Archive retrieval method and system based on cloud computing

Abstract

The invention discloses an archive retrieval method and system based on cloud computing, and the method comprises the steps: obtaining the original data of a drawing document and a technical document through a data collection module of a cloud computing platform, and carrying out the multi-source heterogeneous data fusion, the original data including a CAD drawing, an engineering drawing, a technical specification, a design specification and a product manual.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of archive retrieval, more specifically, to an archive retrieval method and system based on cloud computing. BACKGROUND

[0002] With the advancement of Industry 4.0 and digital transformation, enterprises have generated a large amount of drawing documents and technical documents, which have the characteristics of diversity, complexity and correlation. The traditional archive management system has the following problems:

[0003] Data island problem: The archive data generated by different departments and systems lacks unified management and retrieval mechanism, forming information islands.

[0004] Low retrieval efficiency: The traditional keyword retrieval method is difficult to understand the real intention of the user, and the relevance of the retrieval result is poor.

[0005] Insufficient semantic understanding: The existing system cannot understand the professional terms and semantic relationships in the drawing and technical documents.

[0006] Expansion limitation: The traditional system has performance bottleneck and expansion problem when facing large-scale data.

[0007] The document retrieval method in the prior art only processes text documents and cannot process multi-modal data such as drawings.

[0008] In view of this, we propose an archive retrieval method and system based on cloud computing. SUMMARY

[0009] 1. Technical problem to be solved

[0010] The purpose of the present application is to provide an archive retrieval method based on cloud computing, which solves the technical problems proposed in the background art.

[0011] 2. Technical solution

[0012] The present application provides an archive retrieval method based on cloud computing, comprising the following steps:

[0013] S1: Through the data acquisition module of the cloud computing platform, the original data of the drawing documents and technical documents are obtained, and multi-source heterogeneous data fusion is performed, the original data including CAD drawings, engineering drawings, technical specifications, design manuals and product manuals;

[0014] S2: Adopting a distributed node classification index method based on content awareness, intelligently sharding according to the technical complexity, access frequency and correlation density of the documents, distributing the data to multiple storage nodes of the cloud computing platform through a consistent hashing algorithm, and establishing a dynamic load balancing mechanism across nodes;

[0015] S3: Through the multi-modal fusion of unstructured data structured processing module, combined with image recognition, OCR text recognition and natural language processing technology, extract the geometric features in the drawing, size annotation and semantic information in the technical document, and establish a multi-level feature representation model;

[0016] S4: Based on the domain knowledge driven triple knowledge graph construction method, a multi-dimensional knowledge graph containing technical entities, attribute relationships and time sequence relationships is constructed, the triple uses the semantic structure of <entity-relation-entity>, and the entity alignment and relation reasoning are performed through the graph neural network;

[0017] S5: Establish a knowledge graph index library based on graph embedding, use TransE algorithm for vectorization representation of entities and relationships, and realize multi-hop semantic reasoning through attention mechanism;

[0018] S6: According to the user search request, through the hybrid reasoning engine of fusion semantic similarity and structural similarity, realize the intelligent retrieval based on intention understanding, and provide the explainable retrieval result path.

[0019] Further, the content-aware distributed node classification index in step S2 specifically includes:

[0020] S21: Establish a document complexity evaluation model, calculate the comprehensive complexity score based on the file size, image complexity, text density and reference relationship number of the document;

[0021] S22: Construct an access mode prediction model, analyze user behavior patterns through historical access data, and predict the access heat and time distribution of the document;

[0022] S23: Use an improved consistent hashing algorithm, combine document attributes and node performance indicators, realize intelligent sharding and dynamic migration of data;

[0023] S24: Establish a multi-copy consistency protocol, ensure the strong consistency of cross-node data through Raft algorithm, and realize the automatic recovery mechanism of failure.

[0024] Further, the multi-modal fusion of unstructured data structured processing in step S3 specifically includes:

[0025] S31: For drawing documents, use deep learning-based image segmentation and object detection technology to identify parts, annotation information and symbol elements in the drawing, and establish a hierarchical representation of drawing elements;

[0026] S32: For technical documents, use BERT-based named entity recognition and relationship extraction technology to identify technical terms, parameter values, specification standards and design requirements;

[0027] S33: Establish a domain-adaptive technical terminology ontology library, and realize semantic alignment and standardization of the terminology through word2vec and knowledge graph embedding technology;

[0028] S34: Adopt a multi-modal feature fusion network to jointly represent image features, text features and structural features, and generate a unified document semantic vector.

[0029] Further, the step S4 of constructing the triple knowledge graph based on domain knowledge driving specifically includes:

[0030] S41: Define a multi-level ontology model of the archives management domain, including a concept layer, an instance layer and a rule layer, and establish a mapping relationship between ontologies;

[0031] S42: Adopt an entity alignment method based on graph neural network, and realize identification and linking of the same entity in heterogeneous data sources through joint optimization of structural similarity and attribute similarity;

[0032] S43: Through a relationship extraction method combining remote supervision learning and active learning, automatically discover the implicit relationship between entities, and establish a confidence evaluation mechanism for the relationship;

[0033] S44: Construct a time-series knowledge graph to record the time evolution process of entities and relationships, and support version control and historical tracing;

[0034] S45: Adopt a knowledge graph quality evaluation and optimization method based on reinforcement learning, and realize continuous improvement of the knowledge graph through reward function design.

[0035] Further, the step S6 of the hybrid reasoning engine fusing semantic similarity and structural similarity specifically includes:

[0036] S61: Adopt a query intent recognition model based on attention mechanism to identify explicit and implicit requirements of user queries, and generate vector representation of the query;

[0037] S62: Construct a query subgraph through a graph convolution network, and adopt a subgraph matching algorithm to find similar subgraph structures in the knowledge graph;

[0038] S63: Calculate the comprehensive relevance score combining semantic similarity and structural similarity, wherein the semantic similarity is based on word vector and entity embedding, and the structural similarity is based on graph edit distance;

[0039] S64: Adopt a multi-objective optimization algorithm to sort the retrieval results, considering relevance, diversity and novelty;

[0040] S65: Generate an interpretable search result, show the search reasoning process through the knowledge graph path, and provide an explanation of the relevance score.

[0041] A cloud computing-based archive search system, comprising:

[0042] A data acquisition module for acquiring multi-source heterogeneous drawing documents and technical document data;

[0043] A content-aware distributed node classification index module for intelligent sharding and dynamic load balancing based on document features;

[0044] A multi-modal fusion unstructured data structured processing module for extracting multi-level document feature representations;

[0045] A domain knowledge-driven triple knowledge graph construction module for constructing a multi-dimensional time-series knowledge graph;

[0046] A graph embedding-based knowledge graph index library for storing vectorized entity and relationship representations;

[0047] A hybrid reasoning engine that combines semantic similarity and structural similarity to enable intelligent search and interpretable reasoning.

[0048] Further, the distributed node classification index module comprises:

[0049] A storage node management unit for managing multiple storage nodes and load balancing of the cloud computing platform;

[0050] A document classification unit for multi-dimensional classification based on document type, engineering project, technical field, and timestamp;

[0051] A distributed identifier generation unit for generating a unique distributed identifier for each document;

[0052] An index synchronization unit for cross-node metadata synchronization and index information updating.

[0053] Further, the unstructured data structured processing module comprises:

[0054] An image recognition unit for image recognition and OCR text recognition of drawing documents;

[0055] A natural language processing unit for natural language processing and technical term recognition of technical documents;

[0056] A term standardization unit for establishing a unified technical term dictionary and performing standardization processing;

[0057] A data format conversion unit for converting unstructured information into a structured data format.

[0058] Further, the triple knowledge graph construction module comprises:

[0059] An ontology model definition unit is configured to define an ontology model and a relationship type in the field of archives management;

[0060] An entity recognition unit is configured to recognize entity objects and perform entity linking;

[0061] A relationship extraction unit is configured to recognize semantic relationships between entities and construct a triple structure;

[0062] A graph database storage unit is configured to store a triple knowledge graph and establish a graph query index;

[0063] A knowledge reasoning unit is configured to discover implicit entity relationships through reasoning rules.

[0064] Further, the semantic retrieval engine comprises:

[0065] A query analysis unit is configured to analyze a natural language retrieval request of a user and extract keywords;

[0066] A semantic mapping unit is configured to map retrieval keywords to entity nodes in a knowledge graph;

[0067] A graph traversal unit is configured to find an entity path related to a query in a knowledge graph;

[0068] A similarity calculation unit is configured to calculate semantic similarity between a query entity and a candidate document entity;

[0069] A result sorting unit is configured to sort retrieval results based on similarity and return the results.

[0070] 3. Advantages

[0071] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0072] The present application improves retrieval accuracy through multi-modal fusion and knowledge graph reasoning; the present application enhances system scalability: a distributed architecture supports PB-level data storage and retrieval; the present application improves user experience: provides interpretable retrieval results and intelligent recommendations; the present application reduces operation and maintenance costs: automatic data management and fault recovery mechanisms. BRIEF DESCRIPTION OF DRAWINGS

[0073] Figure 1 A flowchart of an archives retrieval system based on cloud computing is disclosed in a preferred embodiment of the present application; DETAILED DESCRIPTION

[0074] The application will be further described in detail below with reference to the accompanying drawings.

[0075] Referring to Figure 1 A cloud computing-based archive retrieval method, comprising the following steps:

[0076] S1: Obtain the original data of drawing documents and technical documents through the data acquisition module of the cloud computing platform, and perform multi-source heterogeneous data fusion, wherein the original data includes CAD drawings, engineering drawings, technical specifications, design manuals and product manuals;

[0077] S2: Adopt a content-aware distributed node classification indexing method, intelligently shard according to the technical complexity, access frequency and correlation density of the documents, distribute the data to multiple storage nodes of the cloud computing platform through a consistent hashing algorithm, and establish a dynamic load balancing mechanism across nodes;

[0078] S3: Through a multi-modal fusion unstructured data structured processing module, combine image recognition, OCR text recognition and natural language processing technology to extract geometric features, size annotations in drawings and semantic information in technical documents, and establish a multi-level feature representation model;

[0079] S4: Based on the domain knowledge driven triple knowledge graph construction method, a multi-dimensional knowledge graph containing technical entities, attribute relationships and time sequence relationships is constructed, the triple uses a semantic structure of <entity-relation-entity>, and entity alignment and relation reasoning are performed through a graph neural network;

[0080] S5: Establish a knowledge graph index library based on graph embedding, use the TransE algorithm to perform vectorization representation of entities and relationships, and realize multi-hop semantic reasoning through an attention mechanism;

[0081] S6: According to the user retrieval request, through a hybrid reasoning engine that fuses semantic similarity and structural similarity, intelligent retrieval based on intent understanding is realized, and an interpretable retrieval result path is provided.

[0082] As an embodiment of the application, the content-aware distributed node classification indexing in step S2 specifically includes:

[0083] S21: Establish a document complexity evaluation model, calculate the comprehensive complexity score based on the file size, image complexity, text density and reference relationship number of the document;

[0084] S22: Construct an access pattern prediction model, analyze user behavior patterns through historical access data, and predict the access heat and time distribution of the document;

[0085] S23: Use the improved consistent hashing algorithm, combined with document attributes and node performance indicators, to realize intelligent data sharding and dynamic migration;

[0086] S24: Establish a multi-copy consistency protocol to ensure strong consistency of cross-node data through the Raft algorithm and implement an automatic failure recovery mechanism.

[0087] As an embodiment of the present application, the multi-modal fusion unstructured data structuring process in step S3 specifically includes:

[0088] S31: Use deep learning-based image segmentation and object detection techniques on drawing documents to identify parts, annotations, and symbol elements in the drawings, and establish a hierarchical representation of drawing elements;

[0089] S32: Use BERT-based named entity recognition and relationship extraction techniques on technical documents to identify technical terms, parameter values, specifications, and design requirements;

[0090] S33: Establish a domain-adaptive technical term ontology library to achieve semantic alignment and standardization of terms through word2vec and knowledge graph embedding techniques;

[0091] S34: Use a multi-modal feature fusion network to jointly represent image features, text features, and structural features to generate a unified document semantic vector.

[0092] As an embodiment of the present application, the domain knowledge-driven triple knowledge graph construction in step S4 specifically includes:

[0093] S41: Define a multi-level ontology model for the archival management domain, including a concept layer, an instance layer, and a rule layer, and establish mapping relationships between ontologies;

[0094] S42: Use a graph neural network-based entity alignment method to identify and link the same entity in heterogeneous data sources through joint optimization of structural similarity and attribute similarity;

[0095] S43: Use a relationship extraction method combining remote supervision learning and active learning to automatically discover implicit relationships between entities and establish a confidence evaluation mechanism for relationships;

[0096] S44: Construct a time-series knowledge graph to record the time evolution process of entities and relationships, supporting version control and historical tracing;

[0097] S45: Use a reinforcement learning-based knowledge graph quality evaluation and optimization method to achieve continuous improvement of the knowledge graph through reward function design.

[0098] As an embodiment of the present application, the hybrid reasoning engine that fuses semantic similarity and structural similarity in step S6 specifically includes:

[0099] S61: Adopting a query intent recognition model based on attention mechanism, identifying explicit and implicit requirements of user query, and generating vector representation of query;

[0100] S62: Constructing query subgraph through graph convolution network, and adopting subgraph matching algorithm to find similar subgraph structure in knowledge graph;

[0101] S63: Combining semantic similarity and structural similarity to calculate comprehensive relevance score, wherein semantic similarity is based on word vector and entity embedding, and structural similarity is based on graph edit distance;

[0102] S64: Adopting multi-objective optimization algorithm to sort retrieval results, considering relevance, diversity and novelty;

[0103] S65: Generating interpretable retrieval results, showing retrieval reasoning process through knowledge graph path, and providing explanation of relevance score.

[0104] An archive retrieval system based on cloud computing, comprising:

[0105] A data acquisition module for acquiring multi-source heterogeneous drawing and technical document data;

[0106] A distributed node classification index module based on content awareness for intelligent sharding and dynamic load balancing according to document features;

[0107] A multi-modal fusion unstructured data structured processing module for extracting multi-level document feature representation;

[0108] A triple knowledge graph construction module based on domain knowledge driving for constructing multi-dimensional time series knowledge graph;

[0109] A knowledge graph index library based on graph embedding for storing vectorized entity and relationship representation;

[0110] A hybrid reasoning engine that fuses semantic similarity and structural similarity for realizing intelligent retrieval and interpretable reasoning.

[0111] As an embodiment of the present application, the distributed node classification index module includes:

[0112] A storage node management unit for managing multiple storage nodes and load balancing of cloud computing platform;

[0113] A document classification unit for multi-dimensional classification according to document type, engineering project, technical field and time stamp;

[0114] a distributed identifier generation unit configured to generate a unique distributed identifier for each document;

[0115] an index synchronization unit configured to implement cross-node metadata synchronization and index information update.

[0116] As an embodiment of the present application, the unstructured data structuring processing module comprises:

[0117] an image recognition unit configured to perform image recognition and OCR text recognition on the drawing document;

[0118] a natural language processing unit configured to perform natural language processing and technical term recognition on the technical document;

[0119] a term standardization unit configured to establish a unified technical term dictionary and perform standardization processing;

[0120] a data format conversion unit configured to convert the unstructured information into a structured data format.

[0121] As an embodiment of the present application, the triple knowledge graph construction module comprises:

[0122] an ontology model definition unit configured to define an ontology model and relationship types in the field of archives management;

[0123] an entity recognition unit configured to recognize entity objects and perform entity linking;

[0124] a relationship extraction unit configured to recognize semantic relationships between entities and construct a triple structure;

[0125] a graph database storage unit configured to store the triple knowledge graph and establish a graph query index;

[0126] a knowledge reasoning unit configured to discover implicit entity relationships through reasoning rules.

[0127] As an embodiment of the present application, the semantic retrieval engine comprises:

[0128] a query analysis unit configured to analyze a natural language retrieval request of a user and extract keywords;

[0129] a semantic mapping unit configured to map the retrieval keywords to entity nodes in the knowledge graph;

[0130] a graph traversal unit configured to find an entity path related to the query in the knowledge graph;

[0131] a similarity calculation unit configured to calculate semantic similarity between the query entity and candidate document entities;

[0132] A result sorting unit for sorting the search results based on similarity and returning.

[0133] As an embodiment of the present application, it also includes:

[0134] A user permission management module for managing access permissions and operation permissions of different users to the archives;

[0135] A data security module for encrypting sensitive archive data for storage and transmission;

[0136] A system monitoring module for monitoring system running state, performance indicators and abnormal situations;

[0137] A cache optimization module for caching hot query results to improve search response speed.

[0138] As an embodiment of the present application, the method further includes:

[0139] Establishing an archive version control mechanism to track the version change history of drawing documents and technical documents;

[0140] Building an archive association network to identify the reference relationship, dependency relationship and inheritance relationship between different archives;

[0141] Implementing archive lifecycle management, including archive creation, update, review, release and archiving;

[0142] Providing archive statistical analysis functions to generate archive usage, hot spot analysis and trend prediction reports.

[0143] Embodiment 1: Distributed node classification index based on content awareness;

[0144] Document complexity evaluation model: ComplexityScore = α × FileSize + β × ImageComplexity + γ × TextDensity + δ × ReferenceCount Where: -α, β, γ, δ are weight coefficients -FileSize is file size -ImageComplexity is image complexity (based on edge detection and texture analysis) -TextDensity is text density -ReferenceCount is the number of reference relationships

[0145] Embodiment 2: Multi-modal feature fusion network;

[0146] The multi-modal feature fusion network adopts the following architecture:

[0147] Image feature extraction: ResNet-50 is used to extract image features;

[0148] Text feature extraction: using BERT model to extract text features;

[0149] Feature fusion: fusing multi-modal features through attention mechanism;

[0150] Semantic vector generation: generating unified document semantic vector;

[0151] Embodiment 3: Knowledge graph construction process

[0152] Entity recognition: using BiLSTM-CRF model to identify technical entities;

[0153] Relation extraction: using attention mechanism-based relation classification model;

[0154] Entity alignment: using graph neural network for cross-source entity alignment;

[0155] Knowledge fusion: constructing unified knowledge graph.

[0156] The technical scheme of the present application includes content-aware distributed node classification index The traditional distributed storage method is usually based on simple hash algorithm or random allocation, and the method proposed by the present application has the following innovations: a document complexity evaluation model is established, considering factors such as file size, image complexity, and text density; an access pattern prediction model is constructed to predict document access heat based on historical data; an improved consistent hash algorithm is used to realize intelligent sharding and dynamic migration of data.

[0157] The technical scheme of the present application includes multi-modal fusion of unstructured data structure processing The existing technology usually processes image and text data separately, and the innovation of the present application is: using deep learning technology for semantic segmentation and target detection of drawings; combining BERT model for entity recognition and relation extraction of technical documents; realizing joint representation of image and text through multi-modal feature fusion network.

[0158] The technical scheme of the present application includes domain knowledge driven triple knowledge graph construction The traditional knowledge graph construction method lacks domain knowledge guidance, and the present application proposes: a multi-level ontology model, including concept layer, instance layer and rule layer; an entity alignment method based on graph neural network; a time series knowledge graph supporting version control and historical traceability.

[0159] The technical scheme of the present application includes a hybrid reasoning engine that fuses semantic similarity and structural similarity The existing retrieval method usually only considers semantic similarity, and the innovation of the present application is: combining semantic similarity and structural similarity for comprehensive evaluation; using attention mechanism for query intent recognition; generating explainable retrieval result path.

Claims

1. A cloud computing-based document retrieval method, characterized in that, Includes the following steps: S1: The raw data of drawings and technical documents are acquired through the data acquisition module of the cloud computing platform, and multi-source heterogeneous data are fused. The raw data includes CAD drawings, engineering drawings, technical specifications, design instructions and product manuals. S2: Adopt a content-aware distributed node classification indexing method, intelligently shard according to the technical complexity, access frequency and association density of documents, distribute data to multiple storage nodes of the cloud computing platform through a consistent hashing algorithm, and establish a dynamic load balancing mechanism across nodes. S3: Through the unstructured data structuring module of multimodal fusion, combined with image recognition, OCR text recognition and natural language processing technology, it extracts geometric features, dimension annotations and semantic information in technical documents from drawings, and establishes a multi-level feature representation model; S4: A domain knowledge-driven triplet knowledge graph construction method is used to construct a multi-dimensional knowledge graph containing technical entities, attribute relations, and temporal relations. The triplet adopts the semantic structure of <entity-relationship-entity> and performs entity alignment and relation reasoning through graph neural networks. S5: Establish a knowledge graph index based on graph embedding, use the TransE algorithm to vectorize entities and relations, and achieve multi-hop semantic reasoning through an attention mechanism; S6: Based on the user's search request, it realizes intelligent search based on intent understanding through a hybrid reasoning engine that integrates semantic similarity and structural similarity, and provides interpretable search result paths.

2. The cloud computing-based archive retrieval method according to claim 1, characterized in that, The content-aware distributed node classification index in step S2 specifically includes: S21: Establish a document complexity evaluation model and calculate a comprehensive complexity score based on the document's file size, image complexity, text density, and number of references; S22: Build an access pattern prediction model to analyze user behavior patterns through historical access data and predict the access popularity and time distribution of documents; S23: An improved consistent hashing algorithm is adopted, combined with document attributes and node performance indicators, to achieve intelligent data sharding and dynamic migration; S24: Establish a multi-replica consistency protocol, ensure strong consistency of data across nodes through the Raft algorithm, and implement an automatic fault recovery mechanism.

3. The cloud computing-based archive retrieval method according to claim 1, characterized in that, The unstructured data structuring process for multimodal fusion in step S3 specifically includes: S31: Use deep learning-based image segmentation and object detection technology to identify parts, annotation information and symbol elements in the drawings, and establish a hierarchical representation of the drawing elements; S32: Use BERT-based named entity recognition and relation extraction technology to identify technical terms, parameter values, specifications and standards and design requirements in technical documents; S33: Establish a domain-adaptive technical terminology ontology, and achieve semantic alignment and standardization of terms through word2vec and knowledge graph embedding technologies; S34: Employs a multimodal feature fusion network to jointly represent image features, text features, and structural features, generating a unified document semantic vector.

4. The cloud computing-based archive retrieval method according to claim 1, characterized in that, The construction of the domain knowledge-driven triple knowledge graph in step S4 specifically includes: S41: Define a multi-layered ontology model for the field of records management, including a concept layer, an instance layer, and a rule layer, and establish mapping relationships between ontology. S42: An entity alignment method based on graph neural networks is adopted to identify and link the same entity in heterogeneous data sources through joint optimization of structural similarity and attribute similarity. S43: By combining remote supervised learning and active learning, implicit relationships between entities are automatically discovered, and a relationship confidence assessment mechanism is established. S44: Construct a temporal knowledge graph to record the temporal evolution of entities and relationships, supporting version control and historical traceability; S45: Employ a reinforcement learning-based knowledge graph quality assessment and optimization method, and achieve continuous improvement of the knowledge graph through reward function design.

5. The cloud computing-based archive retrieval method according to claim 1, characterized in that, The hybrid reasoning engine that integrates semantic similarity and structural similarity in step S6 specifically includes: S61: Employ an attention-based query intent recognition model to identify the explicit and implicit needs of user queries and generate a vector representation of the query. S62: Construct query subgraphs through graph convolutional networks and use subgraph matching algorithms to find similar subgraph structures in the knowledge graph; S63: Calculate the comprehensive relevance score by combining semantic similarity and structural similarity, where semantic similarity is based on word vectors and entity embeddings, and structural similarity is based on graph edit distance; S64: Use a multi-objective optimization algorithm to rank the search results, taking into account relevance, diversity and novelty; S65: Generate interpretable search results, demonstrate the search reasoning process through knowledge graph paths, and provide an explanation of the relevance score.

6. A cloud computing-based archive retrieval system, characterized in that, include: The data acquisition module is used to acquire multi-source heterogeneous drawing documents and technical document data; A content-aware distributed node classification indexing module is used for intelligent sharding and dynamic load balancing based on document characteristics. A multimodal fusion module for structuring unstructured data is used to extract multi-level document feature representations. A domain knowledge-driven triplet knowledge graph construction module is used to construct multidimensional temporal knowledge graphs. A knowledge graph index library based on graph embedding, used to store vectorized entity and relation representations; A hybrid reasoning engine that integrates semantic similarity and structural similarity is used to achieve intelligent retrieval and interpretable reasoning.

7. The cloud-based archive retrieval system according to claim 6, characterized in that, The distributed node classification index module includes: The storage node management unit is used to manage multiple storage nodes and load balancers on a cloud computing platform. Document classification units are used to classify documents in multiple dimensions, such as document type, project, technical field, and timestamp. A distributed identifier generation unit is used to generate a unique distributed identifier for each document. The index synchronization unit is used to synchronize metadata and update index information across nodes.

8. The cloud-based archive retrieval system according to claim 6, characterized in that, The unstructured data structuring module includes: The image recognition unit is used to perform image recognition and OCR text recognition on drawings and documents. The natural language processing unit is used to perform natural language processing and technical terminology recognition on technical documents. The terminology standardization unit is used to establish a unified technical terminology dictionary and perform standardization processing. The data format conversion unit is used to convert unstructured information into a structured data format.

9. The cloud-based archive retrieval system according to claim 6, characterized in that, The triplet knowledge graph construction module includes: The ontology model definition unit is used to define the ontology model and relation types in the field of records management; The entity recognition unit is used to identify entity objects and link entities. The relation extraction unit is used to identify semantic relationships between entities and construct triple structures. The graph database storage unit is used to store triple knowledge graphs and build graph query indexes. The knowledge reasoning unit is used to discover implicit entity relationships through reasoning rules.

10. The cloud-based archive retrieval system according to claim 6, characterized in that, The semantic retrieval engine includes: The query parsing unit is used to parse the user's natural language search request and extract keywords; Semantic mapping unit, used to map search keywords to entity nodes in the knowledge graph; The graph traversal unit is used to find and query related entity paths in the knowledge graph; The similarity calculation unit is used to calculate the semantic similarity between the query entity and the candidate document entity; The results sorting unit is used to sort and return the search results based on similarity.

Images