Archive information management system based on cloud computing

Abstract

The invention discloses an archive information management system based on cloud computing, and the system comprises the steps: obtaining encrypted archive information, and transmitting the encrypted archive information to a cloud platform; carrying out image noise reduction on the encrypted archive information by utilizing wavelet transform, and converting an image into text data through a multi-head attention OCR engine; processing an archive entity relation graph by utilizing a GCN graph convolutional network in a cloud platform, learning classification topological characteristics, analyzing archive text semantics by adopting a mechanism, capturing context dependency, and establishing an adaptive deep classification model; inputting the processed archive information into an adaptive deep classification model for training to obtain archive information confidence; and performing archive information management according to the archive information confidence, wherein the management at least comprises automatic classification and information inspection. Complex association and context dependence among archives can be accurately captured, accurate classification and deep mining are carried out on archive information, and powerful support is provided for archive management and decision making.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to an archive information management system based on cloud computing. Background Art

[0002] Traditional archival information management models have numerous shortcomings. Traditional archival management relies on manual operations, whether for the archiving and retrieval of paper archives or the initial classification and organization of electronic archives. This is inefficient, error-prone, and difficult to handle with massive amounts of data. Manually organizing a complex archive can take hours and carries the risk of information entry errors. While existing archival management technologies have introduced cloud computing and artificial intelligence, they still have shortcomings in processing procedures and model construction. Some solutions lack effective protection for the encrypted transmission of archival information, which can easily lead to data leaks. Furthermore, archival classification models are simplistic, making it impossible to deeply explore the value of archival data to meet the needs of efficient and accurate management. Summary of the Invention

[0003] The purpose of the present invention is to solve the above problems and design an archive information management system based on cloud computing.

[0004] The technical solution of the present invention to achieve the above object is that, further, in the above-mentioned cloud computing-based archive information management system, the archive information management system method includes the following steps: An information acquisition and transmission module is used to acquire text information and image information of the archive to be stored, obtain archive information to be stored, encrypt the stored archive information using an AES encryption algorithm to obtain encrypted archive information, and transmit the encrypted archive information to the cloud platform; The cloud platform computing module is used to receive encrypted archival information, perform image noise reduction on the encrypted archival information using wavelet transform, and convert the image into text data through the multi-head attention OCR engine to obtain processed archival information; The classification model building module is used to process the archival entity relationship graph using the GCN graph convolutional network in the cloud platform, learn classification topology features, and use a combination of Bi-LSTM and attention mechanism to parse the archival text semantics, capture contextual dependencies, and build a GCN-Bi-LSTM adaptive deep classification model. An archival information classification module is used to input the processed archival information into the GCN-Bi-LSTM adaptive deep classification model for training to obtain the archival information confidence; The archival information management module is used to manage archival information according to the archival information confidence level, including at least automatic classification and information verification.

[0005] Furthermore, in the above-mentioned cloud computing-based archive information management system, the information acquisition and transmission module includes the following submodules: An encryption submodule, configured to encrypt the stored archive information using an AES encryption algorithm with a key length of 256 bits; The generation submodule is used to group the archive information to be stored into 128-bit groups, perform AES encryption on each data group to generate encrypted data blocks, and obtain encrypted archive information; A verification submodule, configured to establish a secure transmission channel using the SSL protocol and perform encryption and integrity verification on the encrypted file information; The transmission submodule is used to divide the encrypted archive information into multiple data blocks and transmit them to the cloud platform using block transmission technology.

[0006] Furthermore, in the above-mentioned cloud computing-based archive information management system, the cloud platform computing module includes the following submodules: A decomposition submodule is used to perform a three-layer db4 wavelet decomposition on the image data in the encrypted archive information using wavelet transform to obtain low-frequency approximate coefficients and high-frequency detail coefficients of each layer; A deletion submodule is used to perform soft threshold processing on the high-frequency detail coefficients of each layer, delete the small coefficients corresponding to the noise, and obtain the processed coefficients; The reconstruction submodule is used to perform wavelet reconstruction using the processed coefficients to obtain denoised image data.

[0007] Furthermore, in the above-mentioned cloud computing-based archive information management system, the cloud platform computing module includes the following submodules: Establish a submodule for building a multi-head attention OCR engine through CNN convolutional neural network and multi-head attention mechanism; The extraction submodule is used to use ResNet-50 as the backbone network of the CNN convolutional neural network to extract features from the denoised image and generate image feature vectors; Set up a submodule to set up 8 attention heads, perform linear transformation on the image feature vector output by CNN to obtain the query, and perform linear transformation on the image feature sequence to obtain the key and value; The conversion submodule is used to convert the image into text data through the multi-head attention OCR engine to obtain processed file information.

[0008] Furthermore, in the above-mentioned cloud computing-based archive information management system, the classification model building module includes the following submodules: The GCN submodule is used in the model, which includes a two-layer GCN network. The first layer maps the initial node features to a low-dimensional space, and the second layer learns the topological feature representation of the nodes; The Bi-LSTM submodule is used in Bi-LSTM. The hidden layer dimension is set to 256. It contains two LSTM layers in two directions to capture the bidirectional semantic dependencies of the text and calculate the attention weight of the text sequence using the attention mechanism; The splicing submodule is used to splice the node topology features learned by GCN and the text semantic features obtained by combining Bi-LSTM and the attention mechanism, and input them into the fully connected layer for classification.

[0009] Furthermore, in the above-mentioned cloud computing-based archive information management system, the archive information classification module includes the following submodules: A division submodule, for dividing the processed archive information into a training set, a validation set, and a test set in a ratio of 7:2:1; Set up the submodule to use the Adam optimization algorithm as the optimization algorithm of the model, set the initial learning rate to 0.001, use the learning rate decay strategy, and use the cross entropy loss function as the loss function of the model; The submodule is used to input the processed archival information into the GCN-Bi-LSTM adaptive deep classification model for training to obtain the archival information confidence.

[0010] Furthermore, in the above-mentioned cloud computing-based archive information management system, the archive information management module includes the following submodules: The classification submodule is used to develop a multi-level classification system. The first level is classified by file type, the second level is classified by file formation time, and the third level is classified by file subject. The evaluation submodule is used to establish an information verification rule base, which includes at least integrity verification, consistency verification, and format verification, and to evaluate the quality of archival information by verifying the pass rate and error rate of archival information; The storage submodule is used to obtain multiple annotation results of the system, determine the target annotation result by using the majority voting method, and store the archive information according to the target annotation result.

[0011] Furthermore, in a method for implementing a cloud computing-based archival information management system, the method includes the following steps: Acquiring text information and image information of the archive to be stored to obtain archive information to be stored, encrypting the stored archive information using an AES encryption algorithm to obtain encrypted archive information, and transmitting the encrypted archive information to the cloud platform; Receive encrypted archival information, perform image noise reduction on the encrypted archival information using wavelet transform, convert the image into text data through a multi-head attention OCR engine, and obtain processed archival information; In the cloud platform, we use the GCN graph convolutional network to process the archival entity relationship graph, learn classification topology features, and use a combination of Bi-LSTM and attention mechanism to parse the archival text semantics, capture contextual dependencies, and establish a GCN-Bi-LSTM adaptive deep classification model. Inputting the processed archival information into the GCN-Bi-LSTM adaptive deep classification model for training to obtain archival information confidence; Archival information management is performed based on the archival information confidence level, including at least automatic classification and information verification.

[0012] Furthermore, in a method for implementing a cloud computing-based archival information management system, the method includes the following steps: Encrypting the stored archive information using an AES encryption algorithm with a key length of 256 bits; The archive information to be stored is grouped into 128-bit groups, and each data group is encrypted by AES to generate an encrypted data block to obtain the encrypted archive information; Use the SSL protocol to establish a secure transmission channel and encrypt and verify the integrity of the encrypted file information; The encrypted archive information is divided into multiple data blocks and transmitted to the cloud platform using block transmission technology.

[0013] Furthermore, in a method for implementing a cloud computing-based archival information management system, the method includes the following steps: Performing 3-layer db4 wavelet decomposition on the image data in the encrypted archive information using wavelet transform to obtain low-frequency approximate coefficients and high-frequency detail coefficients of each layer; Performing soft threshold processing on the high-frequency detail coefficients of each layer, deleting small coefficients corresponding to noise, and obtaining processed coefficients; The processed coefficients are used to perform wavelet reconstruction to obtain denoised image data.

[0014] Its beneficial effects are as follows: by acquiring the text information and image information of the archive to be stored, the archive information to be stored is obtained; the stored archive information is encrypted using the AES encryption algorithm to obtain encrypted archive information; the encrypted archive information is transmitted to the cloud platform; the encrypted archive information is received, the encrypted archive information is subjected to image noise reduction using wavelet transform, the image is converted into text data using a multi-head attention OCR engine, and the processed archive information is obtained; the GCN graph convolutional network is used in the cloud platform to process the archive entity relationship graph, learn classification topological features, and adopt a combination of Bi-LSTM and attention mechanism to parse the archive text semantics, capture contextual dependencies, and establish a GCN-Bi-LSTM adaptive deep classification model; the processed archive information is input into the GCN-Bi-LSTM adaptive deep classification model for training to obtain the archive information confidence; and archive information management is performed based on the archive information confidence, including at least automatic classification and information verification. 1. A strict data security protection system is established to eliminate the risk of archive information being stolen or tampered with during transmission and storage from the source, providing a solid guarantee for the security of archive data. 2. It accurately captures the complex connections and contextual dependencies between archives, enabling precise classification and in-depth mining of archival information, providing strong support for archive management and decision-making. 3. Automatic classification and information verification capabilities automate and intelligentize the archive management process. Automatic classification reduces manual intervention and improves classification efficiency; information verification promptly identifies data issues and ensures the quality of archival information; collaborative annotation promotes multi-user collaboration, improves annotation efficiency and accuracy, and comprehensively optimizes the archive management workflow. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Various other advantages and benefits will become apparent to those skilled in the art by reading the following detailed description of the preferred embodiment.The accompanying drawings are only for the purpose of illustrating the preferred embodiment and are not to be considered as limiting the present invention.

[0016] Figure 1 Schematic diagram of a first embodiment of a cloud computing-based archive information management system in an embodiment of the present invention; Figure 2 Schematic diagram of a second embodiment of a cloud computing-based archive information management system in an embodiment of the present invention; Figure 3 Schematic diagram of a third embodiment of a cloud computing-based archive information management system in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0018] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a", "an", "" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of features, integers, steps, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0019] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 As shown, a cloud computing-based archive information management system includes the following modules: An information acquisition and transmission module is used to acquire text information and image information of the archive to be stored, obtain archive information to be stored, encrypt the stored archive information using an AES encryption algorithm to obtain encrypted archive information, and transmit the encrypted archive information to the cloud platform; The cloud platform computing module is used to receive encrypted archival information, perform image noise reduction on the encrypted archival information using wavelet transform, and convert the image into text data through the multi-head attention OCR engine to obtain processed archival information; The classification model building module is used to process the archival entity relationship graph using the GCN graph convolutional network in the cloud platform, learn classification topology features, and use a combination of Bi-LSTM and attention mechanism to parse the archival text semantics, capture contextual dependencies, and build a GCN-Bi-LSTM adaptive deep classification model. An archival information classification module is used to input the processed archival information into the GCN-Bi-LSTM adaptive deep classification model for training to obtain the archival information confidence; The archival information management module is used to manage archival information according to the archival information confidence level, including at least automatic classification and information verification.

[0020] Specifically, the information acquisition and transmission module also includes an encryption submodule for encrypting the stored archive information using an AES encryption algorithm with a key length of 256 bits; The generation submodule is used to group the archive information to be stored into 128-bit groups, perform AES encryption on each data group to generate encrypted data blocks, and obtain encrypted archive information; The verification submodule is used to establish a secure transmission channel using the SSL protocol and to encrypt and verify the integrity of encrypted file information; The transmission submodule is used to use block transmission technology to divide the encrypted archive information into multiple data blocks and transmit them to the cloud platform.

[0021] Specifically: 1. Obtaining archival information: Diversified information collection channels: Paper archives: A high-speed scanner (Fujitsu fi-7900) equipped with an automatic document feeder supports batch scanning and a resolution of 300dpi is used to ensure clear images. During the scanning process, image pre-processing techniques (deskewing and black edge cropping) are used to optimize the original image. For bound archives, a non-contact scanner (Zeussler OS12000) is used to prevent damage. Electronic archives: Connect with office systems (OA systems) and business systems (finance systems, human resources systems) through API interfaces, supporting automatic extraction of multiple file formats (DOC, PDF, JPEG). Establish a file parsing module to identify text information (using text extraction libraries such as Python's PyPDF2 to extract PDF text) and image information (extracting embedded images from documents). Other sources: Reserved external device interfaces (camera, USB flash drive interface) to support on-site entry of temporary files. For example, real-time collection of file image information through high-speed cameras; Integration of archive information to be stored: Establish a unified archival information data structure, including metadata such as the archive unique identifier (generated by UUID), archive type (paper / electronic), collection time, and original file path. Encapsulate text and image information according to the data structure to form archival information objects to be stored.

[0022] 2. Encryption algorithm selection and encryption processing: Encryption algorithm selection: Use the AES (Advanced Encryption Standard) algorithm with a 256-bit key length to meet high security requirements. The AES algorithm has high encryption and decryption speed and strong anti-attack capabilities, making it suitable for batch encryption of archival information. Encryption process: Key management: AES keys are generated and managed using a key management system (KMS), and keys are regularly updated (once a month). Keys are stored in a secure key vault, and access control mechanisms restrict key usage. Data grouping: The archive information to be stored is grouped into 128-bit groups. The data less than 128 bits is filled using the PKCS#7 padding method. Encryption operation: Perform AES encryption on each data group to generate encrypted archive information data blocks.

[0023] 3. Encrypted file information transmission: Secure network channel: Use SSL / TLS protocol (version no less than 1.3) to establish a secure transmission channel, encrypt and verify the integrity of the transmitted data. Support HTTPS protocol for data transmission to ensure the security of data during network transmission; Transmission optimization: Using block transfer technology, large files are divided into multiple data blocks (each 1MB) for transmission, improving transmission efficiency. During the transmission process, the network status is monitored in real time. If a network anomaly occurs, the transmission is automatically paused and the breakpoint is recorded. After the network is restored, the transmission continues from the breakpoint.

[0024] Specifically, the cloud platform computing module also includes a decomposition submodule for performing a three-layer db4 wavelet decomposition on the image data in the encrypted archive information using wavelet transform to obtain low-frequency approximate coefficients and high-frequency detail coefficients of each layer; The deletion submodule is used to perform soft threshold processing on the high-frequency detail coefficients of each layer, delete the small coefficients corresponding to the noise, and obtain the processed coefficients; The reconstruction submodule is used to perform wavelet reconstruction using the processed coefficients to obtain the denoised image data; Establish a submodule for building a multi-head attention OCR engine through CNN convolutional neural network and multi-head attention mechanism; The extraction submodule is used to use ResNet-50 as the backbone network of the CNN convolutional neural network to extract features from the denoised image and generate image feature vectors; Set up a submodule to set up 8 attention heads, perform linear transformation on the image feature vector output by CNN to obtain the query, and perform linear transformation on the image feature sequence to obtain the key and value; The conversion submodule is used to convert images into text data through the multi-head attention OCR engine to obtain processed archival information.

[0025] Specifically: 1. Image noise reduction (wavelet transform): Wavelet basis function selection: The db4 wavelet basis function is selected, which has good symmetry and regularity and is suitable for image denoising. The support length of the db4 wavelet basis function is 8, which can well preserve the edge and detail information of the image while reducing noise; Wavelet transform parameter settings: Decomposition layers: set to 3. Multiple experiments have shown that 3-layer decomposition can achieve a good balance between noise reduction effect and computational efficiency. Threshold processing: Soft threshold processing method is used for the high-frequency coefficients obtained by decomposition of each layer; Noise reduction processing steps: Wavelet decomposition: Perform 3-layer db4 wavelet decomposition on the image data in the encrypted archive information to obtain low-frequency approximate coefficients and high-frequency detail coefficients of each layer; Threshold processing: Soft threshold processing is performed on the high-frequency detail coefficients of each layer to remove small coefficients corresponding to noise; Wavelet reconstruction: Use the processed coefficients to perform wavelet reconstruction to obtain the denoised image.

[0026] 2. Image to Text (Multi-Head Attention OCR Engine): OCR engine architecture: Adopting the architecture of CNN (convolutional neural network) + multi-head attention + RNN (recurrent neural network); CNN feature extraction: Using ResNet-50 as the backbone network, we extract features from the denoised image and generate image feature vectors. ResNet-50 has a deep network structure that can extract richer image features. Multi-head attention mechanism: 8 attention heads are set up, each of which independently calculates the query, key, and value. The query is obtained by linearly transforming the feature vector output by the CNN, and the key and value are obtained by linearly transforming the image feature sequence. Through the multi-head attention mechanism, the model can capture long-range dependencies between different regions in the image, improving the accuracy of OCR. RNN sequence prediction: A bidirectional LSTM (Bi-LSTM) is used as the sequence prediction layer. It receives the feature vectors output by the multi-head attention network, performs sequence modeling, and predicts text sequences. The Bi-LSTM can simultaneously capture both forward and backward semantic information of the text.

[0027] Training and Optimization: Training data: We used a public OCR dataset (ICDAR) and a self-built archival image dataset for training. We also performed data augmentation (rotation, scaling, and noise addition) on the archival image data to improve the generalization ability of the model. Loss function: Connectionist Temporal Classification Loss (CTCLoss) is used to handle the problem of text sequence prediction of variable length.

[0028] Specifically, the classification model building module also includes a GCN submodule, which is used for a model including a two-layer GCN network. The first layer maps the initial features of the nodes to a low-dimensional space, and the second layer learns the topological feature representation of the nodes; The Bi-LSTM submodule is used in Bi-LSTM. The hidden layer dimension is set to 256. It contains two LSTM layers in two directions to capture the bidirectional semantic dependencies of the text and calculate the attention weight of the text sequence using the attention mechanism; The splicing submodule is used to splice the node topology features learned by GCN and the text semantic features obtained by combining Bi-LSTM and the attention mechanism, and input them into the fully connected layer for classification.

[0029] Specifically: (1) Construction of the archival entity relationship graph: 1. Entity and relationship definition: Entity: includes various entities in the archive (such as the names of people and organizations identified above). Each entity has a unique identifier and attributes (such as the gender of a person or the industry to which an organization belongs). Relationships: Define the relationships between entities, such as "belonging" (a person's name belongs to an organization's name), "chronological order" (event A occurs before event B), and "association" (the association between files). Relationships have a type and a weight, with the weight determined by the strength of the association between the entities (calculated by the number of co-occurrences). 2. Map construction method: Use a graph database (Neo4j) to store entity-relationship graphs. Build a directed weighted graph with entities as nodes and relationships as edges. Node and edge attributes are stored in a defined format. (2) GCN graph convolutional network: Network structure: A two-layer GCN network is used. The first layer maps the initial node features (entity type encoding, attribute encoding) to a low-dimensional space, and the second layer learns the topological feature representation of the node. (3) Combination mechanism of Bi-LSTM and attention mechanism: 1. Bi-LSTM structure: The hidden layer dimension is set to 256 and contains two directional LSTM layers (forward and backward), which can capture the bidirectional semantic dependencies of the text; 2. Attention Mechanism: Calculate the attention weights for a text sequence: (IV) Model Fusion The node topology features learned by GCN and the text semantic features obtained by combining Bi-LSTM and the attention mechanism are spliced and input into the fully connected layer for classification to obtain the confidence of the archival information.

[0030] Specifically, the archive information classification module also includes a division submodule for dividing the processed archive information into a training set, a validation set, and a test set in a ratio of 7:2:1; Set up the submodule to use the Adam optimization algorithm as the optimization algorithm of the model, set the initial learning rate to 0.001, use the learning rate decay strategy, and use the cross entropy loss function as the loss function of the model; This submodule is used to input the processed archival information into the GCN-Bi-LSTM adaptive deep classification model for training to obtain the archival information confidence.

[0031] Specifically: GCN layer (graph convolutional network): processes the archival entity relationship graph and learns the topological features between nodes; Input design: Archival entities (persons, institutions, events) are used as graph nodes, and relationships between entities (subordinate, temporal sequence, association) are used as edges. Node features include entity type codes, attribute vectors, etc. Network structure: It uses a two-layer GCN structure. The first layer maps the initial node features to a low-dimensional space, and the second layer learns the topological feature representation of the nodes. Each GCN layer includes graph convolution operations, nonlinear activation functions (ReLU), and batch normalization operations. Output features: Generate a topological feature vector for each entity node to capture the complex relationship network between entities; Bi-LSTM combined with attention mechanism layer: processes archival text sequences and captures semantic dependencies; Input design: Convert the archive text (abstract, text) into a word vector sequence as input; Bidirectional LSTM: Captures bidirectional contextual information of text, sets the hidden layer dimension to 256, and can learn long-distance dependencies; Attention mechanism: Calculates the attention weight for each position in the text sequence, automatically focusing on keywords and phrases important for classification, and enhancing the model's sensitivity to key information; Output features: Generate semantic representation vectors of text to capture the deep semantic information of the text; Adaptive fusion layer: dynamically integrates the features of GCN and BiLSTM; Feature splicing: Splice the topological feature vector generated by GCN and the semantic feature vector generated by combining Bi-LSTM and attention mechanism; Attention weight learning: The importance weights of the two features are learned through a small neural network. The network takes the concatenated features as input and outputs two weight values, corresponding to the importance of the GCN feature and the BiLSTM feature respectively. Weighted fusion: Based on the learned weights, the two features are weightedly fused to generate the final feature representation.

[0032] 1. Training data preparation: The training data preparation process includes: Feature extraction: Entity features: Extract entity types, attributes, and other features from associated archival information and represent them in vector form. For example, for a person entity, features such as name, position, and department can be extracted; for an organization entity, features such as name, industry, and size can be extracted. Relationship features: Construct an adjacency matrix of the entity relationship graph, where each edge contains relationship type and weight information. Relationship weight can be calculated based on the number of co-occurrences between entities, the strength of the association, etc. Text features: Perform word segmentation and word embedding operations on the archival text and convert it into a fixed-dimensional word vector sequence; Data structure conversion: Convert the extracted entity features and relationship information into a graph data structure suitable for GCN processing. Pad and truncate the text sequence to make it of the same length for batch processing. Construct a label vector to convert the classification label of the file into a numerical form. Dataset division: The preprocessed dataset is divided into training set, validation set and test set in a ratio of 7:2:1; A stratified sampling strategy is adopted to ensure that the distribution ratio of samples of each category in each set is relatively balanced to avoid the problem of category imbalance.

[0033] 2. Model training process: The training process adopts a multi-stage optimization strategy: Optimization goal: Loss function: The cross entropy loss function is used to measure the difference between the model prediction result and the true label; Regularization: L2 regularization and Dropout technology are introduced to prevent model overfitting and improve generalization ability; Early stopping strategy: Monitor the validation set loss. When the validation set loss stops decreasing for 10 consecutive epochs, stop training and save the best model parameters. Training strategy: Batch training: Use mini-batch gradient descent, processing 32 samples at a time, to balance training efficiency and model stability; Optimizer: The Adam optimizer is used, the initial learning rate is set to 0.001, and the learning rate decay strategy is used. When the validation set loss reaches plateau, the learning rate is reduced to 0.5 times the original value. Training rounds: Set the maximum number of training rounds to 100, and determine the actual number of training rounds based on the early stopping strategy; Training Monitoring: After each round of training, the model performance is evaluated on the validation set, and indicators such as accuracy, precision, and recall are recorded; Visualize the training process and plot the curves of training loss, validation loss, and accuracy as the training rounds change, making it easier to analyze the model training status; 3. Confidence calculation and application: The confidence level of the model output has the following characteristics: Confidence calculation: The final output of the model is converted into a probability distribution through the softmax function. The probability value of each category represents the confidence that the file belongs to that category. The category with the highest confidence is the prediction result of the model, and its corresponding probability value is the credibility of the prediction; Adaptive mechanism: The adaptive fusion layer in the model can dynamically adjust the importance weights of GCN features and BiLSTM features according to the characteristics of the input file; For archives with complex relationships, the model automatically assigns higher weights to GCN features; for archives with rich text content, it assigns higher weights to BiLSTM features.

[0034] Specifically, the archive information management module also includes a classification submodule for developing a multi-level classification system, with the first level being based on archive type, the second level being based on archive formation time, and the third level being based on archive subject; The evaluation submodule is used to establish an information verification rule base, which includes at least integrity verification, consistency verification, and format verification, and to evaluate the quality of archival information by verifying the pass rate and error rate of archival information; The storage submodule is used to obtain multiple annotation results of the system, determine the target annotation result using the majority voting method, and store the archive information according to the target annotation result.

[0035] Specifically: 1. Automatic classification: 1. Classification criteria: Develop a multi-level classification system. The first level is based on file type (personnel, finance, documents), the second level is based on file creation time (year, quarter), and the third level is based on file subject (recruitment, assessment, budget). The classification standards can be dynamically adjusted according to user needs. 2. Classification method: Based on the confidence level of the archive information output by the model, the category with the highest confidence level is selected as the archive classification result. For multi-label classification problems, a confidence threshold (0.5) is set, and labels above the threshold are regarded as valid classification labels for archives.

[0036] 2. Information Verification: 1. Inspection rules: Establish an information verification rule base, including completeness verification (whether required fields are missing), consistency verification (whether entity relationships are contradictory), format verification (whether the date format is correct), etc. The rule base can be expanded through user customization; 2. Inspection indicators: Calculate indicators such as the inspection pass rate (number of files that passed the inspection / total number of files) and the error rate (number of files with errors / total number of files) to assess the quality of the file information. For files that fail the inspection, an error report is generated to prompt the user to make corrections.

[0037] 3. Collaborative annotation: 1. Labeling process: Support multiple annotators to annotate the same file information at the same time. Annotators log in to the annotation platform, receive annotation tasks, and submit the annotation results after completion. The system automatically collects the annotation results of multiple annotators and uses the majority voting method (or weighted voting method, setting weights according to the credibility of the annotators) to determine the final annotation result; 2. Collaboration Mechanism: Establish a communication platform for annotators (with built-in chat function) to facilitate communication between annotators and discuss issues encountered during the annotation process. Regularly train and evaluate annotators to improve annotation quality and efficiency. Reward those who accurately annotate to encourage their active participation in the annotation process.

[0038] Its beneficial effects are as follows: by acquiring the text and image information of the archive to be stored, the archive information to be stored is obtained; the stored archive information is encrypted using the AES encryption algorithm to obtain encrypted archive information; the encrypted archive information is received, image noise reduction is performed on the encrypted archive information using wavelet transform, and the image is converted into text data using a multi-head attention OCR engine to obtain processed archive information; in the cloud platform, the GCN graph convolutional network is used to process the archive entity relationship graph, learn classification topological features, and use a combination of Bi-LSTM and attention mechanism to parse the archive text semantics, capture contextual dependencies, and establish a GCN-Bi-LSTM adaptive deep classification model; the processed archive information is input into the GCN-Bi-LSTM adaptive deep classification model for training to obtain the archive information confidence; and archive information management is performed based on the archive information confidence, including at least automatic classification and information verification. 1. Build a strict data security protection system to eliminate the risk of archive information being stolen or tampered with during transmission and storage from the source, providing a solid guarantee for archive data security. 2. It accurately captures the complex connections and contextual dependencies between archives, enabling precise classification and in-depth mining of archival information, providing strong support for archive management and decision-making. 3. Automatic classification and information verification capabilities automate and intelligentize the archive management process. Automatic classification reduces manual intervention and improves classification efficiency; information verification promptly identifies data issues and ensures the quality of archival information; collaborative annotation promotes multi-user collaboration, improves annotation efficiency and accuracy, and comprehensively optimizes the archive management workflow.

[0039] See also Figure 2In a cloud computing-based archive information management system, the information acquisition and transmission module includes the following submodules: The encryption submodule is used to encrypt the stored archive information using the AES encryption algorithm with a key length of 256 bits; The generation submodule is used to group the archive information to be stored into 128-bit groups, perform AES encryption on each data group to generate encrypted data blocks, and obtain encrypted archive information; The verification submodule is used to establish a secure transmission channel using the SSL protocol and to encrypt and verify the integrity of encrypted file information; The transmission submodule is used to use block transmission technology to divide the encrypted archive information into multiple data blocks and transmit them to the cloud platform.

[0040] See also Figure 3 In a cloud computing-based archive information management system, the cloud platform computing module includes the following submodules: Establish a submodule for building a multi-head attention OCR engine through CNN convolutional neural network and multi-head attention mechanism; The extraction submodule is used to use ResNet-50 as the backbone network of the CNN convolutional neural network to extract features from the denoised image and generate image feature vectors; Set up a submodule to set up 8 attention heads, perform linear transformation on the image feature vector output by CNN to obtain the query, and perform linear transformation on the image feature sequence to obtain the key and value; The conversion submodule is used to convert images into text data through the multi-head attention OCR engine to obtain processed archival information.

[0041] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A cloud computing-based archive information management system, characterized in that: The archive information management system includes the following modules: An information acquisition and transmission module is used to acquire text information and image information of the archive to be stored, obtain archive information to be stored, encrypt the stored archive information using an AES encryption algorithm to obtain encrypted archive information, and transmit the encrypted archive information to the cloud platform; The cloud platform computing module is used to receive encrypted archival information, perform image noise reduction on the encrypted archival information using wavelet transform, and convert the image into text data through the multi-head attention OCR engine to obtain processed archival information; The classification model building module is used to process the archival entity relationship graph using the GCN graph convolutional network in the cloud platform, learn classification topology features, and use a combination of Bi-LSTM and attention mechanism to parse the archival text semantics, capture contextual dependencies, and build a GCN-Bi-LSTM adaptive deep classification model. An archival information classification module is used to input the processed archival information into the GCN-Bi-LSTM adaptive deep classification model for training to obtain the archival information confidence; The archival information management module is used to manage archival information according to the archival information confidence level, including at least automatic classification and information verification.

2. The cloud computing-based archive information management system according to claim 1, characterized in that: The information acquisition and transmission module includes the following submodules: An encryption submodule, configured to encrypt the stored archive information using an AES encryption algorithm with a key length of 256 bits; The generation submodule is used to group the archive information to be stored into 128-bit groups, perform AES encryption on each data group to generate encrypted data blocks, and obtain encrypted archive information; A verification submodule, configured to establish a secure transmission channel using the SSL protocol and perform encryption and integrity verification on the encrypted file information; The transmission submodule is used to divide the encrypted archive information into multiple data blocks and transmit them to the cloud platform using block transmission technology.

3. The cloud computing-based archive information management system according to claim 1, characterized in that: The cloud platform computing module includes the following submodules: A decomposition submodule is used to perform a three-layer db4 wavelet decomposition on the image data in the encrypted archive information using wavelet transform to obtain low-frequency approximate coefficients and high-frequency detail coefficients of each layer; A deletion submodule is used to perform soft threshold processing on the high-frequency detail coefficients of each layer, delete the small coefficients corresponding to the noise, and obtain the processed coefficients; The reconstruction submodule is used to perform wavelet reconstruction using the processed coefficients to obtain denoised image data.

4. The cloud computing-based archive information management system according to claim 1, wherein: The cloud platform computing module also includes the following submodules: Establish a submodule for building a multi-head attention OCR engine through CNN convolutional neural network and multi-head attention mechanism; The extraction submodule is used to use ResNet-50 as the backbone network of the CNN convolutional neural network to extract features from the denoised image and generate image feature vectors; Set up a submodule to set up 8 attention heads, perform linear transformation on the image feature vector output by CNN to obtain the query, and perform linear transformation on the image feature sequence to obtain the key and value; The conversion submodule is used to convert the image into text data through the multi-head attention OCR engine to obtain processed file information.

5. The cloud computing-based archive information management system according to claim 1, wherein: The classification model building module includes the following submodules: The GCN submodule is used in the model, which includes a two-layer GCN network. The first layer maps the initial node features to a low-dimensional space, and the second layer learns the topological feature representation of the nodes; The Bi-LSTM submodule is used in Bi-LSTM. The hidden layer dimension is set to 256. It contains two LSTM layers in two directions to capture the bidirectional semantic dependencies of the text and calculate the attention weight of the text sequence using the attention mechanism; The splicing submodule is used to splice the node topology features learned by GCN and the text semantic features obtained by combining Bi-LSTM and the attention mechanism, and input them into the fully connected layer for classification.

6. The cloud computing-based archive information management system according to claim 1, characterized in that: The archive information classification module includes the following submodules: A division submodule, for dividing the processed archive information into a training set, a validation set, and a test set in a ratio of 7:2:1; Set up the submodule to use the Adam optimization algorithm as the optimization algorithm of the model, set the initial learning rate to 0.001, use the learning rate decay strategy, and use the cross entropy loss function as the loss function of the model; The submodule is used to input the processed archival information into the GCN-Bi-LSTM adaptive deep classification model for training to obtain the archival information confidence.

7. The cloud computing-based archive information management system according to claim 1, characterized in that: The archive information management module includes the following submodules: The classification submodule is used to develop a multi-level classification system. The first level is classified by file type, the second level is classified by file formation time, and the third level is classified by file subject. The evaluation submodule is used to establish an information verification rule base, which includes at least integrity verification, consistency verification, and format verification, and to evaluate the quality of archival information by verifying the pass rate and error rate of archival information; The storage submodule is used to obtain multiple annotation results of the system, determine the target annotation result by using the majority voting method, and store the archive information according to the target annotation result.

8. A method for implementing a cloud computing-based archive information management system as claimed in claim 1, characterized in that: The method comprises the following steps: Acquiring text information and image information of the archive to be stored to obtain archive information to be stored, encrypting the stored archive information using an AES encryption algorithm to obtain encrypted archive information, and transmitting the encrypted archive information to the cloud platform; Receive encrypted archival information, perform image noise reduction on the encrypted archival information using wavelet transform, convert the image into text data through a multi-head attention OCR engine, and obtain processed archival information; In the cloud platform, we use the GCN graph convolutional network to process the archival entity relationship graph, learn classification topology features, and use a combination of Bi-LSTM and attention mechanism to parse the archival text semantics, capture contextual dependencies, and establish a GCN-Bi-LSTM adaptive deep classification model. Inputting the processed archival information into the GCN-Bi-LSTM adaptive deep classification model for training to obtain archival information confidence; Archival information management is performed based on the archival information confidence level, including at least automatic classification and information verification.

9. A method for implementing a cloud computing-based archive information management system as claimed in claim 1, characterized in that: The method comprises the following steps: Encrypting the stored archive information using an AES encryption algorithm with a key length of 256 bits; The archive information to be stored is grouped into 128-bit groups, and each data group is encrypted by AES to generate an encrypted data block to obtain the encrypted archive information; Use the SSL protocol to establish a secure transmission channel and encrypt and verify the integrity of the encrypted file information; The encrypted archive information is divided into multiple data blocks and transmitted to the cloud platform using block transmission technology.

10. A method for implementing a cloud computing-based archive information management system as claimed in claim 1, characterized in that: The method comprises the following steps: Performing 3-layer db4 wavelet decomposition on the image data in the encrypted archive information using wavelet transform to obtain low-frequency approximate coefficients and high-frequency detail coefficients of each layer; Performing soft threshold processing on the high-frequency detail coefficients of each layer, deleting small coefficients corresponding to noise, and obtaining processed coefficients; The processed coefficients are used to perform wavelet reconstruction to obtain denoised image data.

Images