Intelligent optimization management method applied to digital archive feature data mining extraction

Abstract

This invention discloses an intelligent optimization management method for feature data mining and extraction in digital archives, belonging to the field of digital information processing technology. It effectively overcomes the inherent defects of traditional digital archive processing, such as fixed and rigid processes, low processing accuracy, missing associations, and inability to evolve, by constructing an optimization management framework integrating intelligent perception, multi-engine parallel processing verification, multi-modal feature fusion, and knowledge graph mining. This significantly improves the accuracy and depth of archive information extraction, achieves intelligent and adaptive processing, automatically constructs semantic associations and knowledge networks between archives, and provides intelligent retrieval services based on semantic understanding and graph association. It achieves a leap from "single keyword matching" to "intent understanding and related recommendation," thereby completing a leapfrog upgrade of digital archive management from "digital storage" to "intelligent knowledge service."

Description

Technical Field

[0001] This invention relates to the field of digital information processing technology, and more specifically, to an intelligent optimization management method applied to the mining and extraction of feature data from digital archives. Background Technology

[0002] Digital archives typically contain a large amount of unstructured or semi-structured data, such as scanned documents, images, and text. Today, with the deepening of digital transformation, massive amounts of digital archives (including scanned images, native electronic documents, audio, and video) have become core information assets for various institutions. Digital archives have also shifted from "preservation" to "utilization." To realize the aforementioned values ​​such as deep structuring, knowledge discovery, and empowerment of intelligent applications, it is necessary to extract feature data from archival information.

[0003] A search revealed that patent number CN118245652A discloses an artificial intelligence-based archive management system. This system uses an intelligent acquisition and input module to efficiently convert and automatically input archive data from physical to digital form, automatically classify archives and generate keyword tags, and uses a natural language processing module to provide accurate semantic search services and track the system's operating status and archive activities in real time.

[0004] However, this patent is mainly a traditional AI application that is modular, linear, and focused on automating fixed processes. It has a single dimension of feature extraction, does not involve real-time interactive verification of multimodal information, and lacks the ability to deeply understand the semantics of the archive content. Secondly, the flat organization based on classification and keywords is to label the archives with categories and keywords through classification models and keyword extraction. When searching, it relies on these tags and full-text indexes, and cannot discover the hidden patterns, relationships and knowledge behind the data.

[0005] To overcome the aforementioned shortcomings, this invention aims to provide an intelligent, adaptive, and optimizable method for mining, extracting, and managing digital archival feature data. The core technology lies in constructing a collaborative technology chain of "dynamic perception and multimodal fusion → deep semantic organization and association → group knowledge mining → intelligent graph services." This enhances the accuracy and intelligence of digital archival information processing, enabling the evolution of archival information from shallow labels to deep semantic association networks. It also automatically constructs semantic associations and knowledge networks between archives, providing intelligent retrieval services based on semantic understanding and graph association, thus achieving a leap from "single keyword matching" to "intent understanding and related recommendation." Summary of the Invention

[0006] The purpose of this invention is to solve existing practical problems and provide an intelligent optimization management method for feature data mining and extraction of digital archives compared with existing technologies.

[0007] The objective of this invention can be achieved through the following technical solution: an intelligent optimization management method applied to feature data mining and extraction of digital archives, comprising the following steps: S1. Archive Collection and Preprocessing: Receive the input raw archives, perform dynamic perception preprocessing operations, and output basic structured text content with multimodal fusion features. S2. Archive organization and arrangement: Based on the basic structured text content, multimodal fusion features are extracted and combined with NLP technology to automatically generate richer and deeper tags and relationships, complete the intelligent organization and arrangement of individual archives, generate the initial structured archive information of the complete set of archives rich in semantic tags and basic relationships, and store it in the database. S3. Group Archive Feature Mining and Analysis: Based on the initial structured information of the complete archives obtained in S2, a deeper level of in-depth mining and analysis is performed with the goal of discovering group knowledge and complex patterns, and high-level feature data and knowledge results are output. S4. Archive Application and Services: Wait for user query operations, obtain the user's original query, and return the archive collection in advanced feature data and knowledge results by combining deep semantic analysis, knowledge graph and multi-level index. It also provides graph-based context navigation and knowledge discovery services, and outputs an information-rich and related visualized search results interface. Users can click on entities or tags in the graph to retrieve their subsequent queries. Based on the visual graph navigation, users can jump from one file to the entire related knowledge network.

[0008] Furthermore, the process of performing dynamic sensing preprocessing on the original archive in S1 includes: format conversion and image enhancement of the original archive to obtain the pre-processed digital image / file; For digital images / files, content information and quality indicators are obtained. Based on the content information and quality indicators, a dedicated OCR model is calibrated. Based on a dedicated OCR engine and an image recognition engine, digital images / files are processed in parallel and cross-modal. The output is a basic structured text content that combines text content, visual elements and the relationship between the two.

[0009] Furthermore, the process of forming the initial structured archive information in S2 specifically includes: Deep semantic analysis is performed on the basic structured text content and multimodal fusion features output in S1 to automatically extract entities, keywords, topics and attributes. Based on the semantic analysis results, classification tags and metadata are intelligently generated to build a multi-level index that supports full text, semantics and entities. By identifying common semantic elements between different archives, a basic relationship network between archives is automatically established to finally generate initial structured archive information.

[0010] Furthermore, the initial structured archive information includes: unique identifier content, original archive file references, core metadata, enhanced content annotations, multimodal fusion feature references, processing traceability data, and related links.

[0011] Furthermore, the specific process of feature mining and analysis of digital archives in S3: S31. Deep Semantic Mining and Knowledge Extraction: Taking the initial structured information generated in S2 as input, the system first extracts precise entities, relationships, and events through deep semantic analysis, constructs a deep relationship network, and completes a unified link to the entire entity database. S32. Global Knowledge Graph Construction and Pattern Discovery: Based on the semantic information above, a global knowledge graph spanning multiple archives is constructed, and graph mining techniques are used to discover the implicit community structure, key nodes, and abnormal patterns within it. S33. Knowledge Fusion, Verification and Productization: The semantic mining results are fused and verified with the previous multimodal physical features to generate high-level feature data with confidence assessment, a visualized knowledge graph and a programmable application interface, and output high-level feature data and knowledge results.

[0012] Furthermore, based on semantic information, the specific process of constructing a global knowledge graph across archives includes: integrating all entities, relationships, and events extracted in S31 using a graph data structure; merging the same entity mentioned in different archives into a node in the graph; and constructing the relationships between entities to form a semantic network that comprehensively reflects people, events, objects, organizations, and their relationships within the archives, outputting a queryable, reasonable, and cross-archive global knowledge graph.

[0013] Furthermore, the specific content of advanced feature data and knowledge results includes: computable advanced feature datasets, interactive / visualized knowledge graphs and reports, structured analytical conclusions and insights, and knowledge service APIs that support system integration.

[0014] Furthermore, the specific processes for applications and services in S4 include: S41, Intelligent Search Interface and Intent Understanding: Users can initiate original queries through natural language questions, keyword combinations, or filtering conditions. The system performs entity recognition, relation extraction, and intent classification on the original queries, and links the identified entities with the full-domain entity database built by S3 to ensure semantic consistency. S42. Multi-engine parallel search and fusion sorting: Based on the parsed query intent, the system calls multi-level indexes in parallel, retrieves candidate result sets from the multi-level indexes, uses a machine learning sorting model, and intelligently sorts the candidate result sets by combining multiple factors to return a set of archives that are related to keywords, semantics, and knowledge. S43. Enhance the context and present knowledge associations of the archive collection: encapsulate and expand the information of the returned archive collection, and provide graph-based context navigation and knowledge discovery services, outputting an information-rich and visually associated search results interface. S44. Interactive Graph Query: Users can click on any entity, tag, or keyword on the results page, and the system will immediately treat it as a new query intent, repeat the search steps, and output a "guided, associative, and exploratory" immersive file browsing and knowledge discovery experience.

[0015] Compared with the prior art, the advantages of this invention are: 1. This invention is based on a multi-engine parallel processing verification preprocessing mechanism, which significantly improves the accuracy and robustness of extracting text and visual information from complex, low-quality original archives. It also utilizes NLP and multimodal fusion to achieve deep semantic understanding and annotation of archive content, automatically generating rich entity, topic, and relationship tags, giving the archives machine-understandable "semantics". Based on the initial structured information of the complete set of archives, it performs deeper mining analysis aimed at discovering collective knowledge and complex patterns, establishing explicit and implicit connections between archives and between entities within archives, forming a knowledge graph. This achieves a leapfrog upgrade of archive management from "digital storage" to "intelligent knowledge service", transforming massive amounts of dormant raw data into computable, associative, and insightful high-level feature data and knowledge results.

[0016] 2. Based on the above, intelligent retrieval based on semantic understanding, knowledge graphs, and multi-level indexes returns a collection of archives from advanced feature data and knowledge results. It also provides graph-based contextual navigation and knowledge discovery services, outputting an information-rich and visually relevant search results interface. Users can jump from one archive to the entire related knowledge network by clicking on any entity or tag, based on the visual graph navigation. This achieves a leap from "single keyword matching" to "intent understanding and related recommendation", significantly improving recall and precision. Attached Figure Description

[0017] Figure 1 The method flow of the present invention Figure 1 ; Figure 2 The method flow of the present invention Figure 2 . Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0019] Example 1: This invention discloses an intelligent optimization management method for feature data mining and extraction from digital archives. Please refer to [link to relevant documentation]. Figures 1-2 It includes the following steps: S1. Archive Collection and Preprocessing: Receive the input raw archives, perform dynamic perception preprocessing operations, and output basic structured text content with multimodal fusion features. S2. Archive organization and arrangement: Based on the basic structured text content, multimodal fusion features are extracted and combined with NLP technology to automatically generate richer and deeper tags and relationships, complete the intelligent organization and arrangement of individual archives, generate the initial structured archive information of the complete set of archives rich in semantic tags and basic relationships, and store it in the database. S3. Group Archive Feature Mining and Analysis: Based on the initial structured information of the complete archives obtained in S2, a deeper level of in-depth mining and analysis is performed with the goal of discovering group knowledge and complex patterns, and high-level feature data and knowledge results are output. S4. Archive Application and Services: Wait for user query operations, obtain the user's original query, and return the archive collection in advanced feature data and knowledge results by combining deep semantic analysis, knowledge graph and multi-level index. It also provides graph-based context navigation and knowledge discovery services, and outputs an information-rich and related visualized search results interface. Users can click on entities or tags in the graph to obtain subsequent queries. Based on the visual graph navigation, users can jump from one file to the entire related knowledge network, which meets users' specific needs for querying, analysis, and decision-making, and greatly improves the recall, precision and knowledge utilization of file information.

[0020] The process of performing dynamic sensing preprocessing on the original files in S1 includes: Perform format conversion (such as converting to standard resolution TIFF or PNG) and image enhancement (such as noise reduction, skew correction, and contrast adjustment) on the original archives to obtain the pre-processed digital image / file; For digital images / files, the system acquires content information and quantifies quality indicators. Based on the content information and quality indicators, it calibrates a dedicated OCR model. This means that the system can intelligently select and call the most suitable OCR engine or model based on the characteristics of the archive (such as font, language, age, and clarity). For example, there is a dedicated OCR for ancient books, a general OCR for printed text, and a dedicated OCR for handwritten text. Based on a dedicated OCR engine and image recognition engine, parallel processing and cross-modal verification of digital images / files are performed to output basic structured text content that combines text content, visual elements and the relationship between the two with multimodal fusion features. Specifically, the selected OCR engine is scheduled to recognize the text region, and the image recognition engine is used to recognize the image region. After receiving the text recognition result from OCR and the visual element result from image recognition, the position of the "text content" recognized by OCR and the "visual element" recognized by image on the page is compared. A decision tree based on spatial relationship and semantic logic is executed. For independent elements that are spatially adjacent or semantically related, the association between them is established. This integrates text, visual elements and their precise positions, which greatly improves the parsing efficiency and accuracy. Through "dynamic perception preprocessing" and "multimodal fusion", the system can automatically adapt to the optimal processing flow according to the quality and content of the archives, and cross-validate the text and visual information, which greatly overcomes the problem of traditional OCR's weak ability to process complex and low-quality archives, and produces basic structured content with high confidence.

[0021] The process of forming the initial structured archive information in S2 specifically includes: Deep semantic analysis is performed on the basic structured text content and multimodal fusion features output in S1 to automatically extract entities, keywords, topics and attributes. Based on the semantic analysis results, classification labels and metadata are intelligently generated to build a multi-level index that supports full text, semantics and entities. Relatively standard NLP tasks are performed to add standardized labels for classification and indexing. By identifying common semantic elements between different archives, a basic association network between archives is automatically established. The basic association network is based on explicit, direct and clear common semantic elements (such as the same entity name and the same item number) to establish links, and finally the initial structured archive information is generated. The process explicitly states that natural language processing (NLP) technology is used to perform in-depth analysis of the text content extracted by OCR, such as entity recognition, relation extraction, and topic classification, thereby achieving intelligent organization and feature mining of archival information. Specifically, it is an intelligent information organization process that takes multimodal fusion features as input, is driven by NLP technology, and aims to build an association network. It is responsible for performing entity recognition, relation extraction, topic classification, etc., transforming the original text into usable knowledge, and finally generating initial structured archival information containing content, metadata, association relationships, and traceability information. This initial structured archival information is stored in a database and can be directly called and searched by application systems. By combining NLP and multimodal features, the system automatically generates deep semantic tags that far exceed those of traditional metadata and establishes basic relationships between archives, transforming archives from isolated files into information nodes rich in semantic relationships and context, laying the foundation for in-depth utilization.

[0022] The initial structured archive information includes: unique identifier content, original archive file references, core metadata, enhanced content annotations, multimodal fusion feature references, processing traceability data, and related links.

[0023] The specific process of feature mining and analysis of digital archives in S3: S31. Deep Semantic Mining and Knowledge Extraction: Taking the initial structured information of the complete archive generated in S2 as input, the system first extracts precise entities, relationships and events through deep semantic analysis, constructs a deep relationship network, completes the unified link of the entire entity database, and runs a more in-depth NLP model than the S2 stage. The goal is to achieve a more refined and logical understanding based on the standardized extraction in S2. S32. Global Knowledge Graph Construction and Pattern Discovery: Based on the semantic information mentioned above, a global knowledge graph across archives is constructed, and graph mining technology is used to discover the hidden community structure, key nodes and abnormal patterns. Through clustering, association rule analysis, social network analysis, etc., the network of relationships between people, the chain of event evolution, and the trend of the rise and fall of themes are automatically discovered. The goal is to discover the hidden complex structure and patterns from the macro perspective of the entire archive. S33. Knowledge Fusion, Verification and Productization: The semantic mining results mentioned above are fused and verified with the previous multimodal physical features to generate high-level feature data with confidence assessment, a visualized knowledge graph and a programmable application interface. This completes the transformation from archival data to operable and decision-making domain knowledge, and outputs high-level feature data and knowledge results. The goal is to ensure the credibility of the knowledge and encapsulate it into a usable product. The specific process of feature mining and analysis is a qualitative leap from "individual information description" to "collective knowledge discovery". Based on the correlation information of the complete archives, through in-depth mining and analysis, the system can automatically discover hidden trends, correlation networks, thematic evolution and abnormal patterns, upgrading the archives from a "data repository" to a "data think tank" that can produce insightful reports. It transforms massive amounts of dormant raw data into high-value knowledge assets that are calculable, correlated and insightful, providing unprecedented support for decision-making.

[0024] The specific process of constructing a global knowledge graph across archives based on semantic information includes: integrating all entities, relationships and events extracted from S31 with a graph data structure; merging the same entity mentioned in different archives into a node in the graph; and constructing the relationships between entities to form a semantic network that comprehensively reflects people, events, things, organizations and their relationships within the archives, and outputting a queryable, reasonable global knowledge graph that spans across archives.

[0025] The specific content of advanced feature data and knowledge results includes: computable advanced feature datasets, interactive / visualized knowledge graphs and reports, structured analytical conclusions and insights, and knowledge service APIs that support system integration.

[0026] S2 involves labeling and indexing individual documents to achieve "findability"; S3 involves analyzing all documents to discover hidden patterns and achieve "comprehension". By utilizing the results of S2 (indexing and association) and S3 (knowledge and graph), we provide accurate, efficient and in-depth access to archival information and knowledge services. The specific processes of applications and services in S4 include: S41, Intelligent Search Interface and Intent Understanding: Users can initiate original queries through natural language questions, keyword combinations, or filtering conditions. The system performs entity recognition, relation extraction, and intent classification on the original queries, and links the identified entities with the full-domain entity database built by S3 to ensure semantic consistency. S42, Multi-engine Parallel Search and Fusion Ranking: Based on the parsed query intent, the system calls multi-level indexes in parallel, including full-text index, entity / tag index, and semantic vector index. Full-text index: quickly matches query terms; entity / tag index: accurately matches labeled entities, topics, and metadata; semantic vector index: this is one of the core capabilities of S3, converting the query statement into a semantic vector. In the document semantic vector space generated by S3, it searches for files with the most similar content semantics, even if they do not have completely identical keywords. It recalls candidate result sets from the multi-level indexes and uses a machine learning ranking model to intelligently rank the candidate result sets by considering multiple factors, including semantic relevance score, keyword matching degree, file authority and freshness, user historical search behavior, and relevance in the knowledge graph. It returns a set of files that are keyword-related, semantically related, and knowledge-related. S43. Enhance the context and present knowledge associations of the archive collection: encapsulate and expand the line information of the returned archive collection, and provide graph-based context navigation and knowledge discovery services, outputting an information-rich and visually associated search results interface. S44. Interactive Graph Query: Users can click on any entity (person, organization), tag (topic), or keyword on the results page. The system will immediately treat it as a new query intent, repeat the search steps, and output a "guided, associative, and exploratory" immersive file browsing and knowledge discovery experience. By integrating deep semantic analysis, knowledge graphs, and multi-level indexes, the search results are no longer simple lists, but rather "knowledge interfaces" containing rich connections and supporting visual graph navigation. Users can conduct "graph-based" exploration, starting from a point and discovering the entire knowledge network, greatly improving the efficiency and depth of knowledge discovery.

[0027] In summary, by constructing an optimized management framework that integrates intelligent perception, multi-engine parallel processing verification, multi-modal feature fusion, and knowledge graph mining, this approach effectively overcomes the inherent defects of traditional digital archive processing, such as fixed and rigid processes, low processing accuracy, missing associations, and inability to evolve. It significantly improves the accuracy and depth of archive information extraction, achieves intelligent and adaptive processing, automatically constructs semantic associations and knowledge networks between archives, and provides intelligent retrieval services based on semantic understanding and graph association. This enables a leap from "single keyword matching" to "intent understanding and related recommendations," thereby completing a leapfrog upgrade of digital archive management from "digital storage" to "intelligent knowledge services."

[0028] The above description is merely a preferred embodiment of the present invention; however, the scope of protection of the present invention is not limited thereto; any equivalent substitutions or modifications made by those skilled in the art within the technical scope disclosed in the present invention, based on the technical solution and its improved concept, should be covered within the scope of protection of the present invention.

Claims

1. An intelligent optimization management method applied to feature data mining and extraction in digital archives, characterized by: Includes the following steps: S1. Archive Collection and Preprocessing: Receive the input raw archives, perform dynamic perception preprocessing operations, and output basic structured text content with multimodal fusion features. S2. Archive organization and arrangement: Based on the basic structured text content, multimodal fusion features are extracted and combined with NLP technology to automatically generate richer and deeper tags and relationships, complete the intelligent organization and arrangement of individual archives, generate the initial structured archive information of the complete set of archives rich in semantic tags and basic relationships, and store it in the database. S3. Group Archive Feature Mining and Analysis: Based on the initial structured information of the complete archives obtained in S2, a deeper level of in-depth mining and analysis is performed with the goal of discovering group knowledge and complex patterns, and high-level feature data and knowledge results are output. S4. Archive Application and Services: Wait for user query operations, obtain the user's original query, and return the archive collection in advanced feature data and knowledge results by combining deep semantic analysis, knowledge graph and multi-level index. It also provides graph-based context navigation and knowledge discovery services, and outputs an information-rich and related visualized search results interface. Users can click on entities or tags in the graph to retrieve their subsequent queries. Based on the visual graph navigation, users can jump from one file to the entire related knowledge network.

2. The intelligent optimization management method for feature data mining and extraction of digital archives according to claim 1, characterized in that: The process of performing dynamic sensing preprocessing on the original archive in S1 includes: format conversion and image enhancement of the original archive to obtain the pre-processed digital image / file; For digital images / files, content information and quality indicators are obtained. Based on the content information and quality indicators, a dedicated OCR model is calibrated. Based on a dedicated OCR engine and an image recognition engine, digital images / files are processed in parallel and cross-modal. The output is a basic structured text content that combines text content, visual elements and the relationship between the two.

3. The intelligent optimization management method for feature data mining and extraction of digital archives according to claim 2, characterized in that: The process of forming the initial structured archive information in S2 specifically includes: Deep semantic analysis is performed on the basic structured text content and multimodal fusion features output in S1 to automatically extract entities, keywords, topics and attributes. Based on the semantic analysis results, classification tags and metadata are intelligently generated to build a multi-level index that supports full text, semantics and entities. By identifying common semantic elements between different archives, a basic relationship network between archives is automatically established to finally generate initial structured archive information.

4. The intelligent optimization management method for feature data mining and extraction of digital archives according to claim 3, characterized in that: The initial structured archive information includes: unique identifier content, original archive file references, core metadata, enhanced content annotations, multimodal fusion feature references, processing traceability data, and related links.

5. The intelligent optimization management method for feature data mining and extraction of digital archives according to claim 4, characterized in that: The specific process of feature mining and analysis of digital archives in S3: S31. Deep Semantic Mining and Knowledge Extraction: Taking the initial structured information generated in S2 as input, the system first extracts precise entities, relationships, and events through deep semantic analysis, constructs a deep relationship network, and completes a unified link to the entire entity database. S32. Global Knowledge Graph Construction and Pattern Discovery: Based on the semantic information above, a global knowledge graph spanning multiple archives is constructed, and graph mining techniques are used to discover the implicit community structure, key nodes, and abnormal patterns within it. S33. Knowledge Fusion, Verification and Productization: The semantic mining results are fused and verified with the previous multimodal physical features to generate high-level feature data with confidence assessment, a visualized knowledge graph and a programmable application interface, and output high-level feature data and knowledge results.

6. The intelligent optimization management method for feature data mining and extraction of digital archives according to claim 5, characterized in that: The specific process of constructing a global knowledge graph across archives based on semantic information includes: integrating all entities, relationships and events extracted from S31 with a graph data structure; merging the same entity mentioned in different archives into a node in the graph; and constructing the relationships between entities to form a semantic network that comprehensively reflects people, events, things, organizations and their relationships within the archives, and outputting a queryable, reasonable global knowledge graph that spans across archives.

7. The intelligent optimization management method for feature data mining and extraction of digital archives according to claim 6, characterized in that: The specific content of advanced feature data and knowledge results includes: computable advanced feature datasets, interactive / visualized knowledge graphs and reports, structured analytical conclusions and insights, and knowledge service APIs that support system integration.

8. The intelligent optimization management method for feature data mining and extraction of digital archives according to claim 7, characterized in that: The specific processes for applications and services in S4 include: S41, Intelligent Search Interface and Intent Understanding: Users can initiate original queries through natural language questions, keyword combinations, or filtering conditions. The system performs entity recognition, relation extraction, and intent classification on the original queries, and links the identified entities with the full-domain entity database built by S3 to ensure semantic consistency. S42. Multi-engine parallel search and fusion sorting: Based on the parsed query intent, the system calls multi-level indexes in parallel, retrieves candidate result sets from the multi-level indexes, uses a machine learning sorting model, and intelligently sorts the candidate result sets by combining multiple factors to return a set of archives that are related to keywords, semantics, and knowledge. S43. Enhance the context and present knowledge associations of the archive collection: encapsulate and expand the information of the returned archive collection, and provide graph-based context navigation and knowledge discovery services, outputting an information-rich and visually associated search results interface. S44. Interactive Graph Query: Users can click on any entity, tag, or keyword on the results page, and the system will immediately treat it as a new query intent, repeat the search steps, and output a "guided, associative, and exploratory" immersive file browsing and knowledge discovery experience.

Images