Professional knowledge graph construction method for medical field

Abstract

The invention relates to a professional knowledge graph construction method for the medical field, and the method comprises the following steps: obtaining a multi-modal medical document, carrying out the deep analysis through a Transform-based multi-modal model, and obtaining an analyzed multi-modal medical document; dynamically segmenting the analyzed multi-modal medical document by adopting a context-based adaptive text segmentation method to obtain independent text blocks; based on the independent text blocks, performing structured extraction, and performing verification by adopting a dynamic dual-verification mechanism to form knowledge sub-graphs; and based on the knowledge sub-graph, further constructing a standardized medical knowledge graph. Compared with the prior art, the method has the advantages of high precision, expandability and the like.

Description

Technical Field

[0001] The present invention relates to the fields of medical information processing and artificial intelligence technology, and in particular to a method for constructing a professional knowledge graph for the medical field. Background Art

[0002] The expression and dissemination of medical knowledge relies heavily on unstructured documents, such as clinical guidelines, medical literature, electronic medical records, and imaging reports. These documents contain complex relationships between entities such as diseases, drugs, symptoms, genes, and treatment plans (e.g., "Drug A treats disease B, but its contraindication is gene mutation C"). Traditional knowledge graph construction methods have the following limitations:

[0003] 1. Data heterogeneity challenge:

[0004] 1) Multimodal data dispersion: Medical knowledge carriers include scanned images (paper medical records), unstructured PDFs (including tables and charts), electronic text, etc. Traditional processing tools have difficulty uniformly parsing and retaining structured information (difficulty in accurately extracting the content of scanned documents, such as the relationship between rows and columns in tables, and the semantic association between charts and text).

[0005] 2) Terminology complexity: Medical entities have a large number of synonyms (such as "non-small cell lung cancer" and "NSCLC"), abbreviations (such as "EGFR" for epidermal growth factor receptor), and cross-language expressions (such as "CKD" for "chronic kidney disease"), which makes entity alignment and normalization difficult.

[0006] 2. Insufficient deep reasoning about relationships:

[0007] 1) Multi-hop relational dependencies: Medical decision-making often requires cross-document reasoning. For example, to determine whether drug D is effective for patients with gene mutation E, it is necessary to integrate drug instructions, genetic test reports, and clinical trial data.

[0008] 2) Context sensitivity: The relationship of the same entity in different contexts may be completely different (e.g., “drug F causes side effect G” and “drug F relieves symptom G” require contextual judgment).

[0009] 3. Dynamic updates and scalability bottlenecks:

[0010] 3) Medical knowledge is rapidly iterating (such as the release of new clinical guidelines), and traditional manual maintenance methods cannot update the atlas in real time.

[0011] In existing technologies, rule-based template extraction methods (such as regular expression matching) have poor generalization capabilities and cannot adapt to the diversity of medical texts; while general large language models have semantic understanding capabilities, they lack professional knowledge in the medical field and are prone to incorrect associations. Summary of the Invention

[0012] The purpose of this invention is to provide a method for constructing a high-precision and scalable medical knowledge graph for the professional knowledge graph in the medical field.

[0013] The purpose of the present invention can be achieved by the following technical solutions:

[0014] A method for constructing a professional knowledge graph in the medical field includes the following steps:

[0015] Obtain multimodal medical documents and perform in-depth analysis using a Transformer-based multimodal model to obtain the parsed multimodal medical documents;

[0016] Dynamically segmenting the parsed multimodal medical document using a context-based adaptive text segmentation method to obtain independent text blocks;

[0017] Based on the independent text blocks, structured extraction is performed and a dynamic double verification mechanism is used for verification to form a knowledge subgraph;

[0018] Based on the knowledge subgraph, a standardized medical knowledge graph is further constructed.

[0019] Furthermore, the step of obtaining the parsed multimodal medical document includes:

[0020] Based on the multimodal medical document, an extraction network extracts visual features to obtain region positioning and classification;

[0021] Based on the region positioning and classification, a natural language processing model is used to perform text embedding;

[0022] Layout encoding is performed based on the visual features and text embedding results to obtain a parsed multimodal medical document.

[0023] Furthermore, the step of obtaining area positioning and classification includes:

[0024] Preprocessing the scanned image of the multimodal medical document, including image denoising, enhancement, and standardization;

[0025] Based on the preprocessed scanned images, a pre-trained visual feature extraction network is used to extract low-level and high-level visual features;

[0026] Based on the visual features, the YOLO object detection model is used to identify text areas, tables and charts, and generate corresponding distinguishing coordinates and classification information.

[0027] Furthermore, the text embedding step includes:

[0028] Performing text preprocessing on the text region, including: word segmentation and removal of stop words;

[0029] A Transformer-based pre-trained language model is used to map the preprocessed text regions into a high-dimensional semantic space. Taking into account contextual information, the attention mechanism is used to capture long-term dependencies in the text to effectively model medical entities and relationships.

[0030] Furthermore, the layout encoding process includes:

[0031] Analyzing the layout information of the multimodal medical document based on the visual features and the text embedding results, including the typesetting relationship of the text area, the row and column structure of the table, and the correspondence between the chart and the text;

[0032] Encoding the spatial position information of the text region and the corresponding image region to represent the relative position of the text region in the image through position coding;

[0033] For tables, restore the row and column structure of the table based on its layout information, and convert the table content into structured data;

[0034] For charts, image recognition and text association technology are used to extract chart titles, labels, and the relationship between charts and text, and establish cross-modal references and semantic associations.

[0035] Furthermore, the step of obtaining an independent text block includes:

[0036] Based on the parsed multimodal medical documents, a specialized large-scale model in the medical field is used to identify the document's chapter structure and segment it into initial independent text blocks according to the chapter structure. In the case of articles without a clear structure, the individual independent text blocks are segmented by setting a token upper limit and a sliding overlapping window.

[0037] For said initial independent text block, assigning unique identifiers to tables and figures and inserting dynamic reference marks;

[0038] A bidirectional index for positioning is established between the tables and charts and the corresponding description texts to form a final independent text block.

[0039] Furthermore, the step of obtaining the verified structured extraction result includes:

[0040] Setting a medical-specific prompt template to determine entity types and relationship types, wherein the entity types include diseases, drugs, genes, and examination indicators, and the relationship types include treatments, contraindications, complications, and biomarkers;

[0041] Inputting the independent text block into the fine-tuned medical language model and combining it with the medical-specific prompt template to generate JSON data containing entities, relations, and confidence scores;

[0042] Filter out relations whose confidence scores are lower than the set relation confidence threshold to form a preliminary knowledge subgraph;

[0043] Based on the preliminary knowledge subgraph, an entity list is extracted, and reverse prompt words are constructed to mine missing entities and relationships;

[0044] For contradictory relationships of the same entity pair, arbitration is performed based on the confidence score, the number of supporting documents, and the latest guideline version to obtain the final knowledge subgraph.

[0045] Furthermore, the steps of constructing the standardized medical knowledge graph include:

[0046] defining a standardized intermediate representation and normalizing the knowledge subgraph;

[0047] Based on the standardized knowledge subgraph, the semantic similarity and edit distance algorithms are used to perform entity normalization to obtain a standardized medical knowledge graph.

[0048] Furthermore, the method further includes performing non-standard character cleaning and terminology standardization on the multimodal medical document, specifically including:

[0049] Use a medical-specific regular expression library to remove advertising characters and non-medical symbols, and retain key symbols;

[0050] Use the medical terminology database to build a synonym mapping table to achieve term abbreviation expansion and language alignment.

[0051] Furthermore, the method further includes dynamically updating and maintaining the standardized medical knowledge graph, specifically including:

[0052] Obtain new medical documents, compare the differences with the standardized medical knowledge graph, synchronize the differences to the standardized medical knowledge graph, and add timestamps to entities and relationships;

[0053] Regularly detect logical conflicts in the standardized medical knowledge graph and make corrections through manual review or based on majority evidence;

[0054] Delete obsolete relationships based on the latest literature and guidelines.

[0055] Compared with the prior art, the present invention has the following beneficial effects:

[0056] (1) The present invention automatically parses multimodal medical documents, accurately extracts medical entities and relationships through dynamic segmentation, and dynamically verifies and optimizes knowledge representation, thereby constructing a high-precision and scalable medical knowledge graph.

[0057] (2) In the dynamic double verification mechanism, the present invention performs double verification by setting confidence scores and constructing reverse prompt words, and also resolves conflicts for contradictory entity pairs. It can not only filter out incorrect relationships but also discover missing entities and relationships, thereby improving the comprehensiveness and accuracy of the knowledge graph.

[0058] (3) The present invention dynamically segments chapters according to the text structure, which can better preserve semantic relationships. For charts and tables, dynamic reference tags are inserted and a bidirectional index with the description text is constructed, so that relevant data can be quickly located during retrieval, thereby improving retrieval efficiency.

[0059] (4) The present invention adopts an incremental update mechanism to add new content in new documents to the knowledge graph, and uses conflict detection and correction operations to achieve rapid iteration of medical knowledge and real-time updating of the graph, so that the knowledge graph has the latest timeliness. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 Schematic diagram of the method of the present invention. DETAILED DESCRIPTION

[0061] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0062] This embodiment provides a method for constructing a professional knowledge graph in the medical field. Figure 1 As shown, the method includes the following steps:

[0063] 1. Deep analysis and structured conversion of multimodal medical documents

[0064] The multimodal data source in this embodiment (including but not limited to documents in the medical literature library (PubMed, CNKI); exporting desensitized electronic medical records (XML / JSON format) from the hospital information system (HIS); scanning paper documents to generate high-resolution images (≥300dpi) or user-defined data sets). Secondly, perform OCR and table restoration: use the multimodal model in the medical professional field to perform OCR recognition on the text area to ensure the professionalism of the extracted content, and convert non-Markdown formats such as scanned documents into Markdown format text. Finally, it is also necessary to associate charts: extract the chart title and adjacent descriptive text, and generate a bidirectional reference index. Specifically, this step includes:

[0065] 1) Semantic processing of scanned documents and PDFs:

[0066] Multimodal model architecture: A Transformer-based multimodal model integrates visual feature extraction, text embedding, and layout encoding to achieve text region detection, table structure restoration, and chart semantic binding in scanned images. The specific implementation process of the multimodal model for visual feature extraction, text embedding, and layout encoding is as follows:

[0067] ① Visual feature extraction

[0068] The multimodal model first processes the scanned image through a Transformer to extract visual features from the image. The specific process includes:

[0069] Image preprocessing: De-noise, enhance, and standardize scanned images to improve image quality and ensure text is clear and accurate.

[0070] Visual Feature Extraction: Use pre-trained visual feature extraction networks to extract low-level and high-level visual features from images. These features can capture spatial information and visual patterns in different regions such as text regions, charts, and tables.

[0071] Region Localization and Classification: Use the YOLO object detection algorithm to identify structured or unstructured data areas such as text areas, tables, and charts in images, generate specific region coordinates and classification information, and support subsequent layout analysis.

[0072] ②Text Embedding

[0073] After the multimodal model extracts the text region, it uses the natural language processing (NLP) model to perform semantic embedding on the text. The specific process is as follows:

[0074] Text preprocessing: Perform standardization operations such as tokenization and stop word removal on the text to prepare it for input into the model.

[0075] Semantic Embedding: Utilizes a pre-trained language model based on the Transformer architecture to map text into a high-dimensional semantic space, enabling text information to be effectively captured and represented in the form of vectors.

[0076] Contextual Modeling: During the text embedding process, the model is able to consider contextual information and capture long-term dependencies in the text through the self-attention mechanism, thereby ensuring effective modeling of complex medical entities and relationships in medical documents.

[0077] ③Layout coding

[0078] The layout encoding step is responsible for understanding the structural layout of the document, especially in the processing of content such as tables and charts. The specific execution process is as follows:

[0079] Layout Analysis: The multimodal model combines the visual features of the image with the text embedding results to analyze the layout information of the document, including the row and column structure of tables, the layout relationship of text blocks, and the correspondence between charts and text.

[0080] Positional encoding: When embedding text, the model encodes the spatial position information of text and image regions. Positional encoding (such as 2D coordinate embedding) is used to express the relative positions of text blocks and visual regions in the image. This helps the model better understand the spatial structure and information flow of the document.

[0081] Table and chart structure restoration: For documents containing tables, the model restores the table's row and column structure based on the table's layout information and converts the table content into structured data (such as Markdown format). For charts, the model uses image recognition and text association to extract chart titles, labels, and the relationship between the chart and other text in the document, establishing cross-modal references and semantic associations.

[0082] 2) Table and chart processing:

[0083] Table parsing: Detect table row and column boundaries, convert cell contents to Markdown table format, and preserve row and column metadata.

[0084] Chart association: Identify chart titles and adjacent text descriptions to establish cross-modal reference relationships.

[0085] 3) Non-standard character cleaning and terminology standardization:

[0086] Noise filtering: Use a medical-specific regular expression library to remove advertising characters and non-medical symbols, retaining key symbols (such as "≥3cm" and "HbA1c>7%").

[0087] Terminology normalization: Based on authoritative medical terminology databases (such as SNOMED CT, UMLS, and MeSH), a synonym mapping table is constructed to achieve abbreviation expansion (such as replacing "NSCLC" with "non-small cell lung cancer") and multilingual alignment (such as mapping "EGFR" to "epidermal growth factor receptor" in Chinese).

[0088] 2. Context-aware adaptive text segmentation strategy

[0089] 1) Dynamic segmentation based on document structure:

[0090] Chapter-level identification: Use professional large-scale models in the medical field (ClinicalBERT, BioBERT) to identify the document chapter structure (such as "1.1 Etiology") and divide it into independent text blocks according to semantic units, ensuring that each block focuses on a single topic.

[0091] 2) Mixed cutting mechanism:

[0092] Fixed window segmentation: For text without a clear structure (such as case descriptions), set 32768 tokens as the upper limit of a single block and fixed window segmentation to avoid model processing bottlenecks.

[0093] Sliding overlapping window: The step size is set to 25% of the window size (8192 tokens) to preserve the correlation between cross-block contexts through the overlapping area.

[0094] 3) Multimodal content reference management:

[0095] ●Identifier generation: Assign unique identifiers to tables and figures (such as "Table_1" and "Figure_2"), and insert dynamic reference marks in text blocks (such as "as shown in the following table" pointing to "Table_1").

[0096] Cross-modal association indexing: Build a bidirectional index between tables / charts and their description text, enabling rapid locating of relevant data during subsequent searches.

[0097] 3. Extraction of large language models in the medical field and dynamic double verification mechanism

[0098] 1) Structured extraction driven by prompt words:

[0099] Text block input: Input the text block into the fine-tuned medical language model to generate JSON data containing entities, relations, and confidence scores.

[0100] Template design: Define medical-specific prompt templates, clarify entity types (e.g., diseases, drugs, genes, examination indicators) and relationship types (e.g., treatments, contraindications, complications, biomarkers).

[0101] Confidence threshold filtering: Set a relationship confidence threshold (e.g., ≥0.85) to filter low-quality output (e.g., “Disease H and drug I may have a therapeutic relationship, confidence level 0.6”).

[0102] 2) Dynamic double verification mechanism:

[0103] First extraction: Generate a preliminary set of entity relationships, filter low-confidence data, and generate a preliminary knowledge subgraph.

[0104] Omission detection: The entity list from the initial extraction is fed into the model, and reverse prompts are constructed (e.g., “Entity list: [Disease O, Drug P, Gene Q], what possible relationships might be missing?”). This allows us to identify entities and relationships between entities that may have been missed after the initial extraction.

[0105] Conflict resolution: Conflicting relationships between pairs of entities (e.g., "drug L treats disease M" and "drug L contraindicates disease M") are adjudicated based on confidence scores, number of supporting references, and the latest guideline version.

[0106] 4. Standardized construction and optimization of knowledge graphs

[0107] 3) Alignment of intermediate representation with entity:

[0108] JSON definition: Standardized output format, including but not limited to fields such as entity ID, name, type, description, source document, etc., supporting heterogeneous data integration.

[0109] Entity normalization: Based on semantic similarity (such as cosine similarity) and edit distance algorithms, synonym merging is achieved (for example, "aspirin" and "acetylsalicylic acid" are mapped to the unique ID "DRUG_001").

[0110] 5. Dynamic update and maintenance of knowledge graph

[0111] 4) Incremental update mechanism:

[0112] Difference extraction: When a new document is imported, only the differences with the existing graph are processed.

[0113] ●Version management: Add timestamp attributes to entities and relationships (such as "Guide version: 2023 vs 2024") to support historical data tracing.

[0114] 5) Conflict detection and automatic correction:

[0115] Conflict Scanning: Regularly detect logical conflicts in the graph (e.g., a drug marked as both "recommended" and "contraindicated"), triggering manual review or automatic correction based on majority evidence.

[0116] Timeliness verification: Automatically eliminate outdated relationships (such as old treatment plans replaced by new ones) based on the latest literature and guidelines.

[0117] In summary, this embodiment provides a solution for building a professional knowledge graph for the medical field. It aims to construct a high-precision, scalable medical knowledge graph through automated parsing of multimodal medical documents, precise extraction of medical entities and relationships, and dynamic verification and optimization of knowledge representation. This solution covers key aspects such as unstructured data processing, optimization of large language models in the medical field, and dynamic updating of knowledge graphs. It addresses core issues such as the fragmentation of medical knowledge, the difficulty of integrating heterogeneous data, and insufficient reasoning capabilities for complex relationships, providing a core knowledge infrastructure for clinical decision support systems, medical research platforms, and intelligent diagnosis and treatment tools.

[0118] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0119] It will be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention may be implemented in various computer languages, for example, the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0120] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0121] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0122] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0123] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0124] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A method for constructing a professional knowledge graph in the medical field, characterized by: The following steps are involved: Obtain multimodal medical documents and perform in-depth analysis using a Transformer-based multimodal model to obtain the parsed multimodal medical documents; Dynamically segmenting the parsed multimodal medical document using a context-based adaptive text segmentation method to obtain independent text blocks; Based on the independent text blocks, structured extraction is performed and a dynamic double verification mechanism is used for verification to form a knowledge subgraph; Based on the knowledge subgraph, a standardized medical knowledge graph is further constructed.

2. A method for constructing a professional knowledge graph in the medical field according to claim 1, characterized in that: The step of obtaining the parsed multimodal medical document includes: Based on the multimodal medical document, an extraction network extracts visual features to obtain region positioning and classification; Based on the region positioning and classification, a natural language processing model is used to perform text embedding; Layout encoding is performed based on the visual features and text embedding results to obtain a parsed multimodal medical document.

3. A method for constructing a professional knowledge graph in the medical field according to claim 2, characterized in that: The steps of obtaining region positioning and classification include: Preprocessing the scanned image of the multimodal medical document, including image denoising, enhancement, and standardization; Based on the preprocessed scanned images, a pre-trained visual feature extraction network is used to extract low-level and high-level visual features; Based on the visual features, the YOLO object detection model is used to identify text areas, tables and charts, and generate corresponding distinguishing coordinates and classification information.

4. A method for constructing a professional knowledge graph in the medical field according to claim 3, characterized in that: The text embedding steps include: Performing text preprocessing on the text region, including: word segmentation and removal of stop words; A Transformer-based pre-trained language model is used to map the preprocessed text regions into a high-dimensional semantic space. Taking into account contextual information, the attention mechanism is used to capture long-term dependencies in the text to effectively model medical entities and relationships.

5. A method for constructing a professional knowledge graph in the medical field according to claim 4, characterized in that: The layout encoding process includes: Analyzing the layout information of the multimodal medical document based on the visual features and the text embedding results, including the typesetting relationship of the text area, the row and column structure of the table, and the correspondence between the chart and the text; Encoding the spatial position information of the text region and the corresponding image region to represent the relative position of the text region in the image through position coding; For tables, restore the row and column structure of the table based on its layout information, and convert the table content into structured data; For charts, image recognition and text association technology are used to extract chart titles, labels, and the relationship between charts and text, and establish cross-modal references and semantic associations.

6. The method for constructing a professional knowledge graph in the medical field according to claim 1, characterized in that: The step of obtaining an independent text block includes: Based on the parsed multimodal medical documents, a specialized large-scale model in the medical field is used to identify the document's chapter structure and segment it into initial independent text blocks according to the chapter structure. In the case of articles without a clear structure, the individual independent text blocks are segmented by setting a token upper limit and a sliding overlapping window. For said initial independent text block, assigning unique identifiers to tables and figures and inserting dynamic reference marks; A bidirectional index for positioning is established between the tables and charts and the corresponding description texts to form a final independent text block.

7. The method for constructing a professional knowledge graph in the medical field according to claim 1, characterized in that: The step of obtaining the verified structured extraction result includes: Setting a medical-specific prompt template to determine entity types and relationship types, wherein the entity types include diseases, drugs, genes, and examination indicators, and the relationship types include treatments, contraindications, complications, and biomarkers; Inputting the independent text block into the fine-tuned medical language model and combining it with the medical-specific prompt template to generate JSON data containing entities, relations, and confidence scores; Filter out relations whose confidence scores are lower than the set relation confidence threshold to form a preliminary knowledge subgraph; Based on the preliminary knowledge subgraph, an entity list is extracted, and reverse prompt words are constructed to mine missing entities and relationships; For contradictory relationships of the same entity pair, arbitration is performed based on the confidence score, the number of supporting documents, and the latest guideline version to obtain the final knowledge subgraph.

8. The method for constructing a professional knowledge graph in the medical field according to claim 1, characterized in that: The steps for constructing the standardized medical knowledge graph include: defining a standardized intermediate representation and normalizing the knowledge subgraph; Based on the standardized knowledge subgraph, the semantic similarity and edit distance algorithms are used to perform entity normalization to obtain a standardized medical knowledge graph.

9. The method for constructing a professional knowledge graph in the medical field according to claim 1, characterized in that: The method also includes performing non-standard character cleaning and terminology standardization on the multimodal medical document, specifically including: Use a medical-specific regular expression library to remove advertising characters and non-medical symbols, and retain key symbols; Use the medical terminology database to build a synonym mapping table to achieve term abbreviation expansion and language alignment.

10. The method for constructing a professional knowledge graph in the medical field according to claim 1, characterized in that: The method also includes the steps of dynamically updating and maintaining the standardized medical knowledge graph, specifically including: Obtain new medical documents, compare the differences with the standardized medical knowledge graph, synchronize the differences to the standardized medical knowledge graph, and add timestamps to entities and relationships; Regularly detect logical conflicts in the standardized medical knowledge graph and make corrections through manual review or based on majority evidence; Delete obsolete relationships based on the latest literature and guidelines.

Images