Intelligent monitoring management method and system based on archive digitization

Abstract

The invention discloses an intelligent monitoring management method and system based on archive digitization, and relates to the technical field of data management, and the method comprises the steps: collecting and preprocessing multi-source archive data, employing a multi-mode BERT model to carry out the feature fusion of different data sources, and generating a unified semantic representation; semantic labeling is performed on archive data through a multi-label classification model, a semantic graph of archive content is constructed by using a graph database, an association relationship between archives is represented, a semantic index tree is constructed based on the semantic graph, and rapid positioning and calling of the archive content are optimized; and recording the change of each file version, positioning the change position based on a semantic index tree, identifying the semantic change of the file through a semantic difference comparison algorithm, recording hash, carrying out granularity division on the file content through the semantic boundary of each level of node in the index tree, and generating a user access strategy. According to the invention, dynamic perception and risk early warning of user behaviors are realized, and the intellectualization and safety of the archive management system are effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data management, and in particular to an intelligent monitoring management method and system based on archive digitization. BACKGROUND

[0002] With the rapid development of information technology, cloud computing and artificial intelligence, traditional archive management methods gradually evolve towards digitization, networking and intelligence. The archive digitization system not only carries the functions of archiving, storing and retrieving structured and unstructured information, but also undertakes the tasks of recording key business processes, knowledge sedimentation and compliance audit. At present, a large number of literatures and products have realized the basic functions of centralized management, keyword search, full-text search and permission control of digital archives. In recent years, the rise of semantic technology, natural language processing (NLP) and graph neural network (GNN) has provided new possibilities for deep semantic modeling of archive content. Using deep models such as BERT to extract features from text content and perform classification, clustering and label annotation has become a mainstream solution. At the same time, with the large amount of multi-modal data (such as images, texts, structured forms, etc.) existing in archives, cross-modal fusion processing and unified representation have become technical difficulties and research hotspots in intelligent archive management systems.

[0003] However, the existing technology still has some deficiencies. On the one hand, most systems only extract and simply classify the surface information of document content, lack the ability to model the semantic hierarchical structure and semantic relationship between archives, and are difficult to support archive calling, comparison and update tracking in complex contexts. On the other hand, the current permission management mechanism mostly relies on static role assignment or rule-driven access control, which cannot refine authorization according to the semantic granularity of archive content, and cannot dynamically adapt to user behavior characteristics. In addition, the change process of archive content version lacks detailed tracking and comparison mechanism, making it difficult to accurately record and identify semantic changes. More importantly, the current archive system mainly focuses on recording access logs for user behavior monitoring, and lacks deep modeling based on semantic paths and behavior graphs, which cannot effectively identify potential risk behaviors. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides an intelligent monitoring management method and system based on archive digitization, which solves the problems of insufficient content semantic understanding, weak version change recognition ability, coarse-grained permission management and limited user behavior monitoring capability in existing archive digitization technology.

[0006] To solve the above technical problems, the present application provides the following technical solutions: In a first aspect, the present application provides an intelligent monitoring management method based on archive digitization, which comprises, Collecting multi-source archive data and preprocessing, using a multi-modal BERT model to fuse features of different data sources, and generating a unified semantic representation; Through a multi-label classification model, the archive data is semantically labeled, and a semantic graph of the archive content is constructed using a graph database to represent the association between archives, a semantic index tree is constructed based on the semantic graph to optimize the rapid positioning and calling of the archive content; Record the changes of each archive version, locate the change position based on the semantic index tree, identify the semantic changes of the archive and record the hash through the semantic difference comparison algorithm, and divide the archive content by granularity and generate a user access strategy through the semantic boundaries of the nodes in the index tree; Record the semantic path of the user accessing the archive and construct a user behavior graph, identify potential abnormal behaviors through behavior graph analysis, and use time series analysis and fuzzy mining algorithms for risk prediction and behavior warning.

[0007] As a preferred scheme of the intelligent monitoring management method based on archive digitization, the multi-label classification model is used to semantically label the archive data, and a semantic graph of the archive content is constructed using a graph database to represent the association between archives, and the semantic index tree is constructed based on the semantic graph, comprising: Construct a Label-GAT model, specifically including defining a label set required for archive management, assigning an embedding vector to each label, taking the joint representation vector A and the label vector as nodes, and establishing a heterogeneous graph structure; Carry out graph attention propagation on the heterogeneous graph structure, and each label node will obtain a semantic representation related to the current sample; Split the joint representation vector A into three sub-vectors of image, text and structured field, respectively as three modal information sources, determine the matching degree of the label under different modalities by calculating the attention weight, and obtain the final fused label representation; Input the fused representation of all labels into a shared fully connected layer, and output the prediction result through the Sigmoid function, train the Label-GAT model, for any new archive, input the corresponding joint representation vector into the trained Label-GAT model, pass through the graph attention propagation and multi-modal weighted interaction, and output a label probability value, according to the set threshold, the label with a probability value greater than the threshold is determined as the classification result of the archive; According to the classification result set, the basic entity type in the graph is determined, the semantic edge relationship between the entities in the graph is established, all entities and their semantic connections are uniformly organized into a triple form in the graph structure, the graph structure is standardized, the initial structure of the graph is constructed, and the semantic reasoning rule is defined; The semantic graph structure is standardized, the knowledge graph embedding model is used to train the graph, each entity and relationship is mapped into a unified low-dimensional vector space, a density-based clustering algorithm is used to cluster all archive entities, and according to the clustering result, the clustering center and the corresponding entity are divided to construct a hierarchical semantic index tree, the upper node of the tree represents a semantic clustering center, each center node is connected to the entity node or the sub-clustering node under it, and a multi-layer index tree is constructed by controlling the branch number and depth of each layer.

[0008] As a preferred scheme of the intelligent monitoring management method based on archive digitization, wherein: the multi-modal BERT model is used to fuse the features of different data sources to generate a unified semantic representation, which collects multi-source archive data, including image data, text data and structured data, and pre-processes the collected data; The pre-processed image is uniformly divided into fixed-size image blocks, and each image block is assigned corresponding position information according to its spatial position in the whole image, so as to generate an image block feature matrix containing image local content and spatial position information, and a multi-layer encoding structure based on attention mechanism is used to model the image block sequence and extract semantic correlation and global feature expression in the image mode; The pre-processed text data is decomposed into a series of word or sub-word units, the BERT model is used to map the tokenized text into a vector representation, each token obtains a fixed-dimensional embedding vector, and an initial text feature matrix is formed, a deep language encoder is used to model the text sequence, and a text feature representation containing context semantics is obtained; After the image feature sequence and the text feature sequence are encoded and mapped to a unified semantic space, information interaction is realized through a cross-modal attention mechanism, and finally aggregated into a graphic-text fusion feature vector F; The pre-processed structured field is spliced into a structured feature vector, the pre-processed structured feature vector is input into a hybrid modeling path, the optimal modeling strategy is selected for different data types of the field, the feature results are spliced and fused, and input into a multi-layer perception encoder to obtain a semantic embedding vector Q of the structured field; The graphic-text fusion feature vector F and the structured semantic embedding vector Q are subjected to vector dimension alignment processing and splicing operation to construct a joint representation vector A.

[0009] As a preferred scheme of the intelligent monitoring management method based on archive digitization, wherein: the record of each archive version change is based on semantic index tree positioning change position, the semantic change of the archive is identified by semantic difference comparison algorithm and the hash pointer is recorded, when the archive is modified, rewritten or updated version, the completed semantic index tree is called immediately, the change content is embedded vector coding, and the semantic similarity calculation is carried out with the embedded vector of the path node of the original archive in the index tree, the nearest path node of the change point in the index tree is determined, and the change anchor point of the semantic position of the archive is located. The content of the current change and the content of the last version of the archive are extracted, and the unified semantic coding model is used to compare the distance of the two versions in the sentence vector or paragraph vector space, and the overall semantic difference is calculated combined with the label path change of the key field. The current change content is generated as a structured snapshot, and full content hash calculation is performed.

[0010] As a preferred scheme of the intelligent monitoring management method based on archive digitization, wherein: the granularity division of the archive content is carried out by the semantic boundary of each node in the index tree, and the user access strategy is generated. The semantic classification boundary is regarded as the semantic classification boundary of each node in the index tree, the semantic break point is identified by analyzing the similarity of the semantic vectors between nodes, and different levels of semantic segments are divided, each archive version is bound to the corresponding semantic level in the corresponding index path, the permission vector is generated according to the user's organization, role and access scene, and the vector comparison is carried out with each semantic segment, and whether the user has access right is judged by similarity threshold.

[0011] As a preferred scheme of the intelligent monitoring management method based on archive digitization, wherein: the potential abnormal behavior is identified by behavior graph analysis, and the risk prediction and behavior warning are carried out by time series analysis and fuzzy mining algorithm. By recording the semantic path node, access time and archive version when the user accesses the archive, the behavior graph centered on the user is continuously constructed, the behavior characteristics of the user in each semantic category, such as access frequency, path jump mode and access depth, are modeled and analyzed, the behavior preference and conventional mode are identified, when the user behavior appears the characteristics of different access path and access density mutation from the historical mode, the time series model is constructed based on the behavior graph, the potential abnormal behavior mode is identified by combining the fuzzy mining algorithm, and the corresponding behavior warning process is generated.

[0012] As a preferred scheme of the intelligent monitoring management method based on archive digitization, wherein: the behavior warning process includes immediately freezing the current access right of the user or forcibly degrading to read-only mode, automatically generating structured alarm information, pushing to the administrator interface, recording the complete behavior link, and appending the risk event to the chain log.

[0013] In a second aspect, the present application provides an intelligent monitoring management system based on archive digitization, comprising, a semantic fusion module for collecting multi-source archive data and performing structured cleaning, and generating unified semantic representation by using a multi-modal BERT model; a graph construction module for performing semantic tagging on archive content by a multi-label classification model, and constructing a graph database to represent semantic association relationships between archives; a semantic index construction module for constructing a semantic index tree based on a semantic graph, realizing fast positioning of archives, and recording each version change and semantic change path and hash information; a permission policy module for dividing archive content granularity according to semantic boundaries at each level in the index tree, and generating fine-grained user access permission policies; a risk early warning module for recording user access behavior to construct a behavior graph, identifying abnormal behavior by combining time series analysis and fuzzy mining, and generating a risk early warning signal.

[0014] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and wherein the computer program is executed by the processor to implement any step of the intelligent monitoring management method based on archive digitization according to the first aspect of the present application.

[0015] In a fourth aspect, the present application provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement any step of the intelligent monitoring management method based on archive digitization according to the first aspect of the present application.

[0016] The present application has the following beneficial effects: the present application realizes semantic unified fusion representation of multi-source archive data by using a multi-modal BERT model, constructs an archive semantic graph and an index tree by combining a graph neural network and a graph database, realizes efficient and traceable content positioning and evolution tracking, continuously models a user behavior semantic path, combines time series analysis and fuzzy mining technology, realizes dynamic perception and risk early warning of user behavior, and effectively improves the intelligence and security of the archive management system. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0018] Fig. 1 The flowchart of the intelligent monitoring management method based on archive digitization in embodiment 1.

[0019] Fig. 2 Structure diagram of the intelligent monitoring management system based on archive digitization in embodiment 1.

[0020] Fig. 3 Flowchart of the construction of the semantic graph and the index tree in embodiment 1. DETAILED DESCRIPTION

[0021] In order to make the above objectives, features and advantages of the present application more apparent, specific embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0022] In the following description, a large number of specific details are set forth in order to facilitate a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0023] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an independent or alternative embodiment that excludes other embodiments.

[0024] Embodiment 1, refer to Figs. 1-3 , the first embodiment of the present application, the embodiment provides an intelligent monitoring management method based on archive digitization, comprising the following steps: S1, collecting multi-source archive data and preprocessing, using multi-modal BERT model to fuse the features of different data sources, and generating a unified semantic representation; Specifically, using a multi-modal BERT model to fuse the features of different data sources and generate a unified semantic representation means collecting multi-source archive data, including image data, text data and structured data, and preprocessing the collected data, wherein the image data comes from scanned paper documents, on-site photos or charts, which needs to be collected uniformly from the file system, cloud storage or scanning equipment, the text data comes from PDF, Word documents or web page contents, which needs to extract pure text content through a parser or an OCR tool, the structured data comes from Excel tables, CSV files or database export results, which needs to be extracted in batches through a database interface or a file reading tool; The collected image data is unified in size, pixel standardized, and each image is uniformly divided into fixed-size image blocks. The text data is disassembled into word or sub-word units by NLP tools, and stop words such as "of", "was", and "and" are removed. The segmented text is mapped to a vector representation using the BERT model, and each token gets a fixed-dimension embedding vector (such as 768 dimensions). Token refers to the smallest processing unit that the text is divided into before inputting the model. For structured data, missing value processing, numerical standardization, category field encoding, and field embedding are performed. The missing value processing refers to using mean filling or median filling to ensure data integrity. Numerical standardization refers to Z-score standardization (subtracting the mean and dividing by the standard deviation) for all numerical fields to ensure consistent dimensions. Category field encoding refers to converting categorical fields to vector form using One-Hot Encoding. Field embedding refers to generating vectors for column names and corresponding numerical values respectively. The preprocessed images are uniformly divided into fixed-size image blocks. According to the spatial position of the image blocks in the whole image, each image block is assigned corresponding position information representation, thereby generating an image block feature matrix containing local content and spatial position information of the image. A multi-layer encoding structure based on attention mechanism is used to model the image block sequence and extract semantic correlation and global feature expression in the image modality. Specifically, it includes: The image block feature sequence is added with corresponding two-dimensional position encoding to explicitly represent the spatial position information of each image region in the original image. The position encoding is realized through a learnable embedding method, so that the model can still maintain the spatial structure of the image when processing unordered sequences. The image block sequence with position encoding is input into a multi-layer encoding structure based on attention mechanism. This structure consists of several stacked self-attention encoding modules. Each module contains multi-head self-attention mechanism and feedforward neural network, which are used to capture local associations and global semantic information between image blocks simultaneously. In each encoding module, the image block feature sequence is mapped to query (Query), key (Key), and value (Value) vectors, and the attention weights between image blocks are calculated to measure their semantic relevance. According to the attention weights, the image block features are reweighted and combined, so that the representation of each image block not only contains its own information, but also integrates the context information of the associated regions.

[0025] Through the multi-layer stacked self-attention structure, the model gradually expands the receptive field, thereby realizing the semantic construction from local to global. Early layers capture local dependencies in adjacent regions, and subsequent layers gradually enhance the connections between distant regions in the image, finally forming an image feature expression with global perception ability. The modeling process not only preserves the independent semantic representation of each image block, but also builds the context structure between regions in the hidden layer, enabling the system to understand the role and relationship of each part in the overall semantic structure of the image, thereby extracting structured global image information; The output from the multi-layer encoding structure is still a sequence of image block features, but the sequence has been re-modeled with higher-level semantic relevance and cross-region information fusion capability, providing a deep expression basis for the image modality in subsequent multi-modal interaction and fusion; The preprocessed text data is decomposed into a series of word or subword units, and the tokenized text is mapped to a vector representation using the BERT model. Each token gets a fixed-dimensional embedding vector, forming an initial text feature matrix. A deep language encoder such as the Transformer encoder is used to model the text sequence and obtain a text feature representation that includes contextual semantics. After the image feature sequence and the text feature sequence are encoded and mapped to a unified semantic space, information interaction is achieved through a cross-modal attention mechanism, and finally aggregated into a text-image fusion feature vector F, which includes the following steps: The image block feature sequence and the text feature sequence are input into independent modal encoders. The image modal encoder models the spatial relationship between image blocks and outputs a deep feature representation sequence for the image. The text modal encoder models the contextual relationship between text tokens and outputs a deep semantic feature representation sequence for the text. Both types of features maintain the same order as the original input and are accompanied by position encoding to ensure that the relative order of semantic information in the sequence is preserved. The image modal feature sequence and the text modal feature sequence are dimensionally aligned to map their respective vector dimensions to a unified semantic space. This mapping process adjusts the image features and text features to the same dimension vector representation by introducing learnable projection parameters, thereby providing a basis for subsequent fusion operations. After semantic space alignment, the image feature sequence is used as the query source and the text feature sequence as the response source to establish an image-guided text cross-modal attention mechanism. Through this mechanism, the semantic similarity between each image block feature and all text token features is calculated, generating an attention weight matrix that reflects the attention degree of image blocks to different text segments, capturing the corresponding relationship between the potential content in the image and the text description. The text feature sequence is weighted and summed based on the attention weights to generate a text enhancement representation corresponding to each image block feature. The original image block feature and its corresponding text enhancement representation are combined to construct new image-text joint features. This process enables the semantic expression capability of the image modality to be further supplemented and guided by the text content, enhancing its understanding of the actual archive content. The global average pooling method is used to extract dominant semantic information in the whole set of features by performing unified aggregation processing on the image-text combined feature sequence. The aggregation operation can compress the original combined sequence to form a compact, multi-dimensional representation vector representing the overall semantics of the current archive sample under the image and text modalities. The aggregated image-text fusion representation vector is output as the final result of the multi-modal feature fusion stage. The preprocessed structured fields are spliced into a structured feature vector, which is input into the hybrid modeling path. The optimal modeling strategy is selected for different data types of the fields, including: numerical fields are sent to the Gradient Boosting Decision Tree (GBDT) model for modeling to extract the nonlinear mapping relationship between them and the target semantic distribution, and output probability features with semantic bias information; categorical fields are input into the Support Vector Machine (SVM) model, and the discrete prediction boundary output by the model is mapped to a continuous probability distribution through the Platt Scaling technique to better capture the similarity information between categories; the probability features output by the two types of models represent the semantic positions of numerical fields and categorical fields in the target label space, respectively; the two types of feature results are spliced and fused, and input into a Multi-Layer Perceptron (MLP) encoder to further extract the global expression ability of structured information in high-dimensional semantic space. This process can be regarded as deep modeling of the context rules and cross-field dependencies implied in the structured fields. The output result after MLP modeling is the semantic embedding vector Q of the structured fields, which has the same dimension as the image-text fusion representation vector. This structured feature representation retains the key information of the original structured fields while having the same expression ability as non-structured information. The Platt Scaling technique is also known as Platt calibration or Sigmoid calibration, which is a technique for converting the output of a classifier into a probability estimate. The image-text fusion feature vector F and the structured semantic embedding vector Q are aligned in vector dimension. If the dimensions are not consistent, a single-layer or multi-layer linear transformation module is introduced to map the structured features to the same dimensional space as the image-text features, ensuring that the three modalities have additivity and splicing ability in the fusion operation. The image-text fusion feature vector F and the structured semantic embedding vector Q are spliced to construct a joint representation vector A. This vector retains the semantic distribution features of images, texts, and structured fields in the feature space and is a complete mapping representation of the archive sample in the multi-modal semantic space.

[0026] By constructing a feature fusion architecture based on a multi-modal BERT model, deep modeling and efficient fusion of image, text and structured data in a unified semantic space are realized, which significantly improves the understanding and processing capability of multi-source archive data. After spatial division and position coding, the image data retains local details and global structure information; the text data obtains context semantic features through BERT, enhancing the accuracy of semantic expression; the structured data extracts non-linear and semantic distribution features in numerical and categorical fields through GBDT and SVM modeling respectively, ensuring the integrity and interpretability of structure information. The cross-modal attention mechanism realizes text interaction guided by images, making image and text information complementary and strengthening semantic alignment, further improving the deep understanding of complex archive content. Finally, a joint representation vector is constructed through unified projection and vector splicing, integrating the semantic features of the three types of data, providing high-quality feature support for downstream tasks such as classification, retrieval, archiving and risk assessment.

[0027] S2, semantic labeling of archive data is performed by a multi-label classification model, and a semantic graph of archive content is constructed using a graph database to represent the association relationship between archives, a semantic index tree is constructed based on the semantic graph to optimize the rapid positioning and calling of archive content; Specifically, semantic labeling of archive data is performed by a multi-label classification model, and a semantic graph of archive content is constructed using a graph database to represent the association relationship between archives, a semantic index tree is constructed based on the semantic graph to optimize the rapid positioning and calling of archive content, including: A Label-GAT model is constructed, specifically including defining a label set required for archive management, assigning an embedding vector to each label, and maintaining the same semantic dimension as the joint representation vector A; The joint representation vector A and the label vector are used as nodes together to establish a heterogeneous graph structure. In this graph, sample nodes are connected to all label nodes, and label nodes are also connected through semantic similarity. In this way, the model can capture the relationship between samples and labels, as well as the correlation between labels; Graph attention propagation is performed on the heterogeneous graph structure. For each label node, its representation is updated based on the information of neighbor nodes. In the updating process, the contribution of neighbor nodes is dynamically determined by attention weights, so that each label representation can fuse key information from sample nodes and other label nodes. After multiple layers of propagation, each label node obtains a semantic representation related to the current sample:

[0028]

[0029] In the formula, is the vector representation of node t after the graph attention network at the g-th layer, N(t) is the neighbor set of node t (including the nodes connected to it by edges), is the attention weight of node r to the center node t, is the linear mapping matrix of the g-th layer, used to transform the input features, is the representation of node r at the previous layer, σ() is an activation function (such as ReLU, ELU) to increase the non-linear representation ability, a is a learnable attention vector to control the weight of the concatenated vector, LeakyReLU is an activation function to process the linear transformation result after concatenation, exp() is the exponential operation of Softmax to normalize the attention weight, is the representation of node t at the previous layer, is the representation of node j at the previous layer; To further combine multi-modal features, after the label representation is updated, it needs to be weighted and interacted with different parts of the three-modal vector, and the joint representation vector A is split into three sub-vectors of image, text and structured field, which are used as three kinds of modal information sources. By calculating the attention weight, the matching degree of the label in different modalities can be determined, and the final fused label representation is obtained:

[0030]

[0031] wherein, is the attention weight of label k to modality p (such as image F, text T, and structured field Q), is the final representation vector of label k in modality p, is the semantic vector representation of modality p, is the fused label k representation vector; The fused representation of all labels is input into a shared fully connected layer, and the prediction result is output through a Sigmoid function. Each label will get a probability value between 0 and 1, indicating the possibility of the profile belonging to the label; The Label-GAT model is trained with the historical joint representation vector A as input and the real multi-label set as output. The model calculates the prediction result through forward propagation and compares it with the real label to get the loss value. Binary cross-entropy is used as the loss function, and the parameters in the model, including label embedding, weight matrix and attention parameter, are updated through back propagation. The training process is repeated until the maximum number of iterations is reached, and the trained Label-GAT model is obtained; For any new file, after obtaining the corresponding joint representation vector, input it into the trained Label-GAT model, and after graph attention propagation and multi-modal weighted interaction, output a label probability value. According to the set threshold (based on experience), the label with a probability value greater than the threshold is determined as the classification result of the file (such as file type, theme, time, sensitivity, etc.); According to the label set, determine the basic entity types in the graph, including archive entity, label entity, and time, event, organization auxiliary semantic entity; The archive entity is used to represent each archive data sample; The label entity represents the label result obtained from the semantic classification stage; The time class entity represents the time attribution of the archive; The event and organization class entities are used to mark specific events and attribution organizations involved in the archive, and each type of entity is uniformly named and coded in a structured manner; Establish semantic edge relationships between entities in the graph. This type of edge relationship is mainly derived directly from the information extracted in the previous step. For example, if a certain archive sample is assigned several labels through multi-label classification, a "belongs to" relationship is established between the archive and the corresponding label entity; If the archive text or structure field indicates that it references another archive, a "reference" relationship is established between the two archive entities; If the archive contains explicit archiving time or occurrence time information, a "belongs to time" relationship is established between it and the time class entity. In addition, if the event or attribution organization involved appears in the structure field, corresponding relationships with the event and organization class entities are also established; All entities and their semantic connections are organized into triplets in the graph structure, that is, each group of relationships consists of a starting entity, a relationship type, and a target entity. In this way, the archive and its semantic labels, related archives, occurrence time, event attribution, etc. are all abstracted as nodes and edges in the graph structure; Normalize the graph structure to meet the data format required by the graph database, and batch import it into the graph database for unified storage. During the import stage, first complete the entry of various entity nodes, and then establish various edge relationships, and mark the connection direction and attribute type of the edges. At this point, multiple semantic units such as archives, labels, times, references, events, and organizations are integrated into a unified graph database structure, and all archive semantic relationships have a queryable and inferable structure foundation; After the initial structure of the graph is completed, semantic reasoning rules are defined to identify indirect relationships that are not explicitly present in the graph but can be logically inferred. The reasoning rules are constructed based on the connection patterns of the graph structure. For example, if an archive C references an archive B and archive B belongs to a certain label, it can be inferred that there is a certain association between archive C and the label. If two archives C and B appear in the same time period, it can be inferred that they are concurrent or related to some extent. If archives C and B are related to the same event or the same organization, it can be inferred that there is some association between them. Such rules are defined uniformly and executed in bulk in the graph structure, generating a batch of newly added semantic relationship edges to enrich the semantic structure of the graph. Once the graph is established and the reasoning is completed, users can search the graph through the query interface of the graph database, especially the Cypher query language. Through the query interface, users can quickly find the information of interest such as archives, labels, times, and reference relationships in the graph. The semantic graph structure is standardized, and all archive entities, labels, events, organizations, and time nodes in the graph are uniformly numbered. The triplets (e.g., archive C, belongs to, project contract) are formatted into the standard knowledge graph input form, and an entity vocabulary, a relationship set, and an adjacency structure matrix are established. The graph is trained using a knowledge graph embedding model (such as TransE, TransR, or ComplEx), which maps each entity and relationship to a unified low-dimensional vector space. The embedding result preserves the semantic proximity between entities, so that semantically similar archives or labels are close to each other in the vector space. The embedding vector (semantic coordinates) corresponding to each graph node is obtained. All archive entities are clustered using a density-based clustering algorithm, which includes the following steps: The semantic embedding vectors of all archive entities are used as input data. The similarity matrix between entities is constructed by calculating the distance (such as Euclidean distance or cosine similarity) between vectors. According to the parameters set by the clustering algorithm (such as eps representing the neighborhood radius and min_samples representing the minimum number of points in a cluster in DBSCAN), the areas with high local density in the vector space are identified, and multiple entity clusters are automatically formed. Each cluster represents a class of semantically similar archive entities. The algorithm automatically marks noise points (i.e., abnormal entities that cannot be assigned to any cluster) as independent samples. After clustering, for each cluster, the average value of all entity vectors in the cluster is calculated to obtain a center vector representing the semantic cluster. The center vector can be regarded as the semantic "center of gravity" of the class in the semantic space. This clustering algorithm can automatically divide semantically similar archive groups based on the density relationship of entities in the vector space, while eliminating isolated points or abnormal samples. The cluster center represents the semantic center of a class of archives. According to the clustering results, the clustering centers and the corresponding entity division are constructed into a hierarchical semantic index tree, the upper nodes of the tree represent semantic clustering centers (i.e. semantic categories), each center node is connected to its subordinate entity nodes or sub-cluster nodes, by controlling the number of branches and depth of each layer, a multi-layer index tree is constructed, which specifically includes: Based on the semantic clusters output by clustering, each clustering cluster is regarded as a semantic category node, and the embedding vector of the clustering center is regarded as the semantic representation of the node, forming the first-level structure of the index tree. On this basis, the distribution of entities in each cluster is further analyzed: If the number of entities in a cluster is large, sub-clustering division is performed again in the cluster to subdivide the entities into multiple sub-class nodes as the next level branch nodes of the clustering center. Conversely, if the cluster is small or the internal entities are semantically close, these entities are directly connected as leaf nodes under the clustering center node. With all clustering clusters expanding layer by layer and connecting specific archive entities, a multi-layer index tree is gradually formed; In order to ensure the stability of the tree structure in terms of query efficiency and maintainability, the number of nodes in each layer, the depth of the tree and the capacity of the leaf nodes need to be controlled during construction. When the number of branches of a node exceeds the set threshold (based on experience), local re-clustering or structure reconstruction is automatically triggered to maintain the balance of the overall structure. While connecting the nodes, the complete semantic path label of each node is constructed from the root node, and its embedding vector and path position in the semantic space are recorded; Finally, the entire index tree is serialized into a structured storage data model, each node contains its ID, semantic representation, parent-child node relationship, path information and related archive attributes, realizing the unification of semantic structure and entity index; Through the above continuous steps, the clustering results naturally transition into an index tree structure with semantic hierarchical perception ability, providing a structural basis for subsequent semantic query optimization and access strategy formulation; In order to enable user queries to be efficiently mapped to semantic paths in the index tree, a semantic query encoder (such as BERT or MPNet as the encoder backbone) is further trained to map query content (keywords, label combinations, etc.) into the same vector space as the tree structure. On this basis, the embedding representation of the query encoder and the index path is updated simultaneously through a joint optimization method, so that semantically similar queries can more efficiently hit the correct tree path.

[0032] By constructing the Label-GAT model, the label nodes and the sample nodes are jointly modeled, and the semantic edge relationship between the labels is introduced, so that the model has stronger label semantic perception ability in the multi-label classification task, and the classification accuracy and the inter-label collaborative expression effect are significantly improved, secondly, the multi-modal features (image, text, structure field) are combined for weighted interaction, effectively alleviating the semantic missing problem caused by single mode, and enhancing the robustness of the system to complex archive content understanding and label matching; In terms of knowledge organization, the constructed semantic graph connects multiple entities such as archives, labels, time, events and organizations in the form of triples, realizes explicit expression and unified management of semantic relationships, and through graph reasoning mechanism, the system can automatically discover the logical derivable relationship between archives which is not explicitly recorded, and improves the semantic association and retrieval breadth of archives, further, with the help of knowledge graph embedding and density clustering algorithm, the semantic similar archive entities are divided into several semantic clusters, and a hierarchical semantic index tree is constructed, so that the system has the ability of fast positioning, semantic clustering retrieval and path navigation; The semantic encoder maps the query content to the same semantic space as the index tree, which significantly improves the matching efficiency and semantic accuracy of the query. Overall, the present application realizes the whole process optimization from semantic understanding to semantic organization and then to semantic retrieval, has the structure extensibility, retrieval intelligence and semantic reasoning ability, and is suitable for large-scale archive management scenarios in government, law, enterprise and other fields.

[0033] S3, record the change of each archive version, locate the change position based on the semantic index tree, identify the semantic change of the archive by the semantic difference comparison algorithm and record the hash, divide the archive content by the semantic boundary of each node in the index tree and generate a user access strategy; Specifically, the change of each archive version is recorded, the change position is located based on the semantic index tree, the semantic change of the archive is identified by the semantic difference comparison algorithm and the hash is recorded. When it is detected that the archive is modified, rewritten or updated, whether the title content, the text field, the label attribute or the attached instructions are changed, the constructed semantic index tree is immediately called, the changed content is embedded vector coded, and the semantic similarity between the embedded vector of the changed content and the embedded vector of the path node where the original archive is located in the index tree is calculated. The closest path node of the change point in the index tree is determined by using the cosine similarity method, so as to accurately locate the change anchor point of the semantic position of the archive; The content of the current change and the content of the previous version of the file are extracted, and a unified semantic encoding model (such as MPNet or BGE) is used to compare the distance of the two versions in the sentence vector or paragraph vector space, and the overall semantic difference is calculated by combining the label path changes of the key fields. Specifically, for each pair of changed fields or paragraphs, the vector cosine similarity is calculated. If the similarity is lower than a threshold value (based on experience) (such as 0.85), it is determined that "semantic changes have occurred". At the same time, the proportion of paragraphs with semantic changes, the coverage of fields, and the hierarchical changes in the index path are calculated to obtain a "structural semantic change score". After comparison, a structured snapshot of the current change content is generated, and full content (including text, metadata, and semantic anchor path) hash calculation is performed. SHA-256 algorithm is used to generate irreversible hash fingerprint. The hash value, together with version ID, timestamp, operation user identity, and change path ID, is submitted to the chain storage system as version metadata, such as the version ledger of Hyperledger Fabric. At the same time, a "sequential connection relationship" is established between the historical version node and the current new version node of the file in the local semantic graph, clearly recording the evolution of the new version from the old version. This connection relationship is used to build a clear version evolution path, and the new version node is hung under the original leaf node in the semantic index tree to form a version branch. If the change results in a change in the label system (such as "purchase contract" to "commission contract"), the file content is re-embedded and the local semantic cluster vector representation is updated to trigger the clustering and attribution adjustment of the relevant subtree in the index tree, ensuring the structural consistency and subsequent retrieval ability. The entire process realizes an integrated closed loop from file change detection to semantic positioning, difference comparison, version snapshot generation, hash signature, and structure update, ensuring that each file change has a clear semantic anchor and tamper-proof record, while supporting the tracing of any version and its evolution path and semantic change process.

[0034] By constructing a semantic index tree and introducing a semantic difference comparison algorithm, accurate detection and positioning of file content changes are achieved. Embedding vector encoding and cosine similarity calculation ensures efficient identification of semantic changes, while generating a structured snapshot and calculating a hash fingerprint to achieve tamper-proof version records. Writing the hash value and version metadata into the blockchain system enables traceability and auditability of the file. By constructing a version evolution path, the history of the file is clearly presented. The system also supports dynamic adjustment of the label system based on semantic clustering, ensuring the index structure and retrieval performance. In addition, combined with user operation records, permission control and behavior audit can be enhanced, comprehensively improving the security, intelligence, and reliability of the file management system.

[0035] Further, each node of each layer of the semantic index tree is regarded as a semantic classification boundary, each node represents a relatively independent semantic domain (such as "project contract", "internal approval document", "financial statement", etc.), and the archive entity is at the last layer leaf node thereof. By performing clustering distance analysis on the embedding vectors of each path from the root node to the leaf node, the breaking point of the hierarchical semantic boundary in the path is automatically identified, i.e. the position of the sudden increase of semantic change between two nodes. For example, if the cosine similarity of the embedding vectors of a parent node and its child node is significantly lower than that of its adjacent layer (e.g. lower than 0.7), the node is marked as an "access control breakpoint", indicating that the content after this layer has a more fine-grained sensitivity and semantic isolation value. Based on this, the archive content is bound to the path thereof, and the content is divided into several semantic segments (such as top-level semantic segment, middle-level semantic segment, and bottom-level detail segment), each of which is labeled with its classification, sensitivity level, and attached semantic path layer number. A user role permission policy model is constructed, and the organization attribute (such as the unit to which the user belongs), the responsibility role (such as the project manager, the auditor), the access scenario (such as daily query, cross-department retrieval, joint editing), and other information of the user are encoded into a set of permission vectors. These permission vectors are compared with each semantic segment vector on the index tree path. The comparison uses threshold authorization logic: if the cosine similarity of the user permission vector and the semantic segment vector exceeds a certain threshold (such as 0.85, based on experience), it is determined that the user has the semantic ability and organizational qualifications to access the semantic segment. If it is lower, access is directly denied. If it is equal, an upper-level manual approval mechanism is triggered or a fuzzy query result (such as a desensitized segment or a label prompt) is provided as an alternative. The permission policy is finally recorded in the form of an access mapping relationship composed of path ID + layer number + role label, such as "path Y-node 3 can be accessed by role X in organization Y". This relationship is mapped into a smart contract rule on the blockchain, realizing on-chain verification of access control. Specifically, in the actual access process, the access request submitted by the user is first converted into a query vector, and is located to the target path in the semantic index tree. Then, according to the user's identity, it is queried in the permission policy whether the access permission of the semantic segment under the path is matched. If the matching is successful, the requested archive segment content is returned, and the access operation (including time, user, path ID, version number, etc.) is recorded and chained. If the matching is not successful, the denial result or alternative path suggestion is returned according to the permission policy. In this way, hierarchical and granular access control based on semantic structure is realized, ensuring that the information obtained by each organization, each user, and in each access scenario is consistent with their authorization, and all access processes have the characteristics of auditability, verifiability, and traceability, thereby ensuring data security while maximizing support for cross-organizational collaboration and sharing.

[0036] S4, record the semantic path of user access archives and build a user behavior graph, identify potential abnormal behaviors through behavior graph analysis, and use time series analysis and fuzzy mining algorithm for risk prediction and behavior warning; Specifically, whenever a user performs any operation, such as accessing archives, downloading attachments, editing and modifying, version submission, forwarding sharing or deleting files, the key elements of the behavior event are recorded, including user identity, operation type, operation time, target archive path (corresponding to node ID of semantic index tree), version number, access mode (interface, Web, API) and other information, the behavior record is encapsulated into standardized event sequence in real time, and the continuous operation trajectory is formed in time sequence as the main body of each user; Based on the continuous operation trajectory, a "user behavior graph" is constructed: each node in the graph represents an operation behavior (such as "download-financial report-v3" or "modify-project contract-v2"), and the node is attached with operation meta information (such as the corresponding label, semantic path, and organizational level). A directed edge is established between any two consecutive operation nodes to express the operation sequence, time interval or behavior frequency statistics. The weight of the edge can be modeled according to the access density, operation inertia and semantic jump degree. This graph not only visually presents the user's behavior preference and path concentration, but also provides structured input for downstream behavior prediction; The constructed behavior graph is then converted into training samples for time series model to predict subsequent operations or identify potential abnormal risks. LSTM is used to model the user's operation sequence, with the input being the continuous operation vector sequence within the time window (obtained by vectorizing the behavior graph nodes), and the output being the predicted operation category or operation path probability distribution; To enhance the abnormal detection capability, a fuzzy behavior mining mechanism is further introduced: the "semantic mutation degree" (i.e. the path jump amplitude of the current operation and its previous operation in the semantic index tree), "time interval abnormal value" (such as sudden operation at night), and "operation intensity" (such as operation frequency per unit time) of each operation behavior are used as features to construct a comprehensive abnormal score function. For example, if a user suddenly jumps from "project progress document" to "contract performance record" and "financial report" directories, and initiates a large number of download or sharing operations in a short time, this behavior trajectory has multiple abnormal characteristics of "path cross-domain + operation over-density + semantic jump", and finally its abnormal score is raised above the set threshold (experimentally optimized setting); Once the anomaly score exceeds the set threshold (experimentally optimized setting), the control response process bound with the user permission contract is triggered immediately, including: immediately freezing the user's current access permission or forcing it to read-only mode to prevent further spread of risky behavior, automatically generating structured alarm information, pushing it to the administrator interface and security audit module, recording the complete behavior link at the same time, and appending the risk event to the on-chain log, including behavior graph fragments, score details, trigger nodes, control actions, etc. Through this risk control mechanism combining behavior graph and time series prediction, the system not only realizes dynamic learning and anomaly detection of user operation mode, but also establishes a complete closed loop from access control to risk response, effectively guaranteeing the data security and behavior controllability of multi-party archive system in an open collaboration environment.

[0037] The embodiment also provides an intelligent monitoring management system based on archive digitization, comprising: A semantic fusion module is configured to collect multi-source archive data and perform structured cleaning, and generate unified semantic representation by using a multi-modal BERT model. A graph construction module is configured to perform semantic tagging on archive content by using a multi-label classification model, and construct a graph database to represent semantic association relationships between archives. A semantic index construction module is configured to construct a semantic index tree based on a semantic graph, realize fast positioning of archives, and record each version change and semantic change path and hash information. A permission policy module is configured to divide archive content granularity according to semantic boundaries at all levels in the index tree, and generate a fine-grained user access permission policy. A risk warning module is configured to record user access behavior to construct a behavior graph, and identify abnormal behavior and generate a risk warning signal by combining time series analysis and fuzzy mining.

[0038] The embodiment also provides a computer device suitable for the intelligent monitoring management method based on archive digitization, comprising a memory and a processor; the memory is configured to store computer executable instructions, and the processor is configured to execute the computer executable instructions to realize the intelligent monitoring management method based on archive digitization proposed in the above embodiment.

[0039] The computer device can be a terminal, which comprises a processor, a memory, a communication interface, a display screen and an input device connected by a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved by WIFI, an operator network, NFC (Near Field Communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0040] The embodiment also provides a storage medium having a computer program stored thereon, the program being executed by a processor to implement the intelligent monitoring management method based on archiving digitization proposed in the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic storage, a flash memory, a magnetic disk or an optical disk.

[0041] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application, and all of them should be covered in the scope of the claims of the present application.

Claims

1. An intelligent monitoring management method based on archive digitization, characterized by: Comprising, Collect multi-source archive data and preprocess, use multi-modal BERT model to fuse features of different data sources, and generate unified semantic representation; Through a multi-label classification model, the archive data is semantically labeled, and a semantic graph of the archive content is constructed using a graph database to represent the relationship between archives. A semantic index tree is constructed based on the semantic graph to optimize the rapid positioning and calling of archive content; Record the changes of each archive version, locate the change position based on the semantic index tree, identify the semantic changes of the archive and record the hash through the semantic difference comparison algorithm, and generate user access strategies by dividing the archive content into different granularities according to the semantic boundaries of each node in the index tree; Record the semantic path of user accessing the archive and construct a user behavior graph, identify potential abnormal behaviors through behavior graph analysis, and use time series analysis and fuzzy mining algorithms for risk prediction and behavior warning.

2. The method for intelligent monitoring management based on archive digitization according to claim 1, characterized in that: The multi-label classification model is used to semantically label the archive data, and a semantic graph of the archive content is constructed using a graph database to represent the relationship between archives. A semantic index tree is constructed based on the semantic graph to optimize the rapid positioning and calling of archive content; A Label-GAT model is constructed, which specifically includes defining a set of labels required for archive management, assigning an embedding vector to each label, using the joint representation vector A and the label vector as nodes, and establishing a heterogeneous graph structure; Graph attention propagation is performed on the heterogeneous graph structure, and each label node obtains a semantic representation related to the current sample; The joint representation vector A is split into three sub-vectors: image, text, and structured field, which are used as three modal information sources. By calculating the attention weight, the matching degree of the label in different modalities is determined, and the final fused label representation is obtained; The fused representation of all labels is input into a shared fully connected layer, and the prediction result is output through the Sigmoid function. The Label-GAT model is trained. For any new archive, after obtaining the corresponding joint representation vector, it is input into the trained Label-GAT model. After graph attention propagation and multi-modal weighted interaction, a label probability value is output. According to the set threshold, the labels with probability values greater than the threshold are determined as the classification results of the archive; According to the classification result set, the basic entity types in the graph are determined, the semantic edge relationship between entities in the graph is established, all entities and their semantic connections are organized into a triple form in the graph structure, and the graph structure is standardized. The initial structure of the graph is constructed, and the semantic reasoning rules are defined; The semantic graph structure is standardized, the knowledge graph embedding model is used to train the graph, each entity and relationship is mapped to a unified low-dimensional vector space, and a density-based clustering algorithm is used to cluster all archive entities. According to the clustering results, the clustering centers and corresponding entities are divided and constructed into a hierarchical semantic index tree. The upper nodes of the tree represent semantic clustering centers, each center node connects its subordinate entity nodes or sub-cluster nodes, and a multi-layer index tree is constructed by controlling the branch number and depth of each layer.

3. The method for intelligent monitoring management based on archive digitization according to claim 2, characterized in that: The multi-modal BERT model is used to fuse features of different data sources to generate a unified semantic representation. The preprocessed images are uniformly divided into fixed-size image blocks, and each image block is assigned corresponding position information representation according to its spatial position in the whole image, thereby generating an image block feature matrix containing image local content and spatial position information. A multi-layer encoding structure based on attention mechanism is used to model the image block sequence and extract semantic correlation and global feature expression in the image mode. The preprocessed text data is decomposed into a series of word or sub-word units, and the BERT model is used to map the tokenized text into vector representation. Each token obtains a fixed-dimension embedding vector to form an initial text feature matrix. A deep language encoder is used to model the text sequence and obtain text feature representation containing context semantics. After the image feature sequence and the text feature sequence are encoded and mapped to a unified semantic space, information interaction is realized through a cross-modal attention mechanism, and finally aggregated into a text-image fusion feature vector F. The preprocessed structured fields are spliced into a structured feature vector, which is input into the hybrid modeling path. The optimal modeling strategy is selected for different data types of the fields, the feature results are spliced and fused, and the semantic embedding vector Q of the structured fields is obtained by inputting into a multi-layer perceptron encoder. The text-image fusion feature vector F and the structured semantic embedding vector Q are subjected to vector dimension alignment processing and splicing operation to construct a joint representation vector A.

4. The method for intelligent monitoring management based on archive digitization according to claim 3, characterized in that: The changes of each archive version are recorded, the change position is located based on the semantic index tree, and the semantic changes of the archive are identified and recorded by a semantic difference comparison algorithm. When the archive is detected to be modified, rewritten or updated, the constructed semantic index tree is immediately called to perform embedding vector coding on the changed content, and semantic similarity calculation is performed with the embedding vector of the path node where the original archive is located in the index tree to determine the nearest path node of the change point in the index tree, and the change anchor point of the semantic position of the archive is located. The current changed content and the content of the previous version of the archive are extracted, and the unified semantic coding model is used to compare the distance of the two versions in the sentence vector or paragraph vector space, and the overall semantic difference is calculated by combining the label path change of the key field. The current changed content is generated as a structured snapshot, and full-content hash calculation is performed.

5. The method for intelligent monitoring management based on archive digitization according to claim 4, characterized in that: The semantic boundaries of the nodes at each level in the index tree are used to divide the archive content by granularity and generate user access strategies. The semantic breakpoints are identified and divided into different levels of semantic segments by analyzing the similarity of the semantic vectors between the nodes. Each archive version is bound to the corresponding semantic level in the corresponding index path, and the permission vector is generated according to the user's organization, role and access scenario. The vector is compared with each semantic segment, and the user's access permission is determined by the similarity threshold.

6. The method for intelligent monitoring management based on archive digitization according to claim 5, characterized in that: The potential abnormal behavior is identified through behavior graph analysis, and time series analysis and fuzzy mining algorithm are used for risk prediction and behavior warning. By recording the semantic path nodes, access time and file version when the user accesses the archive, a user-centered behavior graph is continuously constructed, the behavior characteristics of the user in each semantic category, such as access frequency, path jump mode and access depth, are modeled and analyzed, the behavior preference and regular pattern are identified, when the user behavior appears different from the historical pattern, the access path and the access density mutation characteristics, based on the behavior graph, a time series model is constructed, and a fuzzy mining algorithm is used to identify the potential abnormal behavior pattern, and a corresponding behavior warning process is generated.

7. The method for intelligent monitoring management based on archive digitization according to claim 6, characterized in that: The behavior warning process includes immediately freezing the user's current access permission or forcing it to read-only mode, automatically generating structured alarm information, pushing it to the administrator interface, recording the complete behavior link, and writing the risk event into the chain log.

8. An intelligent monitoring management system based on archive digitization, based on the intelligent monitoring management method based on archive digitization of any one of claims 1-7, characterized in that: It includes, A semantic fusion module is used to collect multi-source archive data and perform structured cleaning, and a multi-modal BERT model is used to generate a unified semantic representation; A graph construction module is used to perform semantic labeling on the archive content through a multi-label classification model, and to construct a graph database to represent the semantic association relationship between archives; A semantic index construction module is used to construct a semantic index tree based on the semantic graph, realize fast positioning of the archive, and record the version change path and hash information of each semantic change; A permission policy module is used to divide the archive content granularity according to the semantic boundary in the index tree, and generate a fine-grained user access permission policy; A risk warning module is used to record user access behavior to construct a behavior graph, and to identify abnormal behavior and generate a risk warning signal by combining time series analysis and fuzzy mining. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The processor executes the computer program to realize the steps of the intelligent monitoring management method based on archive digitization according to any one of claims 1-7.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to realize the steps of the intelligent monitoring management method based on archive digitization according to any one of claims 1-7.

Images