a bus 1005. An input / output interface 1006 is also connected to the bus. Generally, the following systems can be connected to the input / output interface 1006: input apparatuses 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output apparatuses 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; the storage apparatus 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication apparatus 1009. The communication apparatus 1009 can allow the archive intelligent cataloging device to communicate with other devices wirelessly or by wire to exchange data. Although the archive intelligent cataloging device with various systems is shown in the figure, it should be understood that all the shown systems are not required to be implemented or possessed. More or fewer systems can be alternatively implemented or possessed.

[0168] In particular, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product comprising a computer program carried on a computer readable medium, the computer program containing program codes for executing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network through the communication apparatus, or installed from the storage apparatus 1003, or installed from the ROM 1002. When the computer program is executed by the processing apparatus 1001, the above-mentioned functions defined in the methods of the embodiments disclosed in the present application are performed.

[0169] The archive intelligent cataloging device provided in the present application adopts the archive intelligent cataloging method in the above-mentioned embodiments, and aims to solve the technical problems of response delay, privacy leakage risk and poor generalization ability of the archive intelligent cataloging service based on the public cloud. Compared with the prior art, the archive intelligent cataloging device provided in the present application has the same beneficial effects as the archive intelligent cataloging method provided in the above-mentioned embodiments, and other technical features in the archive intelligent cataloging device are the same as the features disclosed in the above-mentioned method, which will not be repeated here.

[0170] It should be understood that various parts of the present application can be realized with hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0171] The above description is merely that of a specific implementation of the present application, and the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, and all such changes or replacements should be encompassed within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0172] The present application provides a computer readable storage medium having stored thereon computer readable program instructions (i.e., computer programs) for performing the archive intelligent cataloging method in the above embodiments.

[0173] The computer readable storage medium provided by the present application may, for example, be a U disk, but is not limited to an electric, magnetic, optical, electromagnetic, infrared, or semiconductor system or device, or any combination thereof. More specific examples of the computer readable storage medium can include, but are not limited to, an electric connection having one or more conductive wires, a portable computer disk, a hard disk, a random access memory (RAM), a read only memory (ROM), an erasable programmable read only memory (EPROM or flash memory), an optical fiber, a portable compact disk read only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present embodiment, the computer readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system or device. The program code contained on the computer readable storage medium can be transmitted by any suitable medium, including but not limited to an electric wire, an optical cable, an RF (Radio Frequency), etc., or any suitable combination thereof.

[0174] The above computer readable storage medium can be contained in the archive intelligent cataloging device; or can exist separately without being assembled into the archive intelligent cataloging device.

[0175] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0176] The flow diagrams and the block diagrams in the drawings are meant as methodological and functional description of implementations of systems, methods, and computer program products according to various embodiments of the present application. In this regard, each block in the flow diagrams or block diagrams can represent a module, a segment, or a portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that in some alternative implementations, the functions noted in the blocks can occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently or the blocks may

[0177] The modules involved in the embodiments of the present application can be implemented in software or hardware. In some cases, the name of the module does not constitute a limitation on the module itself.

[0178] The readable storage medium provided by the present application is a computer readable storage medium, which stores computer readable program instructions (i.e. computer program) for executing the above-mentioned file intelligent cataloging method, and aims to solve the technical problems of response delay, privacy leakage risk and poor generalization ability of the file intelligent cataloging service based on public cloud. Compared with the prior art, the computer readable storage medium provided by the present application has the same beneficial effects as the file intelligent cataloging method provided by the above-mentioned embodiments, and will not be described here.

[0179] The application further provides a computer program product comprising a computer program which, when executed by a processor, implements the steps of the archive intelligent registration method as described above.

[0180] The computer program product provided by the application aims to solve the technical problems of response delay, privacy leakage risk and poor generalization ability of the archive intelligent registration service based on a public cloud. Compared with the prior art, the beneficial effects of the computer program product provided by the application are the same as those of the archive intelligent registration method provided by the above-mentioned embodiments, and are not described here.

[0181] Compared with the prior art, the archive intelligent registration method, device, equipment, medium and computer product provided by the embodiments of the application can extract the service feature information of the target service, perform data standardization processing on the service feature information to obtain standard feature data, perform hash processing on the standard feature data to obtain unique feature data, perform numerical processing and splicing processing on the unique feature data to obtain a first service feature value, accumulate the first service feature value of the target service to obtain a target service feature value, and finally compare the target service feature value with a feature value set to obtain an archive intelligent registration result. Compared with the traditional method of generating a unique key value or a continuous serial number for each service to identify repeated services, the method is more efficient, flexible and reliable. Based on the scheme of the application, the service in a complex scenario is converted through a series of simple conversions, and finally converted into the comparison of two numbers, so that the comparison process is very intuitive and efficient. The system only needs to simply compare whether the two values are equal, and can quickly judge whether the two services are completely the same.

[0182] It should be noted that in this document, the terms "comprising", "including", or any other variant thereof are intended to cover non-exclusive inclusions, so that a process, method, article or system including a series of elements does not only include those elements, but also includes other elements not explicitly listed, or includes elements inherent to such a process, method, article or system. Without more limitations, the element defined by the statement "including a" does not exclude the presence of another identical element in the process, method, article or system including the element.

[0183] The above-mentioned serial numbers of the embodiments of the application are only for description, and do not represent the advantages and disadvantages of the embodiments.

[0184] Through the above description of the embodiments, those skilled in the art can clearly understand that the above-mentioned example method can be realized by means of software and a necessary general hardware platform, and of course, it can also be realized by hardware, but in many cases, the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product in essence or in the form of a contribution to the prior art. The computer software product is stored in a storage medium such as a ROM / RAM, a magnetic disk, or an optical disk, and includes a plurality of instructions for causing a terminal device (which can be a mobile phone, a computer, a server, a controlled terminal, or a network device) to execute the method of each embodiment of the present application.

[0185] The above is only a preferred embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation using the content of the specification and drawings, or direct or indirect application in other related technical fields, is also included in the patent protection scope of the present application.

Claims

1. An archival intellectual property rights registration method, characterized by, The intelligent cataloging method applied to the edge node comprises: Through a pre-deployed lightweight double model, hierarchical feature extraction and fusion are performed on pre-collected archive information to generate a cataloging item set with confidence annotation; Low-confidence cataloging items in the cataloging item set are uploaded to a cloud server; Cataloging correction data transmitted by the cloud server is obtained, and the cataloging correction data is generated based on multi-modal comparative analysis of the cloud server on the low-confidence cataloging items; Cataloging items that are not marked as low-confidence cataloging items are integrated with the cataloging correction data to generate a standard cataloging file according to standard cataloging rules; The step of generating a cataloging item set with confidence annotation by performing hierarchical feature extraction and fusion on pre-collected archive information through a pre-deployed lightweight double model comprises: The lightweight double model comprises a visual understanding model and a text understanding model; Three-level feature analysis is performed on the archive information by using the visual understanding model to generate a structured visual feature vector; wherein, the first-level feature comprises the overall layout and regional division of an image; the second-level feature focuses on local shape and texture; and the third-level feature focuses on detailed visual elements; A text recognition result of the archive information is obtained; Text alignment is performed on the text recognition result and the structured visual feature vector to generate visual enhanced text, and the visual enhanced text is input into the text understanding model; Entity extraction, semantic completion and confidence calculation are performed on the visual enhanced text by using the text understanding model to generate a cataloging item set with confidence annotation; wherein, the semantic completion refers to supplementing and correcting incomplete or ambiguous text content according to context information.

2. The archival intellectual property description method of claim 1, wherein, The step of generating a cataloging item set with confidence annotation by performing entity extraction, semantic completion and confidence calculation on the visual enhanced text by using the text understanding model comprises: Based on the spatial coordinate data in the structured visual feature vector, a spatial constraint operation is performed on the visual enhanced text, an entity extraction region is determined according to archive format features, a regionalized entity set containing a region positioning accuracy parameter is generated; According to the document type identifier of the archive information, corresponding cataloging rules are loaded from a pre-set archive knowledge graph, semantic completion and format standardization operations are performed on the regionalized entity set to generate a standardized cataloging item containing a semantic rationality parameter; Based on the region positioning accuracy parameter and the semantic rationality parameter, a multi-modal scoring matrix is constructed, confidence fusion calculation is performed on each standardized cataloging item to generate a cataloging item set with confidence annotation.

3. The archival intellectual property description method of claim 1, wherein, The intelligent cataloging method further comprises: When a task takeover instruction from the cloud server is received, pre-processing archive information is uploaded to the cloud server.

4. An archival intellectual property rights method, characterized by, The intelligent cataloging method applied to the cloud server comprises: Low-confidence cataloging items transmitted by an edge node are received, and the low-confidence cataloging items are generated based on the edge node performing hierarchical feature extraction and fusion on pre-collected archive information through a pre-deployed lightweight double model; Performing multi-modal comparative analysis on the low-confidence catalog item to generate catalog correction data, and issuing the catalog correction data to the edge node; Integrating catalog items that are not marked as low-confidence catalog items with the catalog correction data through the edge node, and generating a standard catalog file according to standard catalog rules; The step of generating the low-confidence catalog item based on the edge node through pre-deployed lightweight dual models to pre-collected archive information includes: The lightweight dual models include a visual understanding model and a text understanding model; Performing three-level feature analysis on the archive information using the visual understanding model to generate a structured visual feature vector; wherein the first-level feature includes the overall layout and regional division of the image; the second-level feature focuses on local shape and texture; and the third-level feature focuses on detailed visual elements; Obtaining a text recognition result of the archive information; Performing text alignment on the text recognition result and the structured visual feature vector to generate visually enhanced text, and inputting the visually enhanced text into the text understanding model; Performing entity extraction, semantic completion, and confidence calculation on the visually enhanced text through the text understanding model to generate a catalog item set with confidence annotations; wherein the semantic completion refers to supplementing and correcting incomplete or ambiguous text content according to context information.

5. The archival intellectual property description method of claim 4, wherein, The step of performing multi-modal comparative analysis on the low-confidence catalog item to generate catalog correction data, and issuing the catalog correction data to the edge node includes: Calling a similar archive template library to perform multi-modal data alignment on the low-confidence catalog item to extract a key field difference feature vector; Performing catalog rule conflict analysis based on the key field difference feature vector to generate a conflict analysis report containing a difference type identifier; According to the difference type identifier, calling a pre-set catalog correction strategy library to perform data correction operations to generate catalog correction data; Binding the catalog correction data with a corresponding edge node identifier and issuing it to the edge node through an encrypted channel.

6. The archival intellectual property description method of claim 4, wherein, The archive intelligent cataloging method further includes: When detecting that any of the edge nodes is invalid, sending a task takeover instruction to the edge node; Obtaining archive information to be processed by the edge node; Performing hierarchical feature extraction and multi-modal comparative analysis on the archive information to generate a complete catalog item set; Performing format conversion and data encapsulation on the complete catalog item set to generate a standard catalog file.

7. An archival intelligence cataloging device, characterized by, The device includes a memory, a processor, and a computer program stored on the memory and executable on the processor, the computer program being configured to implement the steps of the archive intelligent cataloging method according to any one of claims 1 to 6.

8. A storage medium, characterized by The storage medium is a computer-readable storage medium, and the storage medium stores a computer program, which is executed by a processor to implement the steps of the archive intelligent cataloging method according to any one of claims 1 to 6.

Images