A method and system for automatically extracting critical care field knowledge

Abstract

The present invention relates to the field of medical information processing technology, and in particular to a method and system for automatically extracting knowledge in the field of critical illness. The method comprises the following steps: constructing a critical illness ontology in the field of critical illness to obtain a critical illness ontology data set; performing high-attention ontology mining on the critical illness ontology data set to obtain a high-attention ontology data set; performing topic consistency link analysis based on the high-attention ontology data set to obtain a potential critical illness knowledge source set; performing temporal knowledge fluctuation analysis on the potential critical illness knowledge source set to obtain a dynamic knowledge update time span data set; performing frontier knowledge source hierarchical screening on the potential critical illness knowledge source set based on the dynamic knowledge update time span data set to obtain a critical illness frontier dynamic knowledge source set. The present invention reduces the burden of manual search and literature reading, and improves the efficiency and comprehensiveness of knowledge acquisition.

Description

Technical Field

[0001] The present invention relates to the field of medical information processing technology, and in particular to a method and system for automatically extracting critical care field knowledge. Background Art

[0002] In the field of critical care, acquiring and organizing a large amount of cutting-edge clinical knowledge is of great significance for medical research and clinical practice. The field of critical care involves complex diseases and critical conditions, and it is necessary to keep up to date with the latest medical advances and treatments to provide the best medical care. However, due to the dispersion, dynamism and complexity of knowledge in the field of critical care, the current acquisition and organization methods have certain limitations.

[0003] Knowledge in the field of critical care is distributed in various sources such as literature, clinical guidelines, expert opinions, and academic conferences, and is diverse and complex. Existing acquisition methods mainly rely on manual search and reading of large amounts of literature, which is time-consuming and prone to missing important information. In addition, the rapid update speed of information, the continuous emergence of new research results and clinical guidelines, make the acquisition and organization of knowledge more difficult.

[0004] Existing knowledge extraction methods are mainly based on manual annotation and rule matching, which lacks automation and intelligence. Manual annotation requires a lot of professional knowledge and manpower input, which is costly and inefficient. The rule matching method is limited by pre-set rules and has certain limitations on the coverage and adaptability of knowledge. Summary of the invention

[0005] Based on this, it is necessary for the present invention to provide a method and system for automatically extracting knowledge in the field of critical care to solve at least one of the above technical problems.

[0006] To achieve the above purpose, a method for automatically extracting critical care domain knowledge includes the following steps:

[0007] Step S1: construct a critical care domain ontology for the critical care domain to obtain a critical care domain ontology data set; perform high-attention ontology mining on the critical care domain ontology data set to obtain a high-attention ontology data set;

[0008] Step S2: Perform topic consistency link analysis on the high-attention ontology dataset to obtain a potential critical disease knowledge source set; perform temporal knowledge fluctuation analysis on the potential critical disease knowledge source set to obtain a dynamic knowledge update time span dataset; perform frontier knowledge source hierarchical screening on the potential critical disease knowledge source set based on the dynamic knowledge update time span dataset to obtain a critical disease frontier dynamic knowledge source set;

[0009] Step S3: Based on the high-attention ontology dataset and the dynamic knowledge update time span dataset, a topic-driven crawler is performed on the critical care frontier dynamic knowledge source set to obtain a critical care frontier heterogeneous corpus dataset; heterogeneous knowledge representation learning in the critical care field is performed on the critical care frontier heterogeneous corpus dataset to obtain heterogeneous knowledge representation data in the critical care field;

[0010] Step S4: extract critical care clinical frontier knowledge from the critical care frontier heterogeneous corpus data set according to the critical care field heterogeneous knowledge representation data to obtain clinical frontier knowledge tuple data; perform intelligent triple completion on the clinical frontier knowledge tuple data based on the critical care field heterogeneous knowledge representation data to obtain clinical completion relationship triple data; perform knowledge graph embedding learning on the clinical completion relationship triple data and the critical care field heterogeneous knowledge representation data to obtain a primary critical care knowledge graph;

[0011] Step S5: Perform heuristic graph completion reasoning based on the primary critical illness knowledge graph to obtain critical illness graph defect repair data; perform edge connection repair and node expansion on the primary critical illness knowledge graph based on the critical illness graph defect repair data to obtain a cutting-edge knowledge graph in the critical illness field.

[0012] The present invention can organize the knowledge in the field of critical illness in a systematic and structured manner by constructing an ontology in the field of critical illness. This makes the knowledge in the field of critical illness easier to understand, query and utilize, and improves the accessibility and availability of knowledge. Through high-attention ontology mining, ontologies with importance and attention in the field of critical illness can be identified. This makes the subsequent knowledge extraction process more targeted and efficient, and can focus on the most valuable knowledge sources in the field of critical illness. Through thematic consistency link analysis, knowledge sources with similar themes can be associated to discover potential sets of critical illness knowledge sources. Through temporal knowledge fluctuation analysis, the update speed and time span of critical illness knowledge can be understood. This helps to identify knowledge sources with dynamic update properties, thereby ensuring that the extracted knowledge is timely and cutting-edge. Based on the hierarchical screening of cutting-edge knowledge sources based on the dynamic knowledge update time span data set, the latest and most cutting-edge knowledge sources in the field of critical illness can be determined. The topic-driven crawler captures heterogeneous corpus data related to cutting-edge critical illness knowledge in a targeted manner based on the high-attention ontology data set and the dynamic knowledge update time span data set. This provides a comprehensive and specific source of knowledge extraction, ensuring that the extracted knowledge is closely related to the field of critical illness. By learning from the frontier heterogeneous corpus data set of critical illness, knowledge can be represented in a form that can be understood and processed by computers, namely, heterogeneous knowledge representation data in the field of critical illness. Knowledge tuple data with clinical cutting-edge is extracted from the frontier heterogeneous corpus data of critical illness. This means that the system is able to analyze and understand the latest expertise in the field of critical illness and extract it from a large amount of heterogeneous corpus. These knowledge tuple data include key information such as new disease discoveries, treatments, drug research progress, clinical trials, etc. By extracting clinical cutting-edge knowledge, the system can provide accurate and timely knowledge support for medical researchers, clinicians and other relevant professionals. By completing the frontier clinical knowledge tuple data based on the heterogeneous knowledge representation data in the field of critical illness, the missing parts in the knowledge can be filled and the completeness and availability of knowledge can be improved. In the process of knowledge extraction, some information may be incomplete or missing, which will limit the accuracy and availability of knowledge. The goal of intelligent triple completion is to automatically supplement the missing information by analyzing and integrating heterogeneous knowledge representation data, making the knowledge more comprehensive and accurate. Based on the primary critical illness knowledge graph, the system performs heuristic graph completion reasoning to identify and repair defects in the critical illness graph. Through reasoning and analysis, the missing information in the critical illness graph can be filled to improve the completeness and accuracy of the graph. According to the critical illness graph defect repair data, the primary critical illness knowledge graph is repaired for edge connections and expanded for nodes. This can repair edge connection errors and missing nodes in the original graph, making the graph more complete and coherent. In summary, the present invention provides systematic and structured critical illness field knowledge, which improves the accessibility, usability and accuracy of knowledge. This provides accurate and timely knowledge support for medical researchers, clinicians and other relevant professionals.

[0013] Preferably, the present invention also provides a system for automatically extracting critical disease domain knowledge, which is used to execute the method for automatically extracting critical disease domain knowledge as described above. The system for automatically extracting critical disease domain knowledge includes:

[0014] The critical illness ontology construction module is used to construct the critical illness ontology in the critical illness field to obtain the critical illness ontology data set; perform high-attention ontology mining on the critical illness ontology data set to obtain the high-attention ontology data set;

[0015] The knowledge source analysis module is used to perform topic consistency link analysis based on the high-attention ontology data set to obtain a potential critical disease knowledge source set; perform temporal knowledge fluctuation analysis on the potential critical disease knowledge source set to obtain a dynamic knowledge update time span data set; perform frontier knowledge source hierarchical screening on the potential critical disease knowledge source set based on the dynamic knowledge update time span data set to obtain a critical disease frontier dynamic knowledge source set;

[0016] The heterogeneous corpus collection module is used to perform topic-driven crawling on the critical care frontier dynamic knowledge source set based on the high-attention ontology dataset and the dynamic knowledge update time span dataset to obtain the critical care frontier heterogeneous corpus dataset; perform heterogeneous knowledge representation learning in the critical care field on the critical care frontier heterogeneous corpus dataset to obtain heterogeneous knowledge representation data in the critical care field;

[0017] The frontier knowledge extraction module is used to extract the clinical frontier knowledge of critical illness from the frontier heterogeneous corpus data set of critical illness according to the heterogeneous knowledge representation data in the critical illness field, and obtain the clinical frontier knowledge tuple data; based on the heterogeneous knowledge representation data in the critical illness field, the clinical frontier knowledge tuple data is intelligently completed to obtain the clinical completion relationship triple data; the clinical completion relationship triple data and the heterogeneous knowledge representation data in the critical illness field are embedded in the knowledge graph to obtain the primary critical illness knowledge graph;

[0018] The knowledge graph repair module is used to perform heuristic graph completion reasoning based on the primary critical illness knowledge graph to obtain critical illness graph defect repair data; according to the critical illness graph defect repair data, the primary critical illness knowledge graph is repaired for edge connections and expanded for nodes to obtain a cutting-edge knowledge graph in the critical illness field.

[0019] In the present invention, the critical care ontology construction module can provide a structured critical care knowledge base by constructing an ontology in the critical care field. This ontology data set contains concepts, terms and relationships that describe critical care knowledge. It provides a unified basis for subsequent knowledge extraction and analysis. The knowledge source analysis module can discover knowledge sources related to the critical care field through subject consistency link analysis. These knowledge sources can include literature, clinical guidelines, expert opinions and academic conferences. By obtaining a potential critical care knowledge source set, researchers and doctors can more easily obtain important domain knowledge to support their decision-making and practice. The heterogeneous corpus acquisition module can collect heterogeneous corpus data from the critical care frontier dynamic knowledge source set by adopting topic-driven crawler technology. These data come from different sources and formats, such as text documents, web pages, papers, etc. The goal of the heterogeneous corpus acquisition module is to obtain a data set containing critical care frontier knowledge. This provides researchers with a wide range of data sources to more comprehensively understand and study the latest progress in the critical care field. Frontier knowledge extraction module This module extracts clinical frontier knowledge tuple data by analyzing and processing heterogeneous corpus data. These knowledge tuple data include new treatment methods for diseases, the latest research results, clinical practice guidelines, etc. The intelligent triple completion function of the frontier knowledge extraction module can fill in the missing or incomplete information in the knowledge tuple data, and provide more complete and accurate clinical relationship triple data. This enables researchers to screen out important frontier knowledge from a large amount of heterogeneous corpus and have a more comprehensive knowledge perspective. The knowledge graph repair module uses heuristic graph completion reasoning based on the primary critical illness knowledge graph, which can find defects and incompleteness in the critical illness graph. By reasoning and analyzing the primary critical illness knowledge graph, the repaired data includes edge connections and node expansion information that need to be repaired. In this way, the repaired critical illness knowledge graph is more complete and comprehensive, providing more accurate and comprehensive critical illness field knowledge. Researchers and doctors can obtain deeper critical illness field knowledge from the repaired knowledge graph to support their research and practice. In summary, the present invention reduces the burden of manual search and reading of literature, improves the efficiency and comprehensiveness of knowledge acquisition, and helps users understand the latest medical advances and treatment methods. At the same time, the automated knowledge extraction and repair functions provide users with more accurate and comprehensive knowledge in critical care areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments thereof made with reference to the following drawings:

[0021] Figure 1 A schematic flow chart of the steps of an implementation method for automatic extraction of critical care domain knowledge according to an embodiment is shown.

[0022] Figure 2 A detailed flow chart of step S25 of an embodiment is shown. DETAILED DESCRIPTION

[0023] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0024] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0025] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0026] To achieve this, please refer to Figure 1 to Figure 2 The present invention provides a method for automatically extracting critical care domain knowledge, comprising the following steps:

[0027] Step S1: construct a critical care domain ontology for the critical care domain to obtain a critical care domain ontology data set; perform high-attention ontology mining on the critical care domain ontology data set to obtain a high-attention ontology data set;

[0028] Specifically, for example, domain knowledge and concepts related to the critical care field can be collected, including terms, entities, relationships, etc. Use ontology modeling tools, such as Protégé, to create an ontology model and define the concepts, attributes, and relationships in the field. According to the characteristics and needs of the critical care field, design the hierarchy, categories, and attributes of the ontology, and define the associations between concepts. Use natural language processing technology and machine learning algorithms to identify ontology concepts and relationships that are of high interest and importance in the critical care field. These high-interest ontology data can be specific topics, research hotspots, or key concepts.

[0029] Step S2: Perform topic consistency link analysis on the high-attention ontology dataset to obtain a potential critical disease knowledge source set; perform temporal knowledge fluctuation analysis on the potential critical disease knowledge source set to obtain a dynamic knowledge update time span dataset; perform frontier knowledge source hierarchical screening on the potential critical disease knowledge source set based on the dynamic knowledge update time span dataset to obtain a critical disease frontier dynamic knowledge source set;

[0030] Specifically, for example, the ontology concepts and relationships in the high-attention ontology data set can be used in combination with text mining technology to analyze and process a large number of critical care literature, papers, patents, etc. Through semantic similarity calculation or association rule mining, the concepts and knowledge in the relevant literature are linked to the high-attention ontology data set to form a potential critical care knowledge source set. Perform time series analysis on the knowledge and information in the potential critical care knowledge source set to understand its changing trends and update frequency. Collect historical documents, patents, conference reports, etc. in the critical care field, and use time series analysis methods such as time span and frequency analysis to explore the update time span of knowledge and information. Evaluation and sorting can be performed based on indicators such as the frequency of knowledge updates, the latest research results, and the evaluation of field experts. Use machine learning algorithms or expert systems to classify, cluster, or evaluate the importance of potential critical care knowledge source sets to screen out the critical care frontier dynamic knowledge source set.

[0031] Step S3: Based on the high-attention ontology dataset and the dynamic knowledge update time span dataset, a topic-driven crawler is performed on the critical care frontier dynamic knowledge source set to obtain a critical care frontier heterogeneous corpus dataset; heterogeneous knowledge representation learning in the critical care field is performed on the critical care frontier heterogeneous corpus dataset to obtain heterogeneous knowledge representation data in the critical care field;

[0032] Specifically, for example, high-attention ontology datasets and dynamic knowledge update time span datasets can be used to identify key topics and concerns in the critical care frontier dynamic knowledge source set. Use web crawler tools (such as Scrapy) or specialized crawler frameworks to crawl relevant academic journals, conference papers, patent databases and other network resources based on the identified topics and concerns. Word embedding models (such as Word2Vec, GloVe) or deep learning models (such as BERT, ELMo) can be used to learn vector representations of words and sentences in the critical care field in the critical care frontier heterogeneous corpus dataset. For heterogeneous data (such as structured data, text data, image data, etc.), multimodal learning methods (such as joint representation learning of images and text) can be used to process them.

[0033] Step S4: extract critical care clinical frontier knowledge from the critical care frontier heterogeneous corpus data set according to the critical care field heterogeneous knowledge representation data to obtain clinical frontier knowledge tuple data; perform intelligent triple completion on the clinical frontier knowledge tuple data based on the critical care field heterogeneous knowledge representation data to obtain clinical completion relationship triple data; perform knowledge graph embedding learning on the clinical completion relationship triple data and the critical care field heterogeneous knowledge representation data to obtain a primary critical care knowledge graph;

[0034] Specifically, for example, heterogeneous knowledge representation data in the field of critical illness and heterogeneous corpus data sets of critical illness frontiers can be used to extract entities, relations and events related to clinical frontier knowledge by applying natural language processing and information extraction technologies. Named entity recognition (NER) algorithms, relation extraction algorithms (such as rule-based, pattern-based or machine learning-based methods) and other technologies can be used to identify and extract critical illness clinical frontier knowledge tuple data. Using heterogeneous knowledge representation data in the field of critical illness and clinical frontier knowledge tuple data, intelligent algorithms are used to complete and reason triples. Knowledge graph completion algorithms (such as rule-based, graph neural network-based or reasoning rule-based methods) can be used in combination with heterogeneous knowledge representation data in the field of critical illness to automatically reason and complete missing relations and attributes in clinical frontier knowledge tuples. Using clinical completion relation triple data and heterogeneous knowledge representation data in the field of critical illness, knowledge graph embedding learning is performed. Graph embedding algorithms (such as TransE, TransR, GAT, etc.) can be used to map entities and relations in knowledge graphs into low-dimensional vector spaces to capture semantic and structural information between them.

[0035] Step S5: Perform heuristic graph completion reasoning based on the primary critical illness knowledge graph to obtain critical illness graph defect repair data; perform edge connection repair and node expansion on the primary critical illness knowledge graph based on the critical illness graph defect repair data to obtain a cutting-edge knowledge graph in the critical illness field.

[0036] Specifically, for example, the primary critical illness knowledge graph and the graph completion reasoning algorithm can be used to perform a heuristic reasoning process to repair defects in the critical illness graph. Graph completion reasoning can be performed using rule-based, graph neural network-based, or reasoning rule-based methods. During the reasoning process, based on the existing entities, relationships, and attributes in the graph, possible missing relationships and attributes are inferred and completed accordingly. The reasoning results include the predicted missing relationships and attributes, as well as their confidence or probability. According to the critical illness graph defect repair data, the primary critical illness knowledge graph is repaired with edge connections and expanded with nodes. According to the predicted relationships in the defect repair data, the missing edge connections are added to the primary critical illness knowledge graph. According to the predicted attributes in the defect repair data, the entity nodes in the knowledge graph are expanded to add the missing attribute information. Edge connection repair and node expansion can be performed using a graph database or a graph processing framework (such as Neo4j, NetworkX) to repair and expand the primary critical illness knowledge graph by adding new edges and nodes.

[0037] The present invention can organize the knowledge in the field of critical illness in a systematic and structured manner by constructing an ontology in the field of critical illness. This makes the knowledge in the field of critical illness easier to understand, query and utilize, and improves the accessibility and availability of knowledge. Through high-attention ontology mining, ontologies with importance and attention in the field of critical illness can be identified. This makes the subsequent knowledge extraction process more targeted and efficient, and can focus on the most valuable knowledge sources in the field of critical illness. Through thematic consistency link analysis, knowledge sources with similar themes can be associated to discover potential sets of critical illness knowledge sources. Through temporal knowledge fluctuation analysis, the update speed and time span of critical illness knowledge can be understood. This helps to identify knowledge sources with dynamic update properties, thereby ensuring that the extracted knowledge is timely and cutting-edge. Based on the hierarchical screening of cutting-edge knowledge sources based on the dynamic knowledge update time span data set, the latest and most cutting-edge knowledge sources in the field of critical illness can be determined. The topic-driven crawler captures heterogeneous corpus data related to cutting-edge critical illness knowledge in a targeted manner based on the high-attention ontology data set and the dynamic knowledge update time span data set. This provides a comprehensive and specific source of knowledge extraction, ensuring that the extracted knowledge is closely related to the field of critical illness. By learning from the frontier heterogeneous corpus data set of critical illness, knowledge can be represented in a form that can be understood and processed by computers, namely, heterogeneous knowledge representation data in the field of critical illness. Knowledge tuple data with clinical cutting-edge is extracted from the frontier heterogeneous corpus data of critical illness. This means that the system is able to analyze and understand the latest expertise in the field of critical illness and extract it from a large amount of heterogeneous corpus. These knowledge tuple data include key information such as new disease discoveries, treatments, drug research progress, clinical trials, etc. By extracting clinical cutting-edge knowledge, the system can provide accurate and timely knowledge support for medical researchers, clinicians and other relevant professionals. By completing the frontier clinical knowledge tuple data based on the heterogeneous knowledge representation data in the field of critical illness, the missing parts in the knowledge can be filled and the completeness and availability of knowledge can be improved. In the process of knowledge extraction, some information may be incomplete or missing, which will limit the accuracy and availability of knowledge. The goal of intelligent triple completion is to automatically supplement the missing information by analyzing and integrating heterogeneous knowledge representation data, making the knowledge more comprehensive and accurate. Based on the primary critical illness knowledge graph, the system performs heuristic graph completion reasoning to identify and repair defects in the critical illness graph. Through reasoning and analysis, the missing information in the critical illness graph can be filled to improve the completeness and accuracy of the graph. According to the critical illness graph defect repair data, the primary critical illness knowledge graph is repaired for edge connections and expanded for nodes. This can repair edge connection errors and missing nodes in the original graph, making the graph more complete and coherent. In summary, the present invention provides systematic and structured critical illness field knowledge, which improves the accessibility, usability and accuracy of knowledge. This provides accurate and timely knowledge support for medical researchers, clinicians and other relevant professionals.

[0038] Preferably, step S1 comprises the following steps:

[0039] Step S11: Acquire a critical care medical big data set;

[0040] Specifically, for example, publicly available medical literature databases can be used as data sources, taking PubMed as an example. First, visit the PubMed website and search using keywords and screening conditions related to the field of critical care, such as "critical care", "intensive care" or "ICU". Then, select appropriate documents based on the search results to download or obtain abstract information. These documents can include research papers, clinical guidelines, and expert opinions. This process can be repeated until a sufficient amount and diversity of medical literature data is obtained to construct a large data set for critical care.

[0041] Step S12: performing named entity recognition on the critical care medical big data set to obtain all named entity sets in the critical care field;

[0042] Specifically, for example, natural language processing (NLP) tools and techniques can be used. In this embodiment, spaCy will be used as a commonly used NLP tool for illustration. First, the critical care large data set is preprocessed, including steps such as text cleaning, sentence segmentation, and word segmentation. Then, the data set is loaded using the trained spaCy model. Next, the processed text is traversed to extract named entities and their corresponding entity types, such as disease names, drug names, treatment methods, etc. Finally, the extracted named entity set is saved in an appropriate data structure, such as a list, a database, or an ontology map.

[0043] Step S13: Perform domain semantic filtering on all named entity sets in the critical illness field to obtain a core named entity set in the critical illness field;

[0044] Specifically, for example, professional terms and domain-specific entity nouns related to the field of critical illness can be collected, such as disease names, drug names, treatment methods, etc. A dictionary can be constructed based on medical literature, professional vocabulary, and the knowledge of domain experts. Preprocess all named entity sets in the field of critical illness, including common text processing steps such as text cleaning, sentence segmentation, and word segmentation. Use the constructed domain professional terminology dictionary to filter the named entity set. By matching the entity nouns with the terms in the dictionary, the core named entities related to the field of critical illness are screened out. Save the core named entity set in the field of critical illness obtained through domain semantic filtering into an appropriate data structure, such as a list, database, or ontology graph.

[0045] Step S14: constructing a critical illness domain ontology for the critical illness domain according to the critical illness domain core named entity set to obtain a critical illness domain ontology dataset;

[0046] Specifically, for example, the objectives and application scope of the critical care domain ontology can be clarified. Determine the concepts, attributes, and relationships of the ontology, as well as the associations with other domain ontologies. Design the concept categories (classes), attributes, and relationships of the ontology based on the core named entity set of the critical care domain. You can use ontology editing tools, such as Protégé, to create classes and attributes of the ontology and define the relationships between them. According to the core named entity set, add entities to the ontology and establish relationships between them. For example, you can add entities such as critical illness, drugs, and treatment methods, and define the association relationships between them, such as the relationship between treatment methods and diseases, and the relationship between drugs and diseases. Verify and revise the constructed critical care domain ontology to ensure the consistency and accuracy of the ontology. You can verify it with the help of the knowledge of domain experts and domain specifications, and revise and improve it as needed. Save the constructed critical care domain ontology in the form of an ontology dataset, and use common ontology representation formats such as RDF (Resource Description Framework) or OWL (Web Ontology Language).

[0047] Step S15: Perform high-attention ontology mining on the critical care domain ontology dataset to obtain a high-attention ontology dataset.

[0048] Specifically, for example, the critical care domain ontology dataset can be converted into a graph data structure. Graph database tools, such as Neo4j, can be used to import the ontology dataset as nodes and relationships in the graph database. Based on the concepts and relationships in the ontology dataset, an association network is constructed. Concepts are used as nodes, relationships are used as edges, and the connection relationship between concepts and relationships is represented as a graph structure. The attention of the nodes is calculated using a graph analysis algorithm. Commonly used algorithms include PageRank, BetweennessCentrality, etc. These algorithms can help identify nodes that are important and highly concerned in the network. According to the attention value of the node, the ontology concepts and relationships with high attention are screened out. A threshold can be set to determine the selection of high-attention nodes. The mined high-attention ontology dataset is output as an appropriate data format, such as a graph database or a text file.

[0049] The present invention can obtain data from multiple medical institutions, research institutions and academic journals by acquiring a critical care medical big data set. This makes the knowledge base more comprehensive and diversified, covering critical care medical information of various types and sources. Critical care medical big data sets usually contain a large number of data records, covering a wide range of clinical cases, case data, treatment plans, experimental data, etc. This provides a sufficient data basis for subsequent knowledge extraction and mining, and increases the accuracy and reliability of knowledge. By acquiring real-time critical care medical big data sets, the current critical care medical conditions, research progress and new clinical practices can be quickly reflected. This helps to update and supplement the knowledge in the critical care field in a timely manner, making the knowledge timely and cutting-edge. Through named entity recognition technology, named entities related to the critical care field, such as diseases, symptoms, drugs, etc., can be accurately identified and extracted from the critical care medical big data set. This helps to reduce errors and subjectivity and improve the accuracy and reliability of knowledge extraction. The identified and extracted named entities can be used as the basic units of critical care field knowledge to structure and organize the information in the critical care medical big data set. This makes critical care field knowledge easier to manage, query and analyze, laying the foundation for subsequent knowledge processing and application. Through domain semantic filtering, named entities closely related to the critical care field can be screened out according to the specific semantics and context of the critical care field. This helps to reduce noise and redundant information and improve the accuracy and relevance of knowledge. The core named entity set in the critical care field obtained after screening mainly contains information related to key concepts and entities in the critical care field, which can better focus on the core content of the critical care field. This helps professionals to obtain knowledge closely related to their research and practice more quickly and improve work efficiency. By constructing a critical care field ontology, the knowledge in the critical care field can be organized in a systematic and structured manner. The ontology provides a hierarchical structure and relationship network to define and represent the core named entities in the critical care field and the relationships between them. This helps to better understand and express the knowledge in the critical care field, making the knowledge easier to understand, query and apply. The construction of the critical care field ontology can enhance the relevance and coherence of knowledge by defining the relationships and attributes between entities. This enables the relevant critical care field knowledge to be more closely connected to form a more complete and comprehensive knowledge system. Through high-attention ontology mining, high-attention entities and relationships in the critical care field ontology can be identified. This helps professionals to acquire knowledge related to current research and practice more quickly, and focus on hot issues and cutting-edge advances in the field of critical care. High-attention ontology mining can identify knowledge gaps and incomplete parts in the ontology of the field of critical care, and provide guidance for subsequent knowledge updates and supplements. By filling these gaps, the completeness and accuracy of knowledge can be improved, providing the latest and most comprehensive knowledge support for research and practice in the field of critical care.

[0050] Preferably, step S14 comprises the following steps:

[0051] Step S141: Perform entity concept diffusion analysis on the core named entity set in the critical illness field to obtain critical illness entity concept diffusion trend data;

[0052] Specifically, for example, text mining technology can be used to extract the relationship between core named entities based on the core named entity set in the critical illness field. Natural language processing tools and techniques, such as named entity recognition and relationship extraction, can be used to extract the association between entities. Based on entity relationships, concept diffusion analysis can be performed. Network analysis algorithms, such as propagation models and community discovery, can be used to analyze the degree of propagation and association between entities, thereby obtaining critical illness entity concept diffusion trend data.

[0053] Step S142: constructing an initial entity concept network for the core named entity set in the critical illness field to obtain an initial entity concept network model;

[0054] Specifically, for example, the relationships between core named entities can be extracted based on the core named entity set in the critical care field through text mining or data processing technology. These relationships can be co-occurrence relationships, similarity relationships, association relationships, etc. Relationship extraction can be based on indicators such as co-occurrence frequency and semantic similarity between entities. Based on the entity relationships, an initial concept network model is constructed. Entities are used as nodes and relationships are used as edges, and the connection relationships between entities are represented as a graph structure.

[0055] Step S143: mining constraint knowledge rules on the initial entity concept network model based on the critical illness entity concept diffusion trend data to obtain critical illness domain knowledge constraint rule data;

[0056] Specifically, for example, association rule mining algorithms, such as the Apriori algorithm or the FP-Growth algorithm, can be applied to mine the concept diffusion trend data of critical illness entities. Association rules can help discover the correlation and constraint relationships between entities. Set appropriate support and confidence thresholds to filter out rules with certain correlations. Extract knowledge constraint rules in the critical illness field from the mined association rules. Select constraint rules with practical significance and explanatory power based on business needs and domain expert knowledge. Rules can be sorted and filtered based on indicators such as confidence and support. The data of knowledge constraint rules in the critical illness field obtained by mining can be output and saved as a text file or database. Perform data analysis on the mining results, such as counting the number of rules, frequent item set analysis, etc., to gain insights and understanding about knowledge constraints in the critical illness field.

[0057] Step S144: optimizing the constraint rules of the initial entity concept network model according to the critical disease field knowledge constraint rule data to obtain an optimized entity concept network model;

[0058] Specifically, for example, the critical care domain knowledge constraint rule data can be applied to the initial entity concept network model to perform constraint rule matching. According to the constraints defined in the rules, the entity relationships and attributes in the network model that meet the rules are matched. Rule matching algorithms can be used, such as methods based on pattern matching, logical reasoning, or graph algorithms. According to the matched constraint rules, the constraint rules are applied to the initial entity concept network model. This can include adding, modifying, or deleting entity relationships, attributes, or constraints in the network. According to the operation type and description of the rule, the network model is modified and adjusted accordingly. After applying the constraint rules, the optimized entity concept network model is further optimized and adjusted. Network optimization algorithms, such as graph theory-based algorithms, machine learning, or optimization methods, can be applied to improve the structure, performance, or interpretability of the network.

[0059] Step S145: Acquire a real-time knowledge base in the critical care field; interactively iteratively optimize the optimized entity concept network model and the real-time knowledge base in the critical care field to obtain an ontology data set in the critical care field.

[0060] Specifically, for example, the data source for building a real-time knowledge base in the field of critical illness can be determined. It can include public data sets, medical literature, expert knowledge bases, hospital records, etc. Ensure that the data source is authoritative and reliable. Use data collection tools or methods to collect data related to the field of critical illness from the selected data source. This can include text, structured data, images, real-time sensor data, etc. Ensure the comprehensiveness and timeliness of data collection. Use the optimized entity concept network model as the initial model for interactive optimization with the real-time knowledge base in the field of critical illness. Compare and analyze the real-time knowledge base in the field of critical illness with the initial model. Identify the differences between new entities, relationships, or attributes in the knowledge base and the model. Update and adjust the optimized entity concept network model based on new findings in the knowledge base. This can include adding new entities, establishing new relationships, modifying attributes or constraints, etc. Verify and evaluate the updated model to ensure consistency and accuracy between the model and the knowledge base. Evaluation can be performed using feedback from domain experts, logical reasoning, or model evaluation indicators. Based on the results of interactive iterative optimization, the model is repeatedly updated and verified until a certain accuracy and stability are achieved.

[0061] The present invention can understand the concept evolution process and trend of the core named entities in the critical field through entity concept diffusion analysis. This helps to track and understand the development dynamics of the critical field, grasp the changes and evolution of knowledge in the critical field, and provide a reference for subsequent knowledge updates and supplements. By constructing an initial entity concept network model, the relationship between the core named entities in the critical field can be modeled and represented. This helps to structure and organize the knowledge in the critical field to form a relationship network, providing a more comprehensive and accurate knowledge expression and query method. Through constraint knowledge rule mining, potential knowledge constraint rules can be found from the critical entity concept diffusion trend data. This helps to reveal the associations and laws between entities in the critical field, mine valuable knowledge hidden in the data, and provide support for subsequent knowledge applications and decision-making. Through constraint rule optimization, the knowledge constraint rules in the critical field can be applied to the initial entity concept network model to enhance the relevance and coherence between entities. This helps to form a more accurate and complete knowledge system and improve the availability and application effect of knowledge. By obtaining a real-time knowledge base in the critical field, the latest critical field knowledge can be obtained in a timely manner. By interactively iterating and optimizing the entity concept network model, real-time knowledge can be combined with the model to timely update and supplement the critical care domain ontology dataset, maintaining the timeliness and cutting-edge nature of knowledge. The interactive iterative optimization process can further refine and optimize the critical care domain ontology dataset through interaction with the real-time knowledge base. This helps to improve the quality and accuracy of the critical care domain ontology dataset, making it more adaptable to actual application needs and providing more reliable knowledge support for research, decision-making and practice in the critical care field.

[0062] Preferably, step S2 comprises the following steps:

[0063] Step S21: performing entity semantic granulation on the high-interest ontology data set to obtain a critical disease domain entity semantic granularity set;

[0064] Specifically, for example, you can use machine learning or rule-based entity recognition tools, such as Stanford NER, SpaCy, etc., to perform entity recognition on text in high-profile ontology datasets and extract entities. Entities can be disease names, drug names, treatment methods, etc. Use natural language processing tools, such as WordNet, BERT, etc., to perform semantic analysis on the extracted entities and classify or granulate them. This can be achieved through natural language processing technologies such as part-of-speech tagging, named entity recognition, and word sense disambiguation. Divide entities into different semantic granularities based on domain requirements and semantic relationships between entities. For example, divide diseases into major categories, minor categories, specific diseases, and other levels.

[0065] Step S22: performing heterogeneous critical illness domain knowledge source discovery on the critical illness domain entity semantic granularity set to obtain a candidate heterogeneous critical illness knowledge source set;

[0066] Specifically, for example, multiple heterogeneous critical care domain knowledge sources can be selected as embodiments, including knowledge graphs, medical literature databases, professional medical websites and forums, etc. Data extraction tools (such as web crawlers, API interfaces or database queries) are used to extract data related to the semantic granularity set of entities in the critical care domain from these heterogeneous knowledge sources. These data can be text, structured data or graph data. Subsequently, ETL tools or data conversion scripts are used to convert and integrate the data extracted from different heterogeneous knowledge sources so that they can be processed and analyzed under a unified data model. Next, data mining and analysis techniques (such as association rule mining, text mining or entity relationship extraction) are used to mine the integrated data to discover knowledge related to the semantic granularity set of entities in the critical care domain. Finally, through cooperation and feedback with domain experts, the mined knowledge and heterogeneous knowledge sources are evaluated and screened, and knowledge sources with high quality and credibility are selected as candidate heterogeneous critical care knowledge source sets.

[0067] Step S23: performing topic consistency link analysis on the candidate heterogeneous critical illness knowledge source set based on the critical illness domain entity semantic granularity set to obtain a potential critical illness knowledge source set;

[0068] Specifically, for example, each knowledge source in the candidate heterogeneous critical disease knowledge source set can be traversed. Entities related to the critical disease field can be extracted from the current knowledge source. Natural language processing techniques, such as entity recognition, keyword extraction, and other methods can be used to extract entities. The extracted entities are matched with the semantic granularity set of entities in the critical disease field. This can be done by comparing the entities with the entities in the semantic granularity set to find the matching semantic granularity. Based on the matched semantic granularity, the topic consistency link between the current knowledge source and the candidate heterogeneous critical disease knowledge source set is analyzed. The topic consistency can be evaluated using similarity metrics, semantic association, and other methods. Based on the analysis results of the consistency link, it is determined whether the current knowledge source has topic consistency with the candidate heterogeneous critical disease knowledge source set. If there is topic consistency, the current knowledge source is added to the potential critical disease knowledge source set. All knowledge sources with topic consistency are collected to form a potential critical disease knowledge source set.

[0069] Step S24: performing time series knowledge fluctuation analysis on the potential critical illness knowledge source set to obtain a dynamic knowledge update time span data set;

[0070] Specifically, for example, knowledge update data in the potential critical disease knowledge source set can be collected, which can include the release time, update time or timestamp of related events of the knowledge source. Use time series analysis methods to analyze the knowledge update time span in the potential critical disease knowledge source set. Common time series analysis methods include autoregressive moving average model (ARMA), autoregressive integrated moving average model (ARIMA) and seasonal decomposition. Preprocess the knowledge update time data in the potential critical disease knowledge source set, including removing outliers, filling missing values ​​and smoothing. Then, based on the preprocessed time series data, methods such as ARMA, ARIMA or seasonal decomposition can be applied to perform time series analysis.

[0071] Step S25: Perform frontier knowledge source hierarchical screening on the potential critical disease knowledge source set according to the dynamic knowledge update time span data set to obtain the critical disease frontier dynamic knowledge source set.

[0072] Specifically, for example, for each potential critical disease knowledge source, its corresponding dynamic knowledge update time span data can be obtained. According to the dynamic knowledge update time span data, the potential critical disease knowledge sources are hierarchically screened. A shorter time span usually indicates a more cutting-edge knowledge source. According to the preset time span threshold, the knowledge sources with a shorter time span are screened as cutting-edge knowledge sources. Repeat the above steps to perform time span screening and hierarchical screening operations on all potential critical disease knowledge sources. The screened cutting-edge knowledge sources are collected to form a set of critical disease cutting-edge dynamic knowledge sources.

[0073] The present invention groups entities in high-attention ontology data sets according to their semantic features through entity semantic granulation. This helps to classify and organize entities in the critical care field according to different semantic granularities, and provides a more fine-grained entity description and query method. Through semantic granularization, the meaning and relationship of entities can be better understood, providing a basis for subsequent knowledge discovery and linking. By performing heterogeneous critical care field knowledge source discovery on the entity semantic granularity set, diversified knowledge sources related to the critical care field can be found. This helps to discover knowledge resources of different forms and sources, including literature, experts, databases, etc., to provide diversity and breadth for subsequent knowledge integration and analysis. Through thematic consistency link analysis, the entity semantic granularity set in the critical care field can be associated and linked with the candidate heterogeneous critical care knowledge source set. This helps to discover potential knowledge sources related to the critical care field, establish an association relationship between entities and knowledge sources, and provide a basis for subsequent knowledge acquisition and integration. Through temporal knowledge fluctuation analysis, the changes and update frequency of potential critical care knowledge sources over time can be understood. This helps to identify and analyze the timeliness and dynamics of critical care field knowledge and determine the time span of knowledge update. By understanding the temporal characteristics of critical care knowledge, we can better manage knowledge and formulate update strategies. Based on the dynamic knowledge update time span data set, we can screen the potential critical care knowledge source set for cutting-edge knowledge sources. This helps to identify and select those knowledge sources with shorter time spans and faster updates as the cutting-edge dynamic knowledge sources to focus on. By screening out cutting-edge knowledge sources, we can timely obtain and apply the latest and forward-looking knowledge in the critical care field, and improve the level of research and practice in the critical care field.

[0074] Preferably, step S23 includes the following steps:

[0075] Step S231: Perform entity word unit frequency statistical analysis on the entity semantic granularity set in the critical disease field to obtain critical disease entity word unit frequency statistical data;

[0076] Specifically, for example, text analysis tools or programming languages ​​(such as Python) can be used to perform entity word frequency statistics on the semantic granularity set of entities in the critical illness field. For each entity, count its frequency of occurrence in the corpus. Methods such as Bag-of-Words or TF-IDF (Term Frequency-Inverse Document Frequency) can be used to calculate the word frequency of the entity. According to the word frequency statistics of the entity, organize the data and generate critical illness entity word frequency statistics. This can be a table or data file containing the entity names and their corresponding frequencies.

[0077] Step S232: extracting entity semantic attributes from the entity semantic granularity set in the critical disease field to obtain critical disease entity semantic attribute data;

[0078] Specifically, for example, natural language processing (NLP) tools or programming languages ​​(such as Python) can be used to extract entity semantic attributes from the semantic granularity set of entities in the critical care field. For each entity, its semantic attributes are extracted, such as entity type, associated concepts, attribute features, etc. According to the semantic attribute extraction results of the entity, the data is sorted and critical care entity semantic attribute data is generated. This can be a table or data file containing entity names and their corresponding semantic attributes.

[0079] Step S233: Calculate entity semantic similarity of the entity semantic granularity set in the critical disease field according to the critical disease entity semantic attribute data to obtain critical disease entity semantic similarity matrix data;

[0080] Specifically, for example, a suitable model or method can be selected to construct a semantic representation of the entity corresponding to the semantic attribute data of the critical illness entity. This can include word vector models (such as Word2Vec, GloVe), pre-trained language models (such as BERT, GPT), or other methods based on deep learning or traditional machine learning. The semantic similarity between entities is calculated using the constructed semantic representation model. The semantic similarity between entities can be measured using methods such as cosine similarity, Euclidean distance, and similarity calculation based on attention mechanisms. Based on the calculated semantic similarity, construct the critical illness entity semantic similarity matrix data. This can be a symmetric matrix in which each element represents the semantic similarity between two entities.

[0081] Step S234: Based on the statistical data of critical illness entity word frequency and the critical illness entity semantic similarity matrix data, latent semantic topic link mining is performed on the candidate heterogeneous critical illness knowledge source set to obtain a potential topic link relationship data set;

[0082] Specifically, for example, the statistical data of word frequency of critical entity and the semantic similarity matrix data of critical entity are loaded into the Python environment. The input data is preprocessed, including operations such as removing stop words, stemming or part-of-speech tagging. The NLTK library or SpaCy library can be used to assist in text preprocessing. The preprocessed data is converted into the corpus format required in the Gensim library. This can be a list containing a list of words for each document, or a specific data structure in Gensim, such as Corpus or Dictionary. The LDA topic model is constructed using the LdaModel class in the Gensim library. The parameters of the model are set, such as the number of topics, the number of iterations, the vocabulary size, etc. The constructed LDA model is used to fit the corpus to learn the distribution of topics and the weights of words. According to the fitted LDA model, the topic link relationship between entities is extracted. The link relationship can be determined based on the similarity of topic distribution, the similarity of word weights, or other relevance measures. Based on the extracted topic link relationship, a potential topic link relationship dataset is constructed. This can be a graph structure or table containing topic link relationships between entities.

[0083] Step S235: Compare each potential topic link relationship data in the potential topic link relationship data set with a preset topic consistency threshold; take the candidate heterogeneous critical disease knowledge source corresponding to the potential topic link relationship data equal to or greater than the preset topic consistency threshold as a high-consistency knowledge source, thereby obtaining a potential critical disease knowledge source set.

[0084] Specifically, for example, a potential topic link relationship dataset can be loaded into a Python environment. A topic consistency threshold is preset, which is used to determine a high consistency knowledge source. This threshold can be a predefined fixed value, or it can be adjusted according to actual needs and domain knowledge. Each potential topic link relationship data in the potential topic link relationship dataset is traversed. According to the preset topic consistency threshold, the consistency of the current potential topic link relationship data with the threshold is calculated. This can be calculated based on the similarity of the topic distribution, relevance measurement or other suitable methods. The calculated topic consistency is compared with the preset topic consistency threshold. If the consistency is greater than or equal to the threshold, the corresponding candidate heterogeneous critical disease knowledge source is marked as a high consistency knowledge source. All candidate heterogeneous critical disease knowledge sources marked as high consistency knowledge sources are collected to form a potential critical disease knowledge source set.

[0085] The present invention can understand the frequency of occurrence of different entity tokens in the field of critical illness by performing word-unit frequency statistics analysis on the entity semantic granularity set. This helps to determine the key entities and concepts in the field of critical illness and provide the importance ranking of the entity. Through word-unit frequency statistics, entities that frequently appear in the field of critical illness can be found, providing an important basis for subsequent semantic similarity calculation and topic link mining. By performing semantic attribute extraction on the entity semantic granularity set, the key attributes and features of the entity can be obtained. This helps to deeply understand the semantic meaning and context of the entity, so as to more accurately perform semantic similarity calculation and topic link mining. By extracting the semantic attributes of the entity, a basis can be provided for establishing a semantic association relationship between entities in subsequent steps. By calculating the entity semantic similarity, the degree of semantic similarity between different entities can be quantified. This helps to find semantically related entities and establish associations between them. By calculating the semantic similarity between entities, similarity matrix data can be provided for subsequent topic link mining, further revealing the hidden semantic relationship between entities. By performing latent semantic topic link mining based on entity token frequency statistics and entity semantic similarity matrix data, potential topic link relationships can be discovered. This helps to reveal the potential semantic associations between different entities, even if they do not have obvious semantic similarities on the surface. Through topic link mining, potential associations between knowledge sources can be discovered, providing more comprehensive and in-depth knowledge acquisition and integration for subsequent steps. By comparing with the preset topic consistency threshold, candidate heterogeneous critical knowledge sources that are equal to or greater than the preset topic consistency threshold are screened out, thereby obtaining a set of potential critical knowledge sources with high consistency. This helps to improve the quality and accuracy of knowledge sources and ensure that the selected knowledge sources are highly relevant to specific topics or fields. By screening high-consistency knowledge sources, more reliable and useful critical knowledge can be provided to provide support for further applications and decision-making.

[0086] Preferably, step S25 comprises the following steps:

[0087] Step S251: constructing a dynamic knowledge update tracking mechanism for the potential critical illness knowledge source set according to the dynamic knowledge update time span data set, and obtaining a knowledge source dynamic update time series mechanism;

[0088] Specifically, for example, a mechanism that can track the dynamic update of knowledge sources can be constructed based on the information in the dynamic knowledge update time span dataset. This mechanism can be a data structure, such as a dictionary or a table. Traverse each knowledge source in the potential critical knowledge source set. Extract the update time information of the current knowledge source from the dynamic knowledge update time span dataset. This can include information such as the timestamp of the most recent update, the update frequency, etc. Add the extracted update time information to the dynamic knowledge update tracking mechanism for subsequent dynamic knowledge update frequency analysis. Obtain a knowledge source dynamic update time series mechanism, which contains the update time information of each potential critical knowledge source.

[0089] Step S252: using the knowledge source dynamic update time series mechanism to perform dynamic knowledge update frequency analysis on each potential knowledge source in the potential critical illness knowledge source set, and obtaining knowledge source dynamic update frequency characteristic data;

[0090] Specifically, for example, for each potential knowledge source, the update frequency can be calculated according to its update time information in the knowledge source dynamic update time series mechanism. The update frequency can be measured by calculating the time difference between two consecutive update times. The frequency characteristics of each potential knowledge source are counted. For example, the average update frequency, the maximum update frequency, and the minimum update frequency can be calculated. The knowledge source dynamic update frequency characteristic data is obtained, which contains the update frequency characteristics of each potential critical knowledge source.

[0091] Step S253: performing time series decomposition on the knowledge source dynamic update frequency feature data to obtain a time-frequency representation decomposition result data set, wherein the time-frequency representation decomposition result data set includes high-frequency time mutation component data and low-frequency knowledge evolution trend component data;

[0092] Specifically, for example, the time series decomposition method can be used to decompose the dynamic update frequency characteristic data of the knowledge source. For example, the Holt-Winters seasonal decomposition method can be used. For the dynamic update frequency characteristic data of each knowledge source, the time series decomposition method is applied to split it into a high-frequency time mutation component and a low-frequency knowledge evolution trend component.

[0093] Step S254: constructing a knowledge abnormal mutation detection model based on the high-frequency time mutation component data to obtain a critical disease knowledge mutation detection model; using the critical disease knowledge mutation detection model to perform knowledge mutation analysis on each potential knowledge source in the potential critical disease knowledge source set to obtain critical disease knowledge mutation index data;

[0094] Specifically, for example, high-frequency time mutation component data can be used as training data to construct a knowledge abnormal mutation detection model. Appropriate machine learning algorithms can be selected, such as support vector machines (SVM) or deep learning models, such as recurrent neural networks (RNN). For each knowledge source in the potential critical illness knowledge source set, the constructed knowledge abnormal mutation detection model is used to perform knowledge mutation analysis. The high-frequency time mutation component data of the knowledge source is input into the model, and the model output is used to determine whether the knowledge source has abnormal mutations. According to the abnormal mutation of the knowledge source, the critical illness knowledge mutation index data is calculated. The calculation method of the index can be defined according to the degree of abnormality. For example, the mean or standardized index of the abnormal score can be used to quantify the critical illness knowledge mutation index.

[0095] Step S255: constructing a frontier time decay model based on the low-frequency knowledge evolution trend component data to obtain a critical care knowledge frontier time decay model; using the critical care knowledge frontier time decay model to perform a frontier time decay analysis on each potential knowledge source in the potential critical care knowledge source set to obtain knowledge frontier decay index data;

[0096] Specifically, for example, appropriate tools or methods can be selected, such as exponential smoothing or ARIMA models in time series analysis methods. The low-frequency knowledge evolution trend component data is used as training data to train the selected tools or methods to construct a critical knowledge frontier time decay model. This model can capture changes in the evolution trend of knowledge sources and predict future frontier time decay. For each knowledge source in the potential critical knowledge source set, its corresponding low-frequency knowledge evolution trend component data is obtained. The low-frequency knowledge evolution trend component data of each knowledge source is input into the constructed critical knowledge frontier time decay model. The prediction results of the model are used to analyze the frontier time decay of the knowledge source. According to the output of the model, the knowledge frontier decay index data can be obtained to measure the degree of frontier decay of the knowledge source over time.

[0097] Step S256: performing weighted fusion calculation on the critical illness knowledge mutation index data and the knowledge frontier decay index data to obtain comprehensive critical illness knowledge frontier score data;

[0098] Specifically, for example, the weights of the critical disease knowledge mutation index data and the knowledge frontier decay index data can be determined, and these weights can be set according to actual needs and domain knowledge. For example, a higher weight can be given to the critical disease knowledge mutation index to highlight important knowledge mutations. The critical disease knowledge mutation index data and the knowledge frontier decay index data are weighted and fused. The weighted average or weighted summation method can be used, and the specific calculation formula is as follows: Comprehensive critical disease knowledge frontier score = (critical disease knowledge mutation index data × weight 1) + (knowledge frontier decay index data × weight 2).

[0099] Step S257: The potential knowledge sources corresponding to the comprehensive critical care knowledge frontier score data greater than the preset frontier score threshold are taken as frontier dynamic knowledge sources, thereby obtaining a critical care frontier dynamic knowledge source set.

[0100] Specifically, for example, for each potential knowledge source, its corresponding comprehensive critical knowledge frontier score can be compared with a preset frontier score threshold. If the comprehensive critical knowledge frontier score is greater than the preset frontier score threshold, the potential knowledge source is marked as a frontier dynamic knowledge source. Repeat the above steps to perform score comparison and marking operations on all potential knowledge sources. The potential knowledge sources marked as frontier dynamic knowledge sources are collected to form a critical frontier dynamic knowledge source set.

[0101] The present invention can track the update of each knowledge source in the potential critical disease knowledge source set by constructing a dynamic knowledge update tracking mechanism. This helps to understand the timeliness and latest dynamics of the knowledge source, and ensures that the selected knowledge source can provide the latest critical disease knowledge and information. By tracking the dynamic update of the knowledge source, the latest critical disease field knowledge can be obtained and utilized in a timely manner, and the accuracy and effectiveness of decision-making can be improved. By analyzing the dynamic update frequency of each knowledge source in the potential critical disease knowledge source set, the update speed and frequency of each knowledge source can be understood. This helps to evaluate the activity and quality of the knowledge source and determine whether the knowledge source has the ability to continuously provide high-quality critical disease knowledge. By analyzing the update frequency feature data, it can be determined which knowledge sources have a higher update frequency and reliability in the critical disease field, providing a basis for subsequent knowledge source selection. By performing time series decomposition on the dynamic update frequency feature data of the knowledge source, the data can be decomposed into different components, including a high-frequency time mutation component and a low-frequency knowledge evolution trend component. This helps to understand the update mode and evolution trend of the knowledge source. The high-frequency time mutation component reflects the changes and fluctuations of the knowledge source in a short period of time, while the low-frequency knowledge evolution trend component reveals the long-term evolution trend and change trend of the knowledge source. Through time series decomposition, the dynamic characteristics of knowledge sources can be better grasped, providing a basis for subsequent knowledge analysis and evaluation. By constructing a knowledge abnormal mutation detection model based on high-frequency time mutation component data, abnormal changes and mutations in knowledge sources can be detected. This helps to discover important changes and emergencies in critical knowledge sources and remind decision makers to pay attention to changes in important information. Using the critical knowledge mutation detection model to analyze the knowledge mutation of knowledge sources in the potential critical knowledge source set, critical knowledge mutation index data can be calculated. These index data can help determine the degree of change and mutation of knowledge sources, so that decision makers can understand the changes of important knowledge in a timely manner and take corresponding countermeasures. Based on the low-frequency knowledge evolution trend component data, a frontier time decay model is constructed to predict the frontier decay of knowledge sources. This helps to understand the decay trend of knowledge sources in time, that is, how the novelty and frontier of knowledge weaken over time. By constructing a frontier time decay model and conducting a frontier time decay analysis on knowledge sources in the potential critical knowledge source set, knowledge frontier decay index data can be calculated. These index data provide a basis for evaluating the degree of frontier decay of knowledge sources, helping decision makers to identify and utilize emerging critical care knowledge in a timely manner. By performing weighted fusion calculations on the critical care knowledge mutation index data and the knowledge frontier decay index data, comprehensive critical care knowledge frontier scoring data can be obtained. This scoring data comprehensively considers the changes in knowledge sources and the frontier decay, and can comprehensively evaluate the importance and credibility of knowledge sources. Comprehensive critical care knowledge frontier scoring data can help decision makers sort and screen potential knowledge sources, give priority to knowledge sources with higher scores, and thus more effectively obtain cutting-edge knowledge in the field of critical care.According to the comprehensive critical care knowledge frontier score data and the preset frontier score threshold, potential knowledge sources with scores exceeding the threshold can be identified as frontier dynamic knowledge sources. These frontier dynamic knowledge sources are considered to provide the latest and important critical care knowledge.

[0102] Preferably, step S3 comprises the following steps:

[0103] Step S31: formulating a time series knowledge acquisition strategy for each potential knowledge source in the critical care frontier dynamic knowledge source set according to the dynamic knowledge update time span data set, and obtaining a time series knowledge acquisition strategy data set;

[0104] Specifically, for example, a time series knowledge acquisition strategy can be formulated based on the dynamic knowledge update time span data of each potential knowledge source. Shorter time spans may require more frequent acquisitions, while longer time spans can adopt more sparse acquisition strategies. According to the preset acquisition strategy rules, formulate corresponding time series knowledge acquisition strategies, such as acquisition frequency, acquisition time point, etc. Repeat the above steps to perform time series knowledge acquisition strategy formulation operations on all potential knowledge sources in the critical care frontier dynamic knowledge source set. Collect the formulated time series knowledge acquisition strategy data to form a time series knowledge acquisition strategy data set.

[0105] Step S32: identifying critical illness theme-related acquisition resources for each potential knowledge source in the critical illness frontier dynamic knowledge source set according to the high-attention ontology data set, and obtaining a critical illness theme-related data set;

[0106] Specifically, for example, for each potential knowledge source in the critical care frontier dynamic knowledge source set, a high-attention ontology dataset can be used to identify acquisition resources related to critical care topics. For each potential knowledge source, use the keywords, terms, or entities in the high-attention ontology dataset to filter and match the content of the knowledge source. Text mining or information retrieval technology can be used to match the text in the knowledge source with the keywords in the high-attention ontology dataset to determine its relevance to the critical care topic. Based on the matching results, the content related to the critical care topic is extracted from the knowledge source to form a dataset related to the critical care topic.

[0107] Step S33: Generate a time-series-theme dual-driven acquisition strategy for each potential knowledge source in the critical care frontier dynamic knowledge source set according to the time-series knowledge acquisition strategy data set and the critical care theme-related data set to obtain a frontier heterogeneous knowledge acquisition strategy set;

[0108] Specifically, for example, for each potential knowledge source in the critical care frontier dynamic knowledge source set, a time series and theme dual-driven acquisition strategy can be generated in combination with the time series knowledge acquisition strategy data set and the critical care theme related data set. First, the acquisition time window of each potential knowledge source is determined according to the acquisition frequency and acquisition time point in the time series knowledge acquisition strategy data set. Secondly, based on the keywords, terms or entities in the critical care theme related data set, screening rules related to critical care themes are formulated. Based on the acquisition time window and screening rules, a time series and theme dual-driven acquisition strategy is generated, that is, within the specified time window, specific critical care theme related content is collected. The time series and theme dual-driven acquisition strategies generated for each potential knowledge source in the critical care frontier dynamic knowledge source set are sorted and summarized to form a frontier heterogeneous knowledge acquisition strategy set.

[0109] Step S34: performing intelligent distributed crawler collection on each potential knowledge source in the critical care frontier dynamic knowledge source set according to the frontier heterogeneous knowledge collection strategy set to obtain an original critical care frontier heterogeneous corpus data set;

[0110] Specifically, for example, according to each strategy in the frontier heterogeneous knowledge collection strategy set, an intelligent distributed crawler tool can be used to collect various potential knowledge sources in the critical care frontier dynamic knowledge source set. Configure the intelligent distributed crawler tool, set the starting link, crawling depth, crawling speed and other parameters of the crawling, and ensure that the crawling operation can be automated according to the requirements of the strategy set. During the crawling process, according to the guidance of the strategy set, the intelligent distributed crawler tool will automatically access the critical care frontier dynamic knowledge source according to the set strategy, and filter and extract content related to critical care topics according to the collection rules. The crawler tool will regularly perform collection tasks according to the set collection frequency and collection time window to ensure that the latest information of the critical care frontier dynamic knowledge source is obtained in a timely manner. The collected data can be saved as the original critical care frontier heterogeneous corpus dataset, including data in various forms such as text, pictures, and videos.

[0111] Step S35: performing structured extraction and semantic annotation on the original critical care frontier heterogeneous corpus dataset to obtain a critical care frontier heterogeneous corpus dataset;

[0112] Specifically, for example, first, the original critical care frontier heterogeneous corpus dataset is preprocessed, including text cleaning, word segmentation, and stop word removal. Then, NLP tools such as named entity recognition (NER), part-of-speech tagging, and syntactic analysis are used to semantically parse and annotate the preprocessed original critical care frontier heterogeneous corpus dataset to identify entities, keywords, phrases, and other information in the text. For entities and concepts specific to the critical care field, domain-specific knowledge bases or glossaries can be used for annotation. Further structured extraction can be performed according to task requirements, such as extracting structured information such as relationships, events, and time series. Finally, the data that has undergone structured extraction and semantic annotation is sorted and stored to obtain a critical care frontier heterogeneous corpus dataset.

[0113] Step S36: Perform heterogeneous knowledge representation learning in the critical care field on the critical care frontier heterogeneous corpus dataset to obtain heterogeneous knowledge representation data in the critical care field.

[0114] Specifically, for example, word embedding models, such as Word2Vec or GloVe, can be used to vectorize the critical care frontier heterogeneous corpus dataset. By training the model, the words in the critical care frontier heterogeneous corpus dataset are converted into low-dimensional dense vector representations to capture the semantic relationship between words. Use topic models, such as Latent Dirichlet Allocation (LDA), to perform topic modeling on the critical care frontier heterogeneous corpus dataset. By identifying the distribution of topics in the critical care frontier heterogeneous corpus dataset, the hidden topic structure and relevance are revealed. Use pre-trained language models, such as BERT or GPT, to perform representation learning on the critical care frontier heterogeneous corpus dataset. Through pre-training and fine-tuning of the model, richer and more semantic text representations are obtained, including word-level, sentence-level, or paragraph-level representations. Use Graph Neural Networks to model heterogeneous data. Embed and fuse the text data in the critical care frontier heterogeneous corpus dataset with other data types (such as knowledge graphs and entity relationships) to obtain comprehensive heterogeneous knowledge representation data in the critical care field.

[0115] The present invention can determine the update frequency and time interval of each potential knowledge source in the critical care frontier dynamic knowledge source by analyzing the dynamic knowledge update time span data set. Formulating a time series knowledge acquisition strategy can ensure the timely acquisition of the latest critical care knowledge, so that decision makers and researchers can keep up with the knowledge frontier. Through the analysis of high-attention ontology data sets, acquisition resources related to critical care topics can be identified. These resources may include academic journals, authoritative websites, professional forums, etc., which cover the latest research results, clinical experience and expert opinions related to critical care. Identifying these resources helps to screen out high-quality critical care knowledge sources and improve the reliability and accuracy of knowledge. Combining the time series knowledge acquisition strategy data set and the critical care topic-related data set, a dual-driven acquisition strategy can be generated. This strategy combines time series and topic relevance to ensure that the collected knowledge is both timely and relevant to the critical care topic. The generated frontier heterogeneous knowledge acquisition strategy set can guide the acquisition of knowledge more accurately and improve the practicality and adaptability of knowledge. According to the frontier heterogeneous knowledge acquisition strategy set, the intelligent distributed crawler technology can be used to automatically collect the required knowledge from the critical care frontier dynamic knowledge source. This collection method is efficient and accurate, which can greatly save human resources and ensure the stability of the quality and quantity of the collected knowledge sources. Structured extraction of the original critical care frontier heterogeneous corpus dataset can extract useful information and knowledge from unstructured text data. By extracting entities, relationships, events and other elements in the text, the dataset can be converted into a structured form, making the knowledge in it easier to manage, analyze and apply. Semantic annotation of the critical care frontier heterogeneous corpus dataset can give semantic labels and annotations to the text content in the dataset. By annotating the entity type, relationship, sentiment tendency and other information of the text, the semantic understanding and expression ability of the text can be improved, providing more useful information for subsequent knowledge processing and analysis. By performing knowledge representation learning on the critical care frontier heterogeneous corpus dataset, the knowledge in it can be converted into a form that can be understood and processed by machines. Using machine learning and deep learning techniques, the representation of heterogeneous knowledge in the critical care field can be learned, such as the vector representation of medical entities and the embedding of semantic relationships. Such representation can better capture the association and semantic information between knowledge, providing a basis for subsequent knowledge reasoning, matching and application.

[0116] Preferably, step S4 comprises the following steps:

[0117] Step S41: constructing a heterogeneous knowledge fusion network based on heterogeneous knowledge representation data in the critical care field to obtain heterogeneous knowledge fusion network structure data;

[0118] Specifically, for example, graph neural network methods, such as Graph Convolutional Networks (GCN) or Graph Attention Networks (GAT), can be used to construct a heterogeneous knowledge fusion network structure. The heterogeneous knowledge representation data in the critical care field is used as node features to construct a graph structure, and the connection and feature information between nodes are used for knowledge fusion. If there is knowledge graph data in the critical care field, the heterogeneous knowledge representation data can be fused with the knowledge graph using knowledge graph fusion methods, such as Knowledge Graph Embedding (KGE). The entities, relationships, and attribute information in the knowledge graph are used to embed them into the network and jointly represent and fuse them with the heterogeneous knowledge representation data. If the heterogeneous knowledge representation data contains multiple modalities, such as text, images, or time series data, multimodal fusion methods, such as multimodal neural networks or attention mechanisms, can be used to fuse data of different modalities. The features of different modalities are combined to build a comprehensive heterogeneous knowledge fusion network.

[0119] Step S42: extract clinical frontier feature descriptors from the heterogeneous knowledge representation data in the critical care field to obtain clinical frontier feature descriptor data for critical care; perform transfer learning and fine-grained tuning on the heterogeneous knowledge fusion network structure data based on the clinical frontier feature descriptor data for critical care to obtain a clinical frontier knowledge-aware heterogeneous neural network;

[0120] Specifically, for example, feature descriptors related to the clinical frontier can be designed and extracted according to the characteristics of the critical care field. These feature descriptors can include statistical features, frequency features, text features, semantic features, etc. to capture the key information of the critical care clinical frontier. Use deep learning models, such as convolutional neural networks (CNNs) or long short-term memory networks (LSTMs), to extract features from heterogeneous knowledge representation data in the critical care field. By training deep learning models, learn to extract the representation ability of feature descriptors of the critical care clinical frontier. Using heterogeneous knowledge fusion network structure data as the initial model, fine-tune and transfer learning are performed on the critical care clinical frontier feature descriptor data. By freezing some network layers or setting a smaller learning rate, the network can adapt to the task requirements of the critical care clinical frontier feature descriptor data. On the basis of transfer learning, further fine-grained tuning of heterogeneous neural networks is performed. Adjust the network structure, optimize hyperparameters, increase the training set, or introduce regularization to optimize and improve the network.

[0121] Step S43: deploying the clinical frontier knowledge-aware heterogeneous neural network in a distributed computing cluster to obtain a critical care clinical frontier knowledge tuple detector;

[0122] Specifically, for example, a distributed computing cluster of appropriate size, including multiple computers or servers, can be configured to deploy clinical frontier knowledge-aware heterogeneous neural networks. Use popular deep learning frameworks, such as TensorFlow, PyTorch, or Keras, to deploy models on the distributed computing cluster according to the definition and structure of the heterogeneous neural network. Configure appropriate computing resources and environments to support the training and inference of heterogeneous neural networks. Use the computing power of the distributed computing cluster to perform the training process or inference process in a distributed manner. Technologies such as data parallelism or model parallelism can be used to distribute computing tasks to different computing nodes and perform collaborative computing. Package the clinical frontier knowledge-aware heterogeneous neural network into a container, such as a Docker container. Use container orchestration tools, such as Kubernetes, to deploy the containerized neural network model to the distributed computing cluster and manage its running status.

[0123] Step S44: extracting clinical frontier knowledge tuples from the critical care frontier heterogeneous corpus data set using a critical care frontier knowledge tuple detector to obtain clinical frontier knowledge tuple data;

[0124] Specifically, for example, the heterogeneous corpus data set of the critical care frontier can be preprocessed, such as word segmentation, part-of-speech tagging, syntactic analysis, etc. Use the critical care clinical frontier knowledge tuple detector to process the preprocessed heterogeneous corpus data and extract the clinical frontier knowledge tuples therein. The knowledge tuple can be a data structure containing entities, relationships, and attributes related to the critical care field, which is used to represent the key information of clinical frontier knowledge. The extracted clinical frontier knowledge tuple data is stored and organized, and can be managed and queried using a database or other appropriate data management tools.

[0125] Step S45: performing a topic cluster analysis on the clinical frontier knowledge tuple data to obtain clinical frontier knowledge topic feature data;

[0126] Specifically, for example, text features can be extracted for each clinical frontier knowledge tuple and represented as a numerical vector. Commonly used text feature extraction methods include bag-of-words model, TF-IDF, Word2Vec, etc. Natural language processing (NLP) tool libraries, such as NLTK, spaCy, or Gensim, can be used to extract text features. The extracted text feature data is applied to the topic clustering algorithm to group the clinical frontier knowledge tuples into different topics. Commonly used topic clustering algorithms include K-means, hierarchical clustering, and LDA (Latent Dirichlet Allocation). Machine learning libraries, such as scikit-learn or clustering algorithm implementations in NLTK, can be used to perform topic clustering analysis. Extract the key feature words or feature vectors of each topic from the topic clustering results as knowledge topic feature data.

[0127] Step S46: Based on the heterogeneous knowledge representation data in the critical care field, the heterogeneous association relationship reasoning is performed on the clinical frontier knowledge subject feature data to obtain a heterogeneous relationship completion knowledge base; and the heterogeneous relationship completion knowledge base is used to perform intelligent triple completion on the clinical frontier knowledge tuple data to obtain clinical completion relationship triple data;

[0128] Specifically, for example, knowledge graph reasoning algorithms, such as rule-based reasoning, path reasoning, graph neural networks, and other methods, can be used to perform heterogeneous association relationship reasoning on clinical frontier knowledge subject feature data based on heterogeneous knowledge representation data in the critical care field. Based on the existing domain knowledge and relationship rules, the potential association relationships in the clinical frontier knowledge tuple data are inferred. The inferred heterogeneous association relationships are sorted and stored to build a heterogeneous relationship completion knowledge base. A graph database, a relational database, or other appropriate data storage method can be used to manage and query the heterogeneous relationship completion knowledge base. Using the constructed heterogeneous relationship completion knowledge base, intelligent triple completion is performed on the clinical frontier knowledge tuple data. For each clinical frontier knowledge tuple to be completed, the existing relationships in the heterogeneous relationship completion knowledge base are used for matching and reasoning to find the completion relationship. A graph database query language (such as Cypher) or other query tools can be used to perform intelligent triple completion based on the attributes of the clinical frontier knowledge tuple and the relationships in the heterogeneous relationship completion knowledge base.

[0129] Step S47: Perform knowledge graph embedding learning on the clinical completion relationship triple data and the heterogeneous knowledge representation data in the critical care field to obtain a primary critical care knowledge graph.

[0130] Specifically, for example, a suitable knowledge graph embedding method can be selected to learn the data and embed the entities and relationships in the knowledge graph into a low-dimensional vector space. Common knowledge graph embedding methods include TransE, TransR, ComplEx, DistMult, etc. Open source graph embedding tool libraries, such as OpenKE, GraphVite, or DeepGraphLibrary (DGL), can be used to perform knowledge graph embedding learning. Construct a primary critical care knowledge graph based on the low-dimensional vector representation obtained from the knowledge graph embedding learning. Convert the embedded entities and relationships into nodes and edges of the graph. Graph databases or graph processing libraries, such as Neo4j, NetworkX, or Graph-tool, can be used to build and store primary critical care knowledge graphs.

[0131] The present invention can integrate and fuse different types and forms of knowledge by constructing a heterogeneous knowledge fusion network. This can effectively interact and transmit information from different knowledge sources, and promote the complementarity and synergy between knowledge. This helps to reveal the associations and potential hidden patterns between knowledge, and provide a more comprehensive and integrated perspective on critical knowledge. By extracting clinical frontier feature descriptors from heterogeneous knowledge representation data in the critical field, feature descriptors related to the critical clinical front can be extracted. These descriptors can capture specific features and patterns in the critical field, have strong distinguishing and characterizing capabilities, and help to accurately describe and represent critical clinical frontier knowledge. Based on the critical clinical front feature descriptor data, transfer learning and fine-grained tuning are performed on the heterogeneous knowledge fusion network structure data. Transfer learning uses existing knowledge and models to transfer and reuse knowledge on new tasks, accelerate the learning process and improve performance. Fine-grained tuning further improves the accuracy and generalization ability of the clinical front knowledge perception heterogeneous neural network through fine-tuning and optimization for specific tasks. These technical means can improve the adaptability and reliability of the model, so that it can be better applied to the perception and analysis of critical clinical frontier knowledge. Distributed computing cluster deployment of clinical frontier knowledge-aware heterogeneous neural networks can make full use of computing resources and improve the parallel processing and computing efficiency of the system. Through distributed deployment, the process of knowledge tuple detection can be accelerated to achieve rapid knowledge perception and extraction. The deployed clinical frontier knowledge-aware heterogeneous neural network can be regarded as a knowledge tuple detector. It can automatically discover and extract clinical frontier knowledge tuples, that is, important knowledge fragments or related information, from the critical care frontier heterogeneous corpus dataset. Such a detector is efficient and accurate, and can help medical researchers and clinicians quickly obtain the latest knowledge and cutting-edge information in the field of critical care. Using the critical care clinical frontier knowledge tuple detector, knowledge tuples are extracted from the critical care frontier heterogeneous corpus dataset. These knowledge tuples can contain important information such as key concepts, associations, treatment plans, and prognostic indicators in the field of critical care. The extracted clinical frontier knowledge tuple data can help medical researchers gain a deeper understanding of the latest progress and cutting-edge knowledge in the field of critical care. By performing topic clustering analysis on the clinical frontier knowledge tuple data, knowledge tuples with similar topics can be classified into the same group. Such clustering analysis helps to discover the potential associations and common features between knowledge tuples and extract the subject features of cutting-edge clinical knowledge. Through subject clustering, knowledge can be organized and understood in a more fine-grained manner, which facilitates the subsequent heterogeneous association reasoning and knowledge graph embedding learning. Based on the heterogeneous knowledge representation data in the critical care field, the potential associations and dependencies between different knowledge topics can be revealed by reasoning and analyzing the heterogeneous association relationships in the subject feature data of cutting-edge clinical knowledge. Such a reasoning process can expand the breadth and depth of knowledge, discover hidden association patterns and interactions between knowledge, and enrich the content and structure of the knowledge base.Through heterogeneous association relationship reasoning, a heterogeneous relationship completion knowledge base is obtained. This knowledge base contains the association relationships that are not explicitly expressed in the original clinical frontier knowledge tuple data, thereby enriching the information content of the knowledge base. The establishment of a heterogeneous relationship completion knowledge base provides a more comprehensive and accurate knowledge graph foundation. Using the heterogeneous relationship completion knowledge base, intelligent triple completion is performed on the clinical frontier knowledge tuple data. This means that on the basis of the existing knowledge tuples, by completing the missing relationship triples, the gaps and incomplete parts of the knowledge are filled, and a more complete and consistent knowledge representation is provided. By performing knowledge graph embedding learning on clinical completion relationship triple data and heterogeneous knowledge representation data in the critical care field, the entities and relationships in the knowledge graph can be mapped to a low-dimensional vector space. This helps to capture the semantic relationships and similarities between knowledge, extract the implicit features and potential laws of knowledge, and thus provide a more effective basis for subsequent knowledge reasoning, similarity matching, and data mining.

[0132] Preferably, step S5 comprises the following steps:

[0133] Step S51: performing abnormal structure pattern recognition on the primary critical illness knowledge graph to obtain graph defect pattern data;

[0134] Specifically, for example, the graph structure in the primary critical care knowledge graph can be analyzed to identify abnormal structural patterns therein. Abnormal structural patterns include missing nodes, missing edges, loops, isolated nodes, etc. Graph theory algorithms, graph matching algorithms, or graph analysis tools can be used to identify abnormal structural patterns. The identified abnormal structural patterns are converted into graph defect pattern data. Defect pattern data describes the structural defects or problems existing in the primary critical care knowledge graph. Defect pattern data can be extracted and stored using a custom data structure or a standardized graph defect pattern representation method.

[0135] Step S52: Perform heuristic graph structure calibration on the primary critical disease knowledge graph according to the graph defect pattern data to obtain a graph defect repair atomic operation set;

[0136] Specifically, for example, a series of heuristic rules and operations can be designed to calibrate the graph structure of the primary critical care knowledge graph based on the information provided by the graph defect pattern data. The heuristic rules can be based on expert experience, graph analysis methods, or common graph structure repair strategies. The heuristic graph structure calibration algorithm can be implemented using graph processing libraries and graph analysis tools. The operations applied during the heuristic graph structure calibration process are recorded to form a graph defect repair atomic operation set. The defect repair atomic operation set describes the specific operational steps for repairing graph structure defects in the primary critical care knowledge graph. The defect repair atomic operation set can be extracted and stored using a custom data structure or a standardized operation record format.

[0137] Step S53: Perform subcritical graph expansion on the primary critical disease knowledge graph according to the graph defect repair atomic operation set to obtain critical disease graph defect repair data;

[0138] Specifically, for example, the primary critical disease knowledge graph can be expanded step by step in sequence according to the operation steps in the graph defect repair atomic operation set. Each operation can be to add a new node, add a new edge, or modify the properties of existing nodes and edges. The operation sequence and expansion method can be defined and designed according to the specific defect repair atomic operation set. The intermediate results or final results generated in the process of subcritical graph expansion are extracted to form the critical disease graph defect repair data. The defect repair data contains the structure and content changes of the critical disease knowledge graph after the repair operation.

[0139] Step S54: Repair edge connections and expand nodes of the primary critical illness knowledge graph according to the critical illness graph defect repair data to obtain a cutting-edge knowledge graph in the critical illness field.

[0140] Specifically, for example, the primary critical illness knowledge graph can be repaired for edge connections and expanded for nodes according to the repair operations described in the critical illness graph defect repair data. Edge connection repair can include operations such as adding missing edges and repairing incorrect edge connections. Node expansion can include operations such as adding new nodes and extending the properties of existing nodes. The repair operation can be defined and designed based on the specific defect repair data. Extract part or all of the frontier knowledge graph in the critical illness field from the repaired critical illness knowledge graph. The frontier knowledge graph can contain new nodes, new edges, and expansion of node attributes after repair.

[0141] The present invention can detect potential problems or defects in the graph by performing abnormal structural pattern recognition on the primary critical illness knowledge graph. This helps to discover structural errors, inconsistencies or incompleteness in the knowledge graph. By identifying abnormal structural patterns, the quality and accuracy of the knowledge graph can be improved, providing guidance for subsequent repair and optimization. Based on the graph defect pattern data, the primary critical illness knowledge graph is heuristically calibrated. This can adjust and correct the structure of the graph in an automated manner according to the detected defect pattern. Heuristic calibration can automatically correct structural errors, inaccurate relationships and other problems in the graph, and improve the quality and consistency of the graph. The graph defect repair atomic operation set contains a series of repair operations for graph defects, such as adding, deleting, and modifying relationships. According to the graph defect repair atomic operation set, the primary critical illness knowledge graph is subcritically amplified. This can be a process for repairing and supplementing defects in the graph. By executing the operations in the repair atomic operation set, structural errors, missing or inaccurate relationships in the graph can be repaired. The critical illness graph defect repair data represents the result after repairing the primary critical illness knowledge graph. The repair data contains information such as newly added relationships, modified relationships, and deleted erroneous relationships. According to the critical illness graph defect repair data, the primary critical illness knowledge graph is repaired for edge connections and expanded for nodes. This can repair edge connection errors and node missing problems in the graph. By repairing edge connection errors, the relationships in the graph can be ensured to be accurate; by expanding nodes, the knowledge coverage and depth of the graph can be increased. This helps to improve the integrity and usability of the critical illness knowledge graph. The frontier knowledge graph in the critical illness field is generated during the repair and optimization process, and contains repaired and expanded relationship and node information. It represents the latest knowledge and cutting-edge advances in the critical illness field, and provides a comprehensive and accurate knowledge resource for researchers, experts, and decision makers.

[0142] Preferably, the present invention also provides a system for automatically extracting critical disease domain knowledge, which is used to execute the method for automatically extracting critical disease domain knowledge as described above. The system for automatically extracting critical disease domain knowledge includes:

[0143] The critical illness ontology construction module is used to construct the critical illness ontology in the critical illness field to obtain the critical illness ontology data set; perform high-attention ontology mining on the critical illness ontology data set to obtain the high-attention ontology data set;

[0144] The knowledge source analysis module is used to perform topic consistency link analysis based on the high-attention ontology data set to obtain a potential critical disease knowledge source set; perform temporal knowledge fluctuation analysis on the potential critical disease knowledge source set to obtain a dynamic knowledge update time span data set; perform frontier knowledge source hierarchical screening on the potential critical disease knowledge source set based on the dynamic knowledge update time span data set to obtain a critical disease frontier dynamic knowledge source set;

[0145] The heterogeneous corpus collection module is used to perform topic-driven crawling on the critical care frontier dynamic knowledge source set based on the high-attention ontology dataset and the dynamic knowledge update time span dataset to obtain the critical care frontier heterogeneous corpus dataset; perform heterogeneous knowledge representation learning in the critical care field on the critical care frontier heterogeneous corpus dataset to obtain heterogeneous knowledge representation data in the critical care field;

[0146] The frontier knowledge extraction module is used to extract the clinical frontier knowledge of critical illness from the frontier heterogeneous corpus data set of critical illness according to the heterogeneous knowledge representation data in the critical illness field, and obtain the clinical frontier knowledge tuple data; based on the heterogeneous knowledge representation data in the critical illness field, the clinical frontier knowledge tuple data is intelligently completed to obtain the clinical completion relationship triple data; the clinical completion relationship triple data and the heterogeneous knowledge representation data in the critical illness field are embedded in the knowledge graph to obtain the primary critical illness knowledge graph;

[0147] The knowledge graph repair module is used to perform heuristic graph completion reasoning based on the primary critical illness knowledge graph to obtain critical illness graph defect repair data; according to the critical illness graph defect repair data, the primary critical illness knowledge graph is repaired for edge connections and expanded for nodes to obtain a cutting-edge knowledge graph in the critical illness field.

[0148] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0149] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for automatically extracting critical care domain knowledge, characterized in that: The following steps are involved: Step S1: construct a critical care domain ontology for the critical care domain to obtain a critical care domain ontology data set; perform high-attention ontology mining on the critical care domain ontology data set to obtain a high-attention ontology data set; Step S2: Perform topic consistency link analysis on the high-attention ontology dataset to obtain a potential critical disease knowledge source set; perform temporal knowledge fluctuation analysis on the potential critical disease knowledge source set to obtain a dynamic knowledge update time span dataset; According to the dynamic knowledge update time span data set, the potential critical disease knowledge source set is screened by frontier knowledge source hierarchical level to obtain the critical disease frontier dynamic knowledge source set, wherein the frontier knowledge source hierarchical level screening is specifically as follows: According to the dynamic knowledge update time span data set, a dynamic knowledge update tracking mechanism is constructed for the potential critical illness knowledge source set, and a knowledge source dynamic update time series mechanism is obtained; The knowledge source dynamic update time series mechanism is used to analyze the dynamic knowledge update frequency of each potential knowledge source in the potential critical care knowledge source set, and obtain the knowledge source dynamic update frequency characteristic data; Perform time series decomposition on the dynamic update frequency feature data of the knowledge source to obtain a time-frequency representation decomposition result data set, wherein the time-frequency representation decomposition result data set includes high-frequency time mutation component data and low-frequency knowledge evolution trend component data; Based on the high-frequency time mutation component data, a knowledge abnormal mutation detection model is constructed to obtain a critical disease knowledge mutation detection model; the critical disease knowledge mutation detection model is used to analyze the knowledge mutation of each potential knowledge source in the potential critical disease knowledge source set to obtain critical disease knowledge mutation index data; Based on the low-frequency knowledge evolution trend component data, a frontier time decay model is constructed to obtain the critical care knowledge frontier time decay model; the critical care knowledge frontier time decay model is used to conduct a frontier time decay analysis on each potential knowledge source in the potential critical care knowledge source set to obtain the knowledge frontier decay index data; The critical care knowledge mutation index data and the knowledge frontier decay index data were weighted and fused to obtain the comprehensive critical care knowledge frontier score data; The potential knowledge sources corresponding to the comprehensive critical care knowledge frontier score data greater than the preset frontier score threshold are taken as frontier dynamic knowledge sources, thereby obtaining a critical care frontier dynamic knowledge source set; Based on the high-attention ontology dataset and the dynamic knowledge update time span dataset, we perform topic-driven crawling on the critical care frontier dynamic knowledge source set to obtain the critical care frontier heterogeneous corpus dataset; we perform heterogeneous knowledge representation learning in the critical care field on the critical care frontier heterogeneous corpus dataset to obtain heterogeneous knowledge representation data in the critical care field; Step S4: extract critical care clinical frontier knowledge from the critical care frontier heterogeneous corpus data set according to the critical care field heterogeneous knowledge representation data to obtain clinical frontier knowledge tuple data; perform intelligent triple completion on the clinical frontier knowledge tuple data based on the critical care field heterogeneous knowledge representation data to obtain clinical completion relationship triple data; perform knowledge graph embedding learning on the clinical completion relationship triple data and the critical care field heterogeneous knowledge representation data to obtain a primary critical care knowledge graph; Step S5: Perform heuristic graph completion reasoning based on the primary critical illness knowledge graph to obtain critical illness graph defect repair data; perform edge connection repair and node expansion on the primary critical illness knowledge graph based on the critical illness graph defect repair data to obtain a cutting-edge knowledge graph in the critical illness field.

2. The method for automatically extracting critical care domain knowledge according to claim 1 is characterized in that: Step S1 includes the following steps: Step S11: Acquire a critical care medical big data set; Step S12: performing named entity recognition on the critical care medical big data set to obtain all named entity sets in the critical care field; Step S13: Perform domain semantic filtering on all named entity sets in the critical illness field to obtain a core named entity set in the critical illness field; Step S14: constructing a critical illness domain ontology for the critical illness domain according to the critical illness domain core named entity set to obtain a critical illness domain ontology dataset; Step S15: Perform high-attention ontology mining on the critical care domain ontology dataset to obtain a high-attention ontology dataset.

3. The method for automatically extracting critical care domain knowledge according to claim 2 is characterized in that: Step S14 includes the following steps: Step S141: Perform entity concept diffusion analysis on the core named entity set in the critical illness field to obtain critical illness entity concept diffusion trend data; Step S142: constructing an initial entity concept network for the core named entity set in the critical illness field to obtain an initial entity concept network model; Step S143: mining constraint knowledge rules on the initial entity concept network model based on the critical illness entity concept diffusion trend data to obtain critical illness domain knowledge constraint rule data; Step S144: optimizing the constraint rules of the initial entity concept network model according to the critical disease field knowledge constraint rule data to obtain an optimized entity concept network model; Step S145: Acquire a real-time knowledge base in the critical care field; interactively iteratively optimize the optimized entity concept network model and the real-time knowledge base in the critical care field to obtain an ontology data set in the critical care field.

4. The method for automatically extracting critical care domain knowledge according to claim 1 is characterized in that: Step S2 includes the following steps: Step S21: performing entity semantic granulation on the high-interest ontology data set to obtain a critical disease domain entity semantic granularity set; Step S22: performing heterogeneous critical illness domain knowledge source discovery on the critical illness domain entity semantic granularity set to obtain a candidate heterogeneous critical illness knowledge source set; Step S23: performing topic consistency link analysis on the candidate heterogeneous critical illness knowledge source set based on the critical illness domain entity semantic granularity set to obtain a potential critical illness knowledge source set; Step S24: performing time series knowledge fluctuation analysis on the potential critical illness knowledge source set to obtain a dynamic knowledge update time span data set; Step S25: Perform frontier knowledge source hierarchical screening on the potential critical disease knowledge source set according to the dynamic knowledge update time span data set to obtain the critical disease frontier dynamic knowledge source set.

5. The method for automatically extracting critical care domain knowledge according to claim 4 is characterized in that: Step S23 includes the following steps: Step S231: Perform entity word unit frequency statistical analysis on the entity semantic granularity set in the critical disease field to obtain critical disease entity word unit frequency statistical data; Step S232: extracting entity semantic attributes from the entity semantic granularity set in the critical disease field to obtain critical disease entity semantic attribute data; Step S233: Calculate entity semantic similarity of the entity semantic granularity set in the critical disease field according to the critical disease entity semantic attribute data to obtain critical disease entity semantic similarity matrix data; Step S234: Based on the statistical data of critical illness entity word frequency and the critical illness entity semantic similarity matrix data, latent semantic topic link mining is performed on the candidate heterogeneous critical illness knowledge source set to obtain a potential topic link relationship data set; Step S235: Compare each potential topic link relationship data in the potential topic link relationship data set with a preset topic consistency threshold; take the candidate heterogeneous critical disease knowledge source corresponding to the potential topic link relationship data equal to or greater than the preset topic consistency threshold as a high-consistency knowledge source, thereby obtaining a potential critical disease knowledge source set.

6. The method for automatically extracting critical care domain knowledge according to claim 1 is characterized in that: Step S3 includes the following steps: Step S31: formulating a time series knowledge acquisition strategy for each potential knowledge source in the critical care frontier dynamic knowledge source set according to the dynamic knowledge update time span data set, and obtaining a time series knowledge acquisition strategy data set; Step S32: identifying critical illness theme-related acquisition resources for each potential knowledge source in the critical illness frontier dynamic knowledge source set according to the high-attention ontology data set, and obtaining a critical illness theme-related data set; Step S33: Generate a time-series-theme dual-driven acquisition strategy for each potential knowledge source in the critical care frontier dynamic knowledge source set according to the time-series knowledge acquisition strategy data set and the critical care theme-related data set to obtain a frontier heterogeneous knowledge acquisition strategy set; Step S34: performing intelligent distributed crawler collection on each potential knowledge source in the critical care frontier dynamic knowledge source set according to the frontier heterogeneous knowledge collection strategy set to obtain an original critical care frontier heterogeneous corpus data set; Step S35: performing structured extraction and semantic annotation on the original critical care frontier heterogeneous corpus dataset to obtain a critical care frontier heterogeneous corpus dataset; Step S36: Perform heterogeneous knowledge representation learning in the critical care field on the critical care frontier heterogeneous corpus dataset to obtain heterogeneous knowledge representation data in the critical care field.

7. The method for automatically extracting critical care domain knowledge according to claim 1 is characterized in that: Step S4 includes the following steps: Step S41: constructing a heterogeneous knowledge fusion network based on heterogeneous knowledge representation data in the critical care field to obtain heterogeneous knowledge fusion network structure data; Step S42: extract clinical frontier feature descriptors from the heterogeneous knowledge representation data in the critical care field to obtain clinical frontier feature descriptor data for critical care; perform transfer learning and fine-grained tuning on the heterogeneous knowledge fusion network structure data based on the clinical frontier feature descriptor data for critical care to obtain a clinical frontier knowledge-aware heterogeneous neural network; Step S43: deploying the clinical frontier knowledge-aware heterogeneous neural network in a distributed computing cluster to obtain a critical care clinical frontier knowledge tuple detector; Step S44: extracting clinical frontier knowledge tuples from the critical care frontier heterogeneous corpus data set using a critical care frontier knowledge tuple detector to obtain clinical frontier knowledge tuple data; Step S45: performing a topic cluster analysis on the clinical frontier knowledge tuple data to obtain clinical frontier knowledge topic feature data; Step S46: Based on the heterogeneous knowledge representation data in the critical care field, the heterogeneous association relationship reasoning is performed on the clinical frontier knowledge subject feature data to obtain a heterogeneous relationship completion knowledge base; and the heterogeneous relationship completion knowledge base is used to perform intelligent triple completion on the clinical frontier knowledge tuple data to obtain clinical completion relationship triple data; Step S47: Perform knowledge graph embedding learning on the clinical completion relationship triple data and the heterogeneous knowledge representation data in the critical care field to obtain a primary critical care knowledge graph.

8. The method for automatically extracting critical care domain knowledge according to claim 1, characterized in that: Step S5 includes the following steps: Step S51: performing abnormal structure pattern recognition on the primary critical illness knowledge graph to obtain graph defect pattern data; Step S52: Perform heuristic graph structure calibration on the primary critical disease knowledge graph according to the graph defect pattern data to obtain a graph defect repair atomic operation set; Step S53: Perform subcritical graph expansion on the primary critical disease knowledge graph according to the graph defect repair atomic operation set to obtain critical disease graph defect repair data; Step S54: Repair edge connections and expand nodes of the primary critical illness knowledge graph according to the critical illness graph defect repair data to obtain a cutting-edge knowledge graph in the critical illness field.

9. A system for automatically extracting critical care domain knowledge, characterized in that: The method for automatically extracting critical disease domain knowledge according to claim 1 is used to implement the system for automatically extracting critical disease domain knowledge, comprising: The critical illness ontology construction module is used to construct the critical illness ontology in the critical illness field to obtain the critical illness ontology data set; perform high-attention ontology mining on the critical illness ontology data set to obtain the high-attention ontology data set; The knowledge source analysis module is used to perform topic consistency link analysis based on the high-attention ontology data set to obtain a potential critical disease knowledge source set; perform temporal knowledge fluctuation analysis on the potential critical disease knowledge source set to obtain a dynamic knowledge update time span data set; perform frontier knowledge source hierarchical screening on the potential critical disease knowledge source set based on the dynamic knowledge update time span data set to obtain a critical disease frontier dynamic knowledge source set; The heterogeneous corpus collection module is used to perform topic-driven crawling on the critical care frontier dynamic knowledge source set based on the high-attention ontology dataset and the dynamic knowledge update time span dataset to obtain the critical care frontier heterogeneous corpus dataset; perform heterogeneous knowledge representation learning in the critical care field on the critical care frontier heterogeneous corpus dataset to obtain heterogeneous knowledge representation data in the critical care field; The frontier knowledge extraction module is used to extract the clinical frontier knowledge of critical illness from the frontier heterogeneous corpus data set of critical illness according to the heterogeneous knowledge representation data in the critical illness field, and obtain the clinical frontier knowledge tuple data; based on the heterogeneous knowledge representation data in the critical illness field, the clinical frontier knowledge tuple data is intelligently completed to obtain the clinical completion relationship triple data; the clinical completion relationship triple data and the heterogeneous knowledge representation data in the critical illness field are embedded in the knowledge graph to obtain the primary critical illness knowledge graph; The knowledge graph repair module is used to perform heuristic graph completion reasoning based on the primary critical illness knowledge graph to obtain critical illness graph defect repair data; according to the critical illness graph defect repair data, the primary critical illness knowledge graph is repaired for edge connections and expanded for nodes to obtain a cutting-edge knowledge graph in the critical illness field.

Images