A method and system for managing electronic archives

Abstract

The application relates to the technical field of electronic file management, in particular to a management method and system for electronic files, which comprises the following steps: obtaining a paper file image, performing character recognition on the paper file image based on an image processing technology to obtain a first target text; receiving electronic files uploaded by a plurality of preset data sources every preset sampling period, converting the electronic files into a second target text, and preprocessing the first target text and the second target text to obtain a target text. The paper file is converted into an electronic text through the image processing technology and the character recognition technology, the efficiency of converting the file from paper to an electronic file is greatly improved, the workload of manual input is reduced, and the electronic files from the plurality of data sources can be automatically received and processed at regular intervals through the preset sampling period, so that manual intervention can be avoided, and the real-time performance and comprehensiveness of file management are ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electronic archives management, and particularly relates to a management method and system of electronic archives. BACKGROUND

[0002] Electronic archives refer to archives data saved and managed through digital means, which exist in the form of electronic files and can be stored, managed, retrieved, transmitted and backed up through computer systems, storage devices or networks.

[0003] Traditional methods usually rely on manual input and sorting of paper archives, which is not only time-consuming and laborious, but also prone to errors. The manual input process is time-consuming, and information errors may occur due to human negligence. Traditional methods usually use fixed storage methods and lack a mechanism for dynamically allocating storage resources. The storage location of files is usually static, and cannot be optimized according to the importance of the files, resulting in low resource utilization efficiency. Traditional methods lack an intelligent file importance evaluation mechanism, and the storage and management of documents often rely on manual judgment without clear standards to ensure that important files are stored and protected first. Therefore, critical files may not be managed or backed up in a timely manner, increasing the risk of data loss. SUMMARY

[0004] The technical problem to be solved by the present application is to overcome the shortcomings of the prior art and provide a management method and system of electronic archives.

[0005] The technical solution adopted to solve the above technical problems is as follows: a management method of electronic archives, comprising:

[0006] Obtaining a paper archive image, performing text recognition on the paper archive image based on image processing technology to obtain a first target text;

[0007] Receiving electronic archives uploaded by a plurality of preset data sources every preset sampling period, converting the electronic documents into a second target text, and preprocessing the first target text and the second target text to obtain a target text;

[0008] Performing a word segmentation operation on the target text to obtain a word sequence corresponding to the target text, and classifying the target text based on the word sequence to obtain a text type label corresponding to the target text;

[0009] Obtaining the importance of the target text based on the text type label corresponding to the target text, and matching the target text with a plurality of preset text storage nodes based on the importance of the target text to obtain a text storage node matched with the target text;

[0010] store the target text to the matched text storage node.

[0011] Preferably, the image processing technology is used to recognize the text of the paper file image to obtain the first target text, including:

[0012] The paper file image is enhanced to obtain a paper file enhanced image.

[0013] The first target detection model is used to detect the text box of the paper file enhanced image to obtain a plurality of first text boxes corresponding to the paper file enhanced image, and the plurality of first text boxes are segmented by column to obtain a text box column set.

[0014] The second target detection model is used to detect the text box of the text box column set to obtain a plurality of second text boxes corresponding to the text box column set, and the plurality of second text boxes are segmented by row to obtain a text box row set.

[0015] The text recognition model is used to recognize the text of the text box row set to obtain a text recognition result corresponding to the text box row set.

[0016] The text recognition result is sorted to obtain the first target text.

[0017] Preferably, the text recognition model includes a local feature extraction module, a global feature extraction module, a weighted aggregation module, and a classification module, the local feature extraction module is used to extract the local image features of the text box row set, the global feature extraction module is used to extract the global image features of the text box row set, the weighted aggregation module is used to aggregate the local image features and the global image features to obtain aggregated image features, and the classification module is used to classify the aggregated image features to obtain the text recognition result corresponding to the text box row set.

[0018] Preferably, the local feature extraction module uses a depth separable convolution to extract image features, the global feature extraction module includes a window multi-head attention layer, a sliding window multi-head attention layer, and an MLP layer, each MLP layer is connected with the window multi-head attention layer and the sliding window multi-head attention layer before using a normalization layer, and each layer uses a residual connection, the weighted aggregation module combines the local image features and the global image features to obtain aggregated image features through a GCN, and the classification module classifies the aggregated image features through a fully connected layer and a Softmax function to obtain the text recognition result corresponding to the text box row set.

[0019] Preferably, the target text is classified based on the word sequence to obtain a text type label corresponding to the target text, comprising:

[0020] The context semantic features and the word embedding vectors of the word sequence are obtained based on the trained pre-trained model;

[0021] The similarity between each word in the word sequence is calculated based on the word embedding vectors of the word sequence to obtain the similarity between each word in the word sequence, and if the similarity is greater than a preset similarity threshold, it indicates that there is a similar relationship between the two words;

[0022] A text directed graph is constructed based on the word sequence and the similar relationship between the words, and the text directed graph is aggregated based on a graph convolution operation to obtain text hidden features of the word sequence;

[0023] The context semantic features and the text hidden features of the word sequence are fused to obtain fusion features, and the fusion features are classified to obtain a text type label corresponding to the target text.

[0024] Preferably, the pre-trained model uses BERT, and the text directed graph includes a plurality of nodes and a plurality of directed edges, wherein the nodes correspond to the words in the word sequence, and the directed edges correspond to the similar relationship between the two words.

[0025] Preferably, the text directed graph is aggregated based on a graph convolution operation to obtain text hidden features of the word sequence, comprising:

[0026] The text directed graph is embedded and represented based on a graph convolution operation to obtain node representation, wherein the formula of the node representation is as follows:

[0027] ;

[0028] Wherein, represents the hidden representation of the i-th layer node of the text directed graph, represents an activation function, represents an adjacency matrix of the text directed graph of the list representation type, represents the hidden representation of the i-th layer node of the text directed graph of the list representation type, represents the weight of the i-th layer node of the text directed graph of the list representation type, represents different types of nodes. ​​​​​​

[0029] The type-level attention score of the adjacent node is calculated based on the node representation of each node in the text directed graph, and the calculation formula of the type-level attention score is as follows:

[0030] ;

[0031] wherein, represents the type-level attention score of the adjacent node, represents the attention vector of the type , represents a splicing operation, represents the hidden representation of the current node, represents the embedding representation of the adjacent node of the type , , represents the adjacent node, represents the embedding representation of the directed edge between the current node and the adjacent node, represents the hidden representation of the adjacent node.

[0032] The type-level attention weight of the adjacent node is obtained based on the type-level attention score of the adjacent node, and the calculation formula of the type-level attention weight is as follows:

[0033] ;

[0034] wherein, represents the type-level attention weight;

[0035] The node-level attention score of the adjacent node is calculated based on the type-level attention weight of the adjacent node, and the calculation formula of the node-level attention score is as follows:

[0036] ;

[0037] wherein, represents the type-level attention weight of the adjacent node, represents the attention vector.

[0038] The dual-level attention mechanism of the type-level and the node-level attention is fused to obtain the text hidden feature of the word sequence.

[0039] Preferably, the target text is matched with a plurality of preset text storage nodes based on the importance of the target text to obtain a text storage node matched with the target text, including:

[0040] The behavior score of the plurality of text storage nodes is obtained based on the behavior transaction of the plurality of text storage nodes.

[0041] The behavior score of the plurality of text storage nodes is compared with a preset behavior score threshold, and if the behavior score of the text storage node is less than the preset behavior score threshold, the text storage node is eliminated from the plurality of text storage nodes;

[0042] The remaining text storage nodes in the plurality of text storage nodes are sorted based on the behavior score to obtain a candidate text storage node order set;

[0043] The candidate text storage node order set is divided into a preset number of candidate text storage node order sub-sets, and the target text is matched to a candidate text storage node order sub-set based on the importance of the target text;

[0044] Based on the remaining amount of computing resources of each candidate text storage node in the candidate text storage node order sub-set, the target text is matched to the candidate text storage node with the largest remaining amount of computing resources.

[0045] Preferably, the calculation formula of the behavior score is as follows:

[0046] ;

[0047] Wherein, represents the behavior score of the i-th text storage node in the plurality of text storage nodes, represents the transaction sequence of the i-th text storage node in the plurality of text storage nodes, represents whether the transaction participated by the i-th text storage node in the plurality of text storage nodes is legal, if legal, then , otherwise, , represents whether the transaction participated by the i-th text storage node in the plurality of text storage nodes is illegal, if illegal, then , otherwise, , represents the preset penalty coefficient. The technical scheme adopted to solve the above technical problems is: an electronic archive management system, which is suitable for the electronic archive management method and comprises: A character recognition unit is configured to acquire a paper archive image, perform character recognition on the paper archive image based on image processing technology, and obtain a first target text.

[0048] The technical scheme adopted to solve the above technical problems is: an electronic archive management system, which is suitable for the electronic archive management method and comprises:

[0049] A character recognition unit is configured to acquire a paper archive image, perform character recognition on the paper archive image based on image processing technology, and obtain a first target text.

[0050] ​​The text processing unit is used for receiving electronic archives uploaded by a plurality of preset data sources every preset sampling period, converting the electronic archives into second target texts, and preprocessing the first target texts and the second target texts to obtain target texts;

[0051] The text classification unit is used for performing a word segmentation operation on the target texts to obtain a word sequence corresponding to the target texts, classifying the target texts based on the word sequence to obtain a text type label corresponding to the target texts;

[0052] The storage matching unit is used for obtaining the importance of the target texts based on the text type label corresponding to the target texts, and matching the target texts with a plurality of preset text storage nodes based on the importance of the target texts to obtain a text storage node matched with the target texts;

[0053] The text storage unit is used for storing the target texts into the matched text storage node.

[0054] The present application has the following advantages: (1) The present application converts paper archives into electronic texts through image processing technology and character recognition technology, greatly improves the efficiency of converting archives from paper to electronic archives, reduces the workload of manual input, and can automatically receive and process electronic archives from a plurality of data sources at regular intervals by setting a preset sampling period, which can avoid manual intervention, ensure the real-time and comprehensiveness of archive management, and effectively manage the documents stored in different nodes through matching, reduce the risk of data loss; (2) The present application eliminates errors or inconsistencies that may occur during the conversion of source files by recognizing the text of paper archive images and preprocessing the first target text and the second target text, making the final target text more accurate and reliable, and processing the text based on word segmentation technology and identifying the type of the text through classification can help automatically label and classify archives, thereby reducing the errors of manual classification and improving the accuracy and consistency of archive classification; (3) The present application automatically evaluates the importance of the text according to the type label, which can make the archive management system more flexible, determine the storage strategy according to the importance of the document, ensure that the key document is stored and managed in priority, and match the importance of the target text with the characteristics of the preset storage node, which can dynamically allocate storage resources according to the priority of the document, optimize the utilization rate of storage space and the efficiency of data retrieval. BRIEF DESCRIPTION OF DRAWINGS

[0055] Figure 1A schematic diagram of a step flow of the overall method according to an embodiment of the present application is shown in the figure;

[0056] Figure 2 A schematic diagram of the system architecture of the overall system according to an embodiment of the present application is shown in the figure.

[0057] Reference signs: 1, character recognition unit; 2, text processing unit; 3, text classification unit; 4, storage matching unit; 5, text storage unit. DETAILED DESCRIPTION

[0058] Embodiment one, as shown in the figure, the present application provides a kind of electronic archives management method, comprising: Figure 1

[0059] S1, obtain paper archives image, based on image processing technique to paper archives image carries out character recognition, to obtain first target text;

[0060] S2, every interval preset sampling period receives the electronic archives uploaded by multiple preset data sources, converts electronic document into second target text, pre-processes first target text and second target text, to obtain target text;

[0061] S3, target text is carried out word segmentation operation, to obtain the word sequence corresponding to target text, based on word sequence target text is classified, to obtain the text type label corresponding to target text;

[0062] S4, based on the importance of target text corresponding to the text type label of target text, based on the importance of target text, target text is matched with multiple preset text storage nodes, to obtain the text storage node matched with target text;

[0063] S5, target text is stored to the text storage node matched.

[0064] ​In the present application, paper archive image refers to the conversion of paper documents into digital image format through scanning or photographing and other means; electronic archive refers to the archive files stored in electronic file format (such as PDF, Word document, Excel table, etc.), which are usually directly generated digital documents rather than scanned images; text preprocessing refers to the cleaning and normalization of the original text before further analysis of the text; word segmentation is a very important step in text preprocessing, especially for languages without space separation such as Chinese, word segmentation operation is to divide continuous text string into individual words; word sequence refers to a series of ordered word sets obtained by word segmentation, each word usually carries certain semantic information, through the analysis of word sequence, the meaning and structure of the text can be understood; text classification is a task in natural language processing, the purpose is to classify the text into one or more predefined categories according to the content of the text, for example, classify the article into news, technology, finance and other categories according to the content of the article; text type label is the result of text classification, which is the category label to which the target text is classified, for example, the target text may be classified into contract, report, invoice and other types, each category corresponds to a label; text storage node refers to the physical or virtual storage location of text data, each node may have different storage strategies, permissions and security, etc., according to the importance of the text, the target text will be allocated to the corresponding storage node; text storage refers to saving text data (such as target text) in a storage medium, usually in database, file system or cloud storage, etc.

[0065] In embodiment two, the present application proposes a management method of electronic archives, compared with embodiment one, the present embodiment further includes: performing character recognition on the paper archive image based on image processing technology to obtain the first target text, including:

[0066] A1, image enhancement is performed on the paper archive image to obtain a paper archive enhanced image;

[0067] A2, based on the trained first target detection model, text box detection is performed on the paper archive enhanced image to obtain a plurality of first text boxes corresponding to the paper archive enhanced image, and the plurality of first text boxes are segmented by column to obtain a text box column set;

[0068] A3, based on the trained second target detection model, text box detection is performed on the text box column set to obtain a plurality of second text boxes corresponding to the text box column set, and the plurality of second text boxes are segmented by row to obtain a text box row set;

[0069] A4, based on the trained character recognition model, character recognition is performed on the text box row set to obtain a character recognition result corresponding to the text box row set;

[0070] A5, sort the text recognition result to obtain the first target text.

[0071] In this embodiment, image enhancement is a process of improving image quality through a series of processing means (such as contrast adjustment, noise removal, sharpening, etc.). For paper archive images, the purpose of image enhancement is to make the text and details clearer, which is convenient for subsequent text recognition and detection. For example, during scanning or shooting, the image may be affected by blur, uneven light or noise. Image enhancement technology can optimize the visual effect of the image and improve the recognition accuracy; target detection is a task in computer vision, which aims to locate and identify target objects or regions in an image. In this process, the purpose of target detection is to identify the text box in the image. The target detection model usually labels all possible text areas in the image, i.e. the text box; the first target detection model is used to detect possible text box areas in the enhanced paper archive image. It identifies all possible positions of text in the image and outputs a series of "text box" coordinates; the second target detection model is used to further detect the text box column set. It identifies each line of text in the text box column and further divides the specific position of each line of text; the text box column set refers to the grouping result of multiple text boxes in columns (vertically) obtained by the first target detection model; in the further detection process, the second target detection model divides the text in the text box column by rows (horizontally) to form a text box row set; the text result after text recognition is usually output in a certain order. However, there may be situations such as mixed order of text and character recognition errors in the recognition process, so the recognition result needs to be sorted to ensure the correct order of the text. The sorting process may include reordering the recognized text according to the spatial position of the text box (from top to bottom, from left to right), according to the row and column order.

[0072] In an optional embodiment, the text recognition model includes a local feature extraction module, a global feature extraction module, a weighted aggregation module, and a classification module. The local feature extraction module is used to extract local image features of the text box row set, the global feature extraction module is used to extract global image features of the text box row set, the weighted aggregation module is used to aggregate the local image features and the global image features to obtain aggregated image features, and the classification module is used to classify the aggregated image features to obtain the text recognition result corresponding to the text box row set.

[0073] In an optional embodiment, the local feature extraction module extracts image features using a depth separable convolution, the global feature extraction module includes a window multi-head attention layer, a sliding window multi-head attention layer, and an MLP layer, each MLP layer is connected with the window multi-head attention layer and the sliding window multi-head attention layer before using a normalization layer, and each layer uses a residual connection, the weighted aggregation module combines the local image features and the global image features to obtain aggregated image features through the GCN, and the classification module classifies the aggregated image features through a fully connected layer and a Softmax function to obtain the character recognition result corresponding to the text box row set.

[0074] It should be noted that the role of the local feature extraction module is to extract the features of the local region from the image, especially for each row of text in the text box row set, the local feature usually refers to the detailed information in the image, such as the strokes and shapes of the characters, and the local feature extraction is performed through a convolutional neural network (CNN) or other deep learning technology, which aims to capture the local structural features of each character or character group; the purpose of the global feature extraction module is to extract global features from the entire image or the entire text box region, i.e. more extensive information, the global features contain more extensive context information in the image, not limited to the features of a single character, but also including the arrangement of characters, line spacing, and layout of text, etc. The global features are helpful for recognition in more complex situations, especially in the case of similar or ambiguous character shapes, the overall context can provide more clues.

[0075] In an optional embodiment, the target text is classified based on the word sequence to obtain a text type label corresponding to the target text, including:

[0076] B1, obtaining the context semantic features and word embedding vectors of the word sequence based on the trained pre-trained model;

[0077] B2, calculating the similarity between each word in the word sequence based on the word embedding vectors of the word sequence to obtain the similarity between each word in the word sequence, and if the similarity is greater than a preset similarity threshold, it indicates that there is a similar relationship between the two words;

[0078] B3, constructing a text directed graph based on the word sequence and the similar relationship between the words, and performing graph convolution operation to aggregate the text directed graph to obtain text hidden features of the word sequence;

[0079] B4, fusing the context semantic features and the text hidden features of the word sequence to obtain fusion features, and classifying the fusion features to obtain a text type label corresponding to the target text.

[0080] It should be noted that pre-trained models refer to models that have been trained on large-scale datasets and are able to learn rich language features and representations. These models are usually deep learning models that are trained using a large amount of text data and learn semantic information about words, sentences and context.

[0081] In an optional embodiment, the pre-trained model uses BERT, and the text directed graph includes multiple nodes and multiple directed edges, wherein the nodes correspond to words in the word sequence, and the directed edges correspond to the similarity relationship between two words.

[0082] It should be noted that BERT is a pre-trained language model based on the Transformer architecture, proposed by Google in 2018. It is an important advancement in the field of natural language processing (NLP) and has achieved significant performance improvements in many NLP tasks.

[0083] In an optional embodiment, a directed graph of aggregated text is passed based on graph convolution operations to obtain text-hidden features of word sequences, including:

[0084] C1. Embedding representation of the directed text graph based on graph convolution operations to obtain node representation, where the formula for node representation is as follows:

[0085] ;

[0086] in, The first character represents the directed graph of text. Hidden representation of layer nodes, This represents the activation function. The column represents the type. The adjacency matrix of the directed text graph. The column represents the type. The first text directed graph Hidden representation of layer nodes, The column represents the type. The first text directed graph The weight of layer nodes, Represents different types of nodes;

[0087] C2. Calculate the type-level attention score of neighboring nodes based on the node representation of each node in the text-directed graph. The formula for calculating the type-level attention score is as follows:

[0088] ;

[0089] in, This represents the type-level attention score of adjacent nodes. The column represents the type. Attention vector, This indicates a splicing operation. This represents the hidden representation of the current node. The column represents the type. The embedding representation of the adjacent nodes, and , Indicates adjacent nodes. This represents the embedding of the directed edges between the current node and its neighboring nodes. Represents the hidden representation of adjacent nodes;

[0090] C3. Obtain the type-level attention weights of neighboring nodes based on their type-level attention scores. The formula for calculating the type-level attention weights is as follows:

[0091] ;

[0092] in, This represents the type-level attention weights;

[0093] C4. Calculate the node-level attention score of neighboring nodes based on the type-level attention weights of neighboring nodes. The formula for calculating the node-level attention score is as follows:

[0094] ;

[0095] in, The attention weights represent the type-level values ​​of adjacent nodes. Represents the attention vector;

[0096] C5. The dual attention mechanism of type-level and node-level attention is fused to obtain the hidden text features of word sequences.

[0097] In an optional embodiment, the target text is matched with a plurality of preset text storage nodes based on the importance of the target text to obtain text storage nodes that match the target text, including:

[0098] D1. Obtain the behavior scores of multiple text storage nodes based on the behavior transactions of multiple text storage nodes;

[0099] D2. Compare the behavior scores of multiple text storage nodes with the preset behavior score threshold. If the behavior score of a text storage node is less than the preset behavior score threshold, then the text storage node will be eliminated from the multiple text storage nodes.

[0100] D3. Sort the remaining text storage nodes among multiple text storage nodes based on behavior scores to obtain a candidate text storage node order set;

[0101] D4. Divide the candidate text storage node sequence set into a preset number of candidate text storage node sequence subsets, and match the target text to the candidate text storage node sequence subset based on the importance of the target text;

[0102] D5. Based on the remaining computing resources of each candidate text storage node in the sequential subset of candidate text storage nodes, match the target text to the candidate text storage node with the largest remaining computing resources.

[0103] It should be noted that the candidate text storage node sequence set is a set of text storage nodes sorted according to their behavior scores, with nodes arranged in descending order to indicate the order of priority for use; the candidate text storage node sequence subset is a partitioning of the candidate text storage node sequence set, dividing the sorted node set into several subsets; the importance of the target text refers to the priority or processing value of the text in a specific context; and the remaining computing resources refer to the remaining computing power that each text storage node can currently use to process tasks.

[0104] In an optional embodiment, the behavioral score is calculated using the following formula:

[0105] ;

[0106] in, Represents the first of multiple text storage nodes The behavioral score of each text storage node Represents the first of multiple text storage nodes A sequence of transactions involving individual text storage nodes. Represents the first of multiple text storage nodes If a transaction involves a text storage node, is it legal? If legal, then... Otherwise, , Represents the first of multiple text storage nodes If a transaction involving a text storage node is illegal, then... Otherwise, , This indicates the preset penalty coefficient.

[0107] It should be noted that the behavior score is a quantitative indicator used to evaluate the performance of text storage nodes during task execution. It reflects the overall health or work efficiency of the node. The transactions that text storage nodes participate in include, but are not limited to, data storage, data reading, data updating, data deletion, data synchronization, data compression, and log recording.

[0108] Example 3, as Figure 2As shown, the electronic archive management system provided by the present application is suitable for the electronic archive management method, and comprises:

[0109] The character recognition unit 1 is configured to acquire the paper archive image, perform character recognition on the paper archive image based on image processing technology, and obtain the first target text.

[0110] The text processing unit 2 is configured to receive the electronic archive uploaded by a plurality of preset data sources at every preset sampling period, convert the electronic document into the second target text, and preprocess the first target text and the second target text to obtain the target text.

[0111] The text classification unit 3 is configured to perform word segmentation on the target text to obtain a word sequence corresponding to the target text, and classify the target text based on the word sequence to obtain a text type label corresponding to the target text.

[0112] The storage matching unit 4 is configured to acquire the importance of the target text based on the text type label corresponding to the target text, and match the target text with a plurality of preset text storage nodes based on the importance of the target text to obtain a text storage node matched with the target text.

[0113] The text storage unit 5 is configured to store the target text into the matched text storage node.

[0114] The above detailed description of the embodiments of the present application is made in combination with the accompanying drawings, but the present application is not limited thereto, and various changes can be made within the knowledge of those skilled in the art without departing from the spirit of the present application.

Claims

1. A method of managing electronic archives, characterized by, The method comprises the following steps: acquiring a paper file image, performing text recognition on the paper file image based on image processing technology to obtain first target text; receiving electronic files uploaded by a plurality of preset data sources every preset sampling period, converting the electronic files into second target text, and preprocessing the first target text and the second target text to obtain target text; performing word segmentation on the target text to obtain a word sequence corresponding to the target text, classifying the target text based on the word sequence to obtain a text type label corresponding to the target text; obtaining the importance of the target text based on the text type label corresponding to the target text, and matching the target text with a plurality of preset text storage nodes based on the importance of the target text to obtain a text storage node matched with the target text; storing the target text to the matched text storage node; matching the target text with a plurality of preset text storage nodes based on the importance of the target text to obtain a text storage node matched with the target text, comprising: obtaining behavior scores of the plurality of text storage nodes based on behavior transactions of the plurality of text storage nodes; comparing the behavior scores of the plurality of text storage nodes with a preset behavior score threshold, and if the behavior score of the text storage node is less than the preset behavior score threshold, eliminating the text storage node from the plurality of text storage nodes; sorting the remaining text storage nodes in the plurality of text storage nodes based on the behavior scores to obtain a candidate text storage node order set; dividing the candidate text storage node order set into a preset number of candidate text storage node order sub-sets, and matching the target text to a candidate text storage node order sub-set based on the importance of the target text; matching the target text to the candidate text storage node with the largest remaining computing resources based on the remaining computing resources of each candidate text storage node in the candidate text storage node order sub-set; the calculation formula of the behavior score is as follows: ； wherein, represents a behavior score of the th text storage node in the plurality of text storage nodes, represents a transaction sequence in which the th text storage node in the plurality of text storage nodes participates, represents whether a transaction in which the th text storage node in the plurality of text storage nodes participates is legal, if legal, then , otherwise, then , represents whether a transaction in which the th text storage node in the plurality of text storage nodes participates is illegal, if illegal, then , otherwise, then , represents a preset penalty coefficient.

2. The method of claim 1, wherein, performing text recognition on the paper file image based on image processing technology to obtain first target text, comprising: performing image enhancement on the paper file image to obtain a paper file enhanced image; performing text box detection on the paper file enhanced image based on a trained first target detection model to obtain a plurality of first text boxes corresponding to the paper file enhanced image, and dividing the plurality of first text boxes by columns to obtain a text box column set; performing text box detection on the text box column set based on a trained second target detection model to obtain a plurality of second text boxes corresponding to the text box column set, and dividing the plurality of second text boxes by rows to obtain a text box row set; performing text recognition on the text box row set based on a trained text recognition model to obtain text recognition results corresponding to the text box row set; sorting the text recognition results to obtain first target text.

3. The method of claim 2, wherein, The character recognition model comprises a local feature extraction module, a global feature extraction module, a weighted aggregation module and a classification module, the local feature extraction module is used for extracting local image features of the text box row set, the global feature extraction module is used for extracting global image features of the text box row set, the weighted aggregation module is used for weighting and aggregating the local image features and the global image features to obtain aggregated image features, and the classification module is used for classifying the aggregated image features to obtain a character recognition result corresponding to the text box row set.

4. The method of claim 3, wherein, The local feature extraction module extracts image features by using a depth separable convolution, the global feature extraction module comprises a window multi-head attention layer, a sliding window multi-head attention layer and an MLP layer, each MLP layer is connected with the window multi-head attention layer and the sliding window multi-head attention layer before normalization, and each layer adopts residual connection, the weighted aggregation module combines the local image features and the global image features by GCN to obtain aggregated image features, and the classification module classifies the aggregated image features by a full connection layer and a Softmax function to obtain the character recognition result corresponding to the text box row set.

5. The method of claim 1, wherein, The target text is classified based on the word sequence to obtain a text type label corresponding to the target text, comprising: Contextual semantic features and word embedding vectors of the word sequence are obtained based on the trained pre-training model; The similarity between each word in the word sequence is calculated based on the word embedding vectors of the word sequence to obtain the similarity between each word in the word sequence, and if the similarity is greater than a preset similarity threshold, it indicates that there is a similar relationship between the two words; A text directed graph is constructed based on the word sequence and the similar relationship between the words, and the text directed graph is aggregated based on a graph convolution operation to obtain text hidden features of the word sequence; The contextual semantic features and the text hidden features of the word sequence are fused to obtain fusion features, and the fusion features are classified to obtain a text type label corresponding to the target text.

6. The method of claim 5, wherein, The pre-training model adopts BERT, and the text directed graph comprises a plurality of nodes and a plurality of directed edges, wherein the nodes correspond to words in the word sequence, and the directed edges correspond to the similar relationship between the two words.

7. The method of claim 6, wherein, The text directed graph is aggregated based on a graph convolution operation to obtain text hidden features of the word sequence, comprising: The text directed graph is embedded based on a graph convolution operation to obtain node representation, wherein the formula of the node representation is as follows: ； wherein, represents the hidden representation of the node at layer represents the hidden representation of the node at layer represents an activation function, represents the adjacency matrix of the text directed graph of list represents the adjacency matrix of the text directed graph of list represents the hidden representation of the node at layer represents the hidden representation of the node at layer represents the weight of the node at layer represents the weight of the node at layer represents the weight of the node at layer represents the weight of the node at layer represents different types of nodes; The type-level attention score of adjacent nodes is calculated based on the node representation of each node in the text directed graph, wherein the calculation formula of the type-level attention score is as follows: ； wherein, denotes a type-level attention score for a neighboring node, denotes a list of attention vectors of type , denotes a concatenation operation, denotes a hidden representation of a current node, denotes a list of embedding representations of neighboring nodes of type , , denotes a neighboring node, denotes an embedding representation of a directed edge between a current node and a neighboring node, denotes a hidden representation of a neighboring node; The type-level attention weight of adjacent nodes is obtained based on the type-level attention score of adjacent nodes, wherein the calculation formula of the type-level attention weight is as follows: ； wherein, denotes the type-level attention weight; The node-level attention score of the adjacent node is calculated based on the attention weight of the type level of the adjacent node, and a calculation formula of the node-level attention score is as follows: ； wherein, denotes an attention weight for the type level of the neighboring node, denotes an attention vector; The double-level attention mechanism of the type level and the node level is fused to obtain the text hidden feature of the word sequence.

8. A system for managing electronic files, which is adapted to the method for managing electronic files according to any one of claims 1 to 7, characterized in that, Comprise: A character recognition unit (1) is used to obtain a paper archive image, and character recognition is performed on the paper archive image based on image processing technology to obtain a first target text; A text processing unit (2) is used to receive electronic archives uploaded by a plurality of preset data sources every preset sampling period, convert the electronic archives into a second target text, and preprocess the first target text and the second target text to obtain a target text; A text classification unit (3) is used to perform a word segmentation operation on the target text to obtain a word sequence corresponding to the target text, classify the target text based on the word sequence, and obtain a text type label corresponding to the target text; A storage matching unit (4) is used to obtain the importance of the target text based on the text type label corresponding to the target text, match the target text with a plurality of preset text storage nodes based on the importance of the target text, and obtain a text storage node matched with the target text; A text storage unit (5) is used to store the target text to the matched text storage node.

Images