Method and device for constructing medical knowledge graph based on hybrid intelligent technology

Abstract

The present invention provides a method and device for constructing a medical knowledge graph based on hybrid intelligent technology, which relates to the technical field of knowledge graph construction at the intersection of artificial intelligence and medicine. The method includes using a hybrid intelligent model to identify key entities in medical knowledge materials and extract semantic relationships, and automatically construct a knowledge graph; generating preliminary proofreading suggestions through an expert interactive proofreading mechanism driven by a large language model, and experts make adjustments and feedback in a visual interface, while recording the proofreading process and performing version management; accumulating expert proofreading information and re-importing it into a training set, updating the hybrid intelligent model, and looping the above steps to achieve human-machine collaboration to improve model performance and integrate expert experience into the knowledge graph. The present invention uses hybrid intelligent technology to integrate artificial intelligence and expert collaboration, efficiently construct a high-precision medical knowledge graph, support dynamic updates and reduce data dependence, and improve interpretability and cross-domain applicability.

Description

Technical Field

[0001] The present invention relates to the technical field of knowledge graph construction at the intersection of artificial intelligence and medicine, and in particular to a method and device for constructing a medical knowledge graph based on hybrid intelligence technology. Background Art

[0002] Medical knowledge graphs, as an important method for expressing medical knowledge in a structured form, have been widely used in the medical field, supporting semantic query, reasoning, decision support, and intelligent question and answer. With the rapid development of artificial intelligence technology, the construction methods of medical knowledge graphs have also evolved, from traditional manual and rule-driven methods to automated and intelligent ones. This has improved construction efficiency to a certain extent, but many problems still exist.

[0003] The application of large-scale data-based machine learning, natural language processing, and deep learning models enables medical knowledge graphs to automatically extract information from unstructured and semi-structured data and construct entity relationships and semantic networks. However, due to the complex sources and inconsistent standards of medical data, these approaches face significant challenges in terms of data integrity, diversity, and accuracy. Ensuring the high credibility and applicability of constructed knowledge graphs is a pressing issue in the construction of medical knowledge graphs.

[0004] Currently, all types of single technical solutions have significant defects in the construction of medical knowledge graphs:

[0005] 1. Rule-based approaches: These methods extract entities and relationships from structured and unstructured data using predefined rules and templates. While relatively simple to implement and highly interpretable, they suffer from numerous drawbacks. Firstly, they require the manual definition of numerous rules, making them difficult to adapt to the rapid dynamics and diverse needs of the medical field. Secondly, rule matching is inefficient when dealing with large amounts of data, and their processing capabilities for unstructured data, such as clinical documents and research papers, are significantly insufficient.

[0006] 2. Machine learning-based methods: These methods utilize labeled data to train models to extract entities and relationships from text data, offering high automation and adaptability. However, the high cost of acquiring labeled data in the medical field severely limits the widespread application of these methods. Furthermore, these models suffer from black-box nature, lacking transparency, making it difficult to understand and explain the decision-making process. Furthermore, their semantic understanding capabilities are limited, making it difficult to comprehend the complex semantics of medical texts, particularly cross-disciplinary or specialized terminology.

[0007] 3. Methods based on deep learning models (such as BERT and GPT): These are widely used in knowledge graph construction and offer strong semantic understanding capabilities and excellent performance in large-scale data processing. However, these methods are extremely expensive to train, requiring extensive computing resources and taking a long time to train. Furthermore, the models are highly sensitive to noise in the input data, making them prone to propagating misinformation. Furthermore, they lack domain knowledge supervision, and their output may not align with actual medical knowledge.

[0008] Overall, existing single technology solutions lay the foundation for building medical knowledge graphs, but their respective shortcomings limit their effectiveness and widespread adoption in practical applications. Existing technologies lack effective mechanisms for human-machine collaboration, neglect the critical role of human experts in verifying and supplementing medical knowledge, and struggle to meet the strict requirements of the medical field for high-precision data. Furthermore, knowledge graphs, once constructed, require continuous updating to reflect the latest research findings in the medical field, but existing technologies lack effective means for automated updates and dynamic maintenance. Summary of the Invention

[0009] To this end, an embodiment of the present invention provides a method and device for constructing a medical knowledge graph based on hybrid intelligent technology, which is used to solve the problems of data integrity and accuracy existing in the construction of medical knowledge graphs using a single technology in the existing technology, as well as the lack of human-computer collaboration mechanism, difficulty in meeting high-precision requirements and insufficient dynamic update capabilities.

[0010] To solve the above problems, an embodiment of the present invention provides a method for constructing a medical knowledge graph based on hybrid intelligent technology, which includes:

[0011] A hybrid intelligent model is used to identify medical entities, extract relationships, and build a knowledge graph. The hybrid intelligent model includes a large language model, a multimodal model, a time series model, and a large image model. It identifies key knowledge entities in medical knowledge materials and extracts the semantic relationships between them to automatically build a knowledge graph.

[0012] Through an expert interactive proofreading mechanism driven by a large language model, the constructed knowledge graph is proofread and optimized. The large language model generates preliminary proofreading suggestions, and experts make interactive adjustments and provide feedback through a visual proofreading interface. At the same time, the proofreading process is recorded and versioned.

[0013] The graph information proofread by experts is accumulated and re-imported into the training set, the hybrid intelligent model is updated, and the above steps are repeated. The performance of the hybrid intelligent model is continuously improved through human-computer collaboration, and the expert experience knowledge is absorbed into the knowledge graph.

[0014] Preferably, the use of a hybrid intelligent model for medical entity recognition and relationship extraction includes:

[0015] Combining large language models with multimodal models, key medical entities, including disease names, medication information, onset sites, and abnormality descriptions, can be extracted from text and imaging data.

[0016] A mechanism based on the combination of time series models and text data is designed to achieve the entity extraction and alignment of key medical knowledge concepts, and to combine the descriptive relationships between medical entities mined from text to achieve the alignment of medical entities and the association of their related relationships.

[0017] Preferably, the large language model generates preliminary proofreading suggestions, including:

[0018] The large language model provides preliminary proofreading suggestions for each entity and relationship based on semantic analysis and relational reasoning, and marks low-confidence areas;

[0019] The large language model outputs the logical basis for each proofreading suggestion, including the text source, reasoning path, and referenced knowledge base, providing decision-making support information for experts.

[0020] Preferably, the expert performs interactive adjustments and feedback through a visual proofreading interface, including:

[0021] Experts can quickly understand the recommendations generated by the model and the basis behind them through a visual proofreading interface that displays nodes and relationships in a graphical knowledge graph.

[0022] Experts directly adjust graph nodes and relationships, including adding, modifying, or deleting entities and relationships, and provide feedback on each adjustment operation so that the hybrid intelligent model can absorb expert experience in subsequent training.

[0023] Preferably, the recording and proofreading process and performing version management include:

[0024] Keep detailed records of every step and decision of the proofreading process, including the operation time, the knowledge entities or relationships involved, the revisions made by the experts and their basis;

[0025] After each proofreading is completed, a new version of the knowledge graph is generated and compared with the previous version, and the change points are marked to facilitate subsequent tracking and review.

[0026] Preferably, the accumulation of the atlas information verified and confirmed by the experts and re-importing it into the training set to update the hybrid intelligent model includes:

[0027] The expert-proofread graph information accumulated during the proofreading process is stored in the knowledge base as important data for further training the hybrid intelligent model;

[0028] As feedback information accumulates, the hybrid intelligent model gradually optimizes its entity recognition and relationship extraction capabilities, reducing errors in subsequent construction and improving efficiency.

[0029] An embodiment of the present invention further provides a device for constructing a medical knowledge graph based on hybrid intelligence technology, which is used to implement the above-mentioned method for constructing a medical knowledge graph based on hybrid intelligence technology, specifically comprising:

[0030] A hybrid intelligent model construction module is used to use a hybrid intelligent model to perform medical entity recognition and relationship extraction, as well as knowledge graph construction. The hybrid intelligent model includes a large language model, a multimodal model, a time series model, and a large image model. It identifies key knowledge entities in medical knowledge materials and extracts the semantic relationships between them to achieve automatic construction of the knowledge graph.

[0031] The expert interactive proofreading module is used to proofread and optimize the constructed knowledge graph through an expert interactive proofreading mechanism driven by a large language model. The large language model generates preliminary proofreading suggestions, and experts make interactive adjustments and provide feedback through a visual proofreading interface. At the same time, the proofreading process is recorded and versioned.

[0032] The model update module is used to accumulate the graph information proofread by experts and re-import it into the training set, update the hybrid intelligent model, repeat the above steps in a loop, continuously improve the performance of the hybrid intelligent model through human-computer collaboration, and absorb expert experience knowledge into the knowledge graph.

[0033] Preferably, the hybrid intelligent model building module includes:

[0034] The entity recognition and relationship extraction unit is used to combine large language models with multimodal models to extract key medical entities from text and imaging data. It also designs a mechanism based on the combination of time series models and text data to achieve the entity extraction and alignment of key medical knowledge concepts.

[0035] The knowledge graph construction unit is used to generate a structured knowledge graph after semantic classification and standardization of the extracted entities and relationships.

[0036] Preferably, the expert interactive proofreading module includes:

[0037] The proofreading suggestion generation unit is used by the large language model to provide preliminary proofreading suggestions for each entity and relationship based on semantic analysis and relational reasoning, mark low-confidence areas, and output the logical basis for each proofreading suggestion;

[0038] A visual proofreading interface unit is used to provide experts with a graphical representation of knowledge graph nodes and relationships, enabling them to quickly understand the recommendations generated by the model and the basis behind them;

[0039] An interactive adjustment and feedback unit, which allows experts to directly adjust graph nodes and relationships and provide feedback on each adjustment operation;

[0040] The proofreading record and version management unit is used to record every step and decision of the proofreading in detail, and after each proofreading is completed, a new version of the knowledge graph is generated, compared with the previous version, and the change points are marked.

[0041] Preferably, the model updating module includes:

[0042] The knowledge accumulation unit is used to store the atlas information proofread by experts accumulated during the proofreading process;

[0043] The model optimization unit is used to gradually optimize the entity recognition and relationship extraction capabilities of the hybrid intelligent model as feedback information accumulates, reduce errors in subsequent construction, and improve efficiency.

[0044] It can be seen from the above technical solutions that the present invention has the following beneficial effects:

[0045] (1) This invention combines multiple artificial intelligence models and expert knowledge through hybrid intelligent technology to achieve automated information extraction and structured processing, significantly reducing manual operation time, while reducing dependence on large-scale labeled data, reducing data collection and labeling costs, and improving the scalability of the technology.

[0046] (2) The present invention introduces an expert feedback mechanism to optimize the model prediction results, reduce the impact of data noise and erroneous information, ensure the high consistency between the knowledge graph and medical field knowledge, and improve the credibility of the knowledge graph.

[0047] (3) This invention combines the semantic understanding ability of deep learning models with the domain knowledge of experts to solve the problem of insufficient understanding of complex medical terminology and cross-domain knowledge, and supports the dynamic update and real-time maintenance of knowledge graphs, so that it can reflect the latest research progress in the medical field in real time and meet the ever-changing actual needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the implementation cases of the present invention or the technical solutions in the prior art, the following is a brief description of the drawings required for use in the embodiments. By referring to the drawings, the features and advantages of the present invention will be more clearly understood. The drawings are schematic and should not be understood as limiting the present invention in any way. Those skilled in the art can derive other drawings based on these drawings without inventive effort. Among them:

[0049] Figure 1A flowchart of a method for constructing a medical knowledge graph based on hybrid intelligent technology provided in an embodiment;

[0050] Figure 2 This is a block diagram of a medical knowledge graph construction device based on hybrid intelligent technology provided in an embodiment. DETAILED DESCRIPTION

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only 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 ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention. Example 1

[0052] In order to solve the problems of data integrity and accuracy in the existing single technology for building medical knowledge graphs, as well as the lack of human-computer collaboration mechanism, difficulty in meeting high-precision requirements and insufficient dynamic update capabilities. Figure 1 As shown, an embodiment of the present invention proposes a method for constructing a medical knowledge graph based on hybrid intelligent technology, the method comprising:

[0053] Use hybrid intelligent models for medical entity recognition and relationship extraction, as well as knowledge graph construction. The hybrid intelligent model includes a large language model, a multimodal model, a time series model, and a large image model. It identifies key knowledge entities in medical knowledge materials and extracts the semantic relationships between them, realizing the automatic construction of the knowledge graph.

[0054] Through an expert interactive proofreading mechanism driven by a large language model, the constructed knowledge graph is proofread and optimized. The large language model generates preliminary proofreading suggestions, and experts make interactive adjustments and provide feedback through a visual proofreading interface. At the same time, the proofreading process is recorded and versioned.

[0055] Accumulate the graph information proofread by experts and re-import it into the training set, update the hybrid intelligent model, and repeat the above steps repeatedly. Through human-computer collaboration, the performance of the hybrid intelligent model is continuously improved, and the expert experience and knowledge are absorbed into the knowledge graph.

[0056] As can be seen from the above technical solution, the present invention proposes a method for constructing a medical knowledge graph based on hybrid intelligence technology. This method utilizes a hybrid intelligent model to identify medical entities and extract relationships, thereby constructing a knowledge graph. This step, through the collaborative work of multiple AI models, efficiently identifies key entities and extracts semantic relationships from medical knowledge materials, enabling automated knowledge graph construction, significantly improving construction efficiency and reducing the time consumed by manual operations in traditional methods. The constructed knowledge graph is proofread and optimized through an expert interactive proofreading mechanism driven by a large language model. The large language model generates preliminary proofreading suggestions, and experts interactively adjust and provide feedback through a visual proofreading interface. The proofreading process is also recorded and versioned. This step incorporates human expert participation to ensure high consistency between model outputs and medical domain knowledge, optimize model predictions, reduce errors caused by data noise and erroneous information, and ensure high credibility of the knowledge graph. The expert-proofread graph information is accumulated and re-imported into the training set, and the hybrid intelligent model is updated. This process is repeated repeatedly, continuously improving the performance of the hybrid intelligent model through human-machine collaboration and incorporating expert experience into the knowledge graph. This step uses a dynamic update mechanism to enable the knowledge graph to reflect the latest research progress in the medical field in real time, meet the ever-changing actual needs, and at the same time reduce the dependence on large-scale annotated data, reduce data collection and annotation costs, and improve the scalability of the technology. In summary, the present invention effectively overcomes many shortcomings in the existing technology through the innovative application of hybrid intelligent technology, and provides a new technical solution for the efficient, accurate and dynamic construction of medical knowledge graphs, which has significant technological advancement and application value.

[0057] In this embodiment, the present invention utilizes a hybrid intelligent model to perform medical entity recognition and relationship extraction, as well as knowledge graph construction, wherein the hybrid intelligent model includes a large language model, a multimodal model, a time series model, and a large image model, which identifies key knowledge entities in medical knowledge materials and extracts the semantic relationships therebetween, thereby realizing the automatic construction of a knowledge graph.

[0058] Specifically, data preparation begins with collecting medical text data (such as medical literature and clinical records) and imaging data (such as medical images and waveforms). Preprocessing of text data involves operations such as word segmentation, part-of-speech tagging, and stop word removal; and preprocessing of imaging data involves operations such as image cropping, normalization, and enhancement.

[0059] Further model construction is carried out: a hybrid intelligent model is constructed, including a large language model, a multimodal model, a time series model, and a large imaging model. The large language model is used to extract key medical entities and their semantic relationships from text data; the multimodal model is used to extract key medical entities and their features from imaging data; the time series model is used to analyze the temporal variation characteristics of medical data, such as the progression of diseases and changes in drug efficacy; and the large imaging model is used to extract the visual features of medical entities from imaging data.

[0060] Furthermore, a hybrid intelligent model is used for medical entity recognition and relationship extraction, specifically including:

[0061] Combining large language models with multimodal models, key medical entities, including disease names, medication information, onset sites, and abnormality descriptions, can be extracted from text and imaging data.

[0062] A mechanism based on the combination of time series models and text data is designed to achieve the entity extraction and alignment of key medical knowledge concepts, and to combine the descriptive relationships between medical entities mined from text to achieve the alignment of medical entities and the association of their related relationships.

[0063] Specifically, by combining a large language model with a multimodal model, key medical entities are extracted from preprocessed text and image data. The large language model uses semantic analysis and contextual understanding to identify key medical entities in the text, such as disease names, medication information, onset sites, and abnormality descriptions, and extracts semantic relationships between entities, such as causal and associative relationships. The multimodal model uses feature extraction and fusion to identify key medical entities in the images, such as lesion areas and organ structures, and aligns them with entities in the text. The time series model analyzes the temporal variation characteristics of medical data, providing temporal information support for entity recognition and relationship extraction. The large image model extracts visual features of medical entities in the image data, providing visual information support for entity recognition and relationship extraction. Ultimately, the medical entities and their relationships extracted from the text and images are integrated to form a preliminary medical knowledge graph structure.

[0064] Furthermore, semantic classification and standardization are performed: the extracted medical entities and relationships are semantically classified and categorized into predefined medical knowledge systems, such as disease classification and drug classification. At the same time, the entities and relationships are standardized, with unified naming conventions and formats to ensure the consistency and scalability of the knowledge graph.

[0065] Knowledge graph generation: Organize semantically classified and standardized medical entities and relationships into a structured knowledge graph. The knowledge graph represents medical entities as nodes and semantic relationships between entities as edges. Using technologies such as graph databases to store and manage the knowledge graph, efficient query and reasoning operations are supported.

[0066] In this embodiment, the constructed knowledge graph is proofread and optimized through an expert interactive proofreading mechanism driven by a large language model. The large language model generates preliminary proofreading suggestions, and experts make interactive adjustments and provide feedback through a visual proofreading interface. At the same time, the proofreading process is recorded and version management is performed.

[0067] Specifically, the large language model generates preliminary proofreading suggestions, including: Based on semantic analysis and relational reasoning, the large language model performs preliminary proofreading on each entity and relationship in the constructed knowledge graph and marks low-confidence areas. The model judges the accuracy of entities and relationships based on existing knowledge and logical rules, and generates preliminary proofreading suggestions for parts that may contain errors or uncertainties. For example, when the model recognizes that the name of an entity is inconsistent with a known standard name, it will generate modification suggestions; when the model finds that there are doubts about the reasoning path of a certain relationship, it will propose possible correction directions.

[0068] Furthermore, the logical justification output includes: The large language model simultaneously outputs the logical justification for each proofreading suggestion, including the text source, reasoning path, and referenced knowledge base. The logical justification is presented in clear text form, providing experts with detailed decision-making support information. For example, for a suggestion to modify an entity name, the model will explain the information in the text data that the suggestion is based on and how the conclusion was reached through semantic analysis. For a suggestion to modify a relationship, the model will display the knowledge rules and data evidence used in the reasoning process.

[0069] The present invention also includes interface design: an intuitive visual proofreading interface is designed to graphically display the nodes and relationships of the knowledge graph. The proofreading interface adopts an interactive design, allowing experts to easily browse and operate the knowledge graph. The interface includes an area displaying detailed information about entity nodes, a visual display area for relationships, a prompt area for proofreading suggestions, and an interactive area for expert operation.

[0070] Information Display and Understanding Assistance: In the visual proofreading interface, different entity and relationship types are distinguished by different colors, shapes, or icons, allowing experts to quickly identify and understand the structure and content of the knowledge graph. Furthermore, the interface provides features such as text highlighting and pop-up prompts for related information, helping experts better understand the proofreading suggestions generated by the model and the underlying rationale.

[0071] Furthermore, experts make interactive adjustments and provide feedback through a visual proofreading interface, including:

[0072] Experts can directly adjust nodes and relationships within the knowledge graph through a visual proofreading interface. Experts can modify entity names and attributes, adjust relationship types and directions, and add or delete entities and relationships. For example, if an expert discovers an error in the translation of a disease name, they can modify it directly within the interface. If an expert believes the association between a drug and a disease is inaccurate, they can adjust the relationship type or delete it.

[0073] Experts provide detailed feedback on each adjustment, explaining the reasoning and rationale behind the change. Feedback is entered through interactive features within the interface and stored in association with the corresponding adjustment. For example, when modifying an entity name, experts provide a text explaining the correct origin and standard specifications for the name; when adjusting a relationship, experts explain the medical rationale and reasoning behind the adjustment.

[0074] Furthermore, the proofreading process is recorded and versioned, including:

[0075] Detailed records are kept of every proofreading action and decision made by the expert, including the time of the action, the knowledge entities or relationships involved, the revisions made by the expert, and the rationale behind them. These records are stored as structured data to facilitate subsequent query and analysis. For example, the records will detail when the expert modified which entity, the specific content before and after the modification, and the rationale behind the modification.

[0076] After each proofreading session, a new version of the knowledge graph is generated and compared with the previous version, noting changes. The new version of the knowledge graph is stored in the graph database and associated with the previous version to facilitate subsequent tracking and review. For example, the version comparison function clearly shows the differences between the new and previous versions, including newly added entities and relationships, modified entities and relationships, and deleted entities and relationships.

[0077] In this embodiment, the graph information proofread by experts is accumulated and re-imported into the training set, the hybrid intelligent model is updated, and the above steps are repeated. The performance of the hybrid intelligent model is continuously improved through human-computer collaboration, and the expert experience knowledge is absorbed into the knowledge graph.

[0078] Specifically, the expert-verified graph information is accumulated and re-imported into the training set to update the hybrid intelligent model, including:

[0079] The expert confirmation information accumulated during the proofreading process is stored in the knowledge base. This information includes entity names, attributes, relationship types, semantic annotations, and other information that has been verified by the experts. The knowledge base uses a structured database format to categorize, store, and manage expert confirmation information for subsequent query and use.

[0080] The expert-confirmed information is consolidated and annotated, converting it into a data format suitable for model training. For example, the entity names and attributes confirmed by the experts are used as annotated data to train the entity recognition part of the large language model; the relationship types and semantic annotations confirmed by the experts are used as annotated data to train the relationship extraction part of the model.

[0081] As feedback accumulates, the hybrid intelligent model's entity recognition and relationship extraction capabilities are gradually optimized. Expert-verified information from the knowledge base is used as new training data to retrain the hybrid intelligent model. During training, techniques such as semi-supervised learning and transfer learning are employed to leverage expert-labeled data and the model's prior knowledge to improve model performance and accuracy. For example, semi-supervised learning can use a small amount of expert-labeled data to guide model learning while simultaneously leveraging a large amount of unlabeled data for self-learning, improving the model's ability to recognize medical entities and relationships.

[0082] The updated hybrid intelligent model is evaluated for performance, including metrics such as entity recognition accuracy, recall, relationship extraction accuracy, and F1 score. Based on the evaluation results, the model's parameters and structure are further adjusted for iterative optimization. For example, if the model is found to have deficiencies in certain types of medical entity recognition, the feature extraction component can be adjusted accordingly, increasing the weight of relevant features. If errors are found in the extraction of certain types of relationships, the relationship reasoning component can be optimized to improve the inference algorithm. Example 2

[0083] like Figure 2 As shown, the present invention provides a medical knowledge graph construction device based on hybrid intelligent technology, which is used to implement the medical knowledge graph construction method based on hybrid intelligent technology in the above embodiment 1, specifically comprising:

[0084] A hybrid intelligent model construction module is used to use a hybrid intelligent model for medical entity recognition and relationship extraction, as well as knowledge graph construction. The hybrid intelligent model includes a large language model, a multimodal model, a time series model, and a large image model. It identifies key knowledge entities in medical knowledge materials and extracts the semantic relationships between them, realizing the automatic construction of the knowledge graph.

[0085] The expert interactive proofreading module is used to proofread and optimize the constructed knowledge graph through an expert interactive proofreading mechanism driven by a large language model. The large language model generates preliminary proofreading suggestions, and experts make interactive adjustments and provide feedback through a visual proofreading interface. At the same time, the proofreading process is recorded and versioned.

[0086] The model update module is used to accumulate the graph information proofread by experts and re-import it into the training set, update the hybrid intelligent model, and repeat the above steps repeatedly to continuously improve the performance of the hybrid intelligent model through human-computer collaboration and absorb expert experience knowledge into the knowledge graph.

[0087] Furthermore, the hybrid intelligent model building module includes:

[0088] The entity recognition and relationship extraction unit is used to combine large language models with multimodal models to extract key medical entities from text and imaging data. It also designs a mechanism based on the combination of time series models and text data to achieve the entity extraction and alignment of key medical knowledge concepts.

[0089] The knowledge graph construction unit is used to generate a structured knowledge graph after semantic classification and standardization of the extracted entities and relationships.

[0090] Furthermore, the expert interactive proofreading module includes:

[0091] The proofreading suggestion generation unit is used by the large language model to provide preliminary proofreading suggestions for each entity and relationship based on semantic analysis and relational reasoning, mark low-confidence areas, and output the logical basis for each proofreading suggestion;

[0092] A visual proofreading interface unit is used to provide experts with a graphical representation of knowledge graph nodes and relationships, enabling them to quickly understand the recommendations generated by the model and the basis behind them;

[0093] An interactive adjustment and feedback unit, which allows experts to directly adjust graph nodes and relationships and provide feedback on each adjustment operation;

[0094] The proofreading record and version management unit is used to record every step and decision of the proofreading in detail, and after each proofreading is completed, a new version of the knowledge graph is generated, compared with the previous version, and the change points are marked.

[0095] Furthermore, the model updating module includes:

[0096] The knowledge accumulation unit is used to store the atlas information proofread by experts accumulated during the proofreading process;

[0097] The model optimization unit is used to gradually optimize the entity recognition and relationship extraction capabilities of the hybrid intelligent model as feedback information accumulates, reduce errors in subsequent construction, and improve efficiency.

[0098] A medical knowledge graph construction device based on hybrid intelligent technology in this embodiment is used to implement the aforementioned medical knowledge graph construction method based on hybrid intelligent technology. Therefore, the specific implementation method of the medical knowledge graph construction device based on hybrid intelligent technology can be found in the embodiment part of the medical knowledge graph construction method based on hybrid intelligent technology in the previous text. In order to avoid redundancy, it will not be repeated here.

[0099] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0100] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, 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 flowchart and / or block diagram. 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.

[0101] 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 These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the process. Figure 1 a process or multiple processes and / or boxes Figure 1A step that specifies a function in one or more boxes.

[0102] Obviously, the above embodiments are merely examples for clarity of explanation and are not intended to limit the implementation methods. Those skilled in the art will appreciate that other variations or modifications can be made based on the above description. It is not necessary and impossible to enumerate all implementation methods here. Obvious variations or modifications derived therefrom remain within the scope of protection of the present invention.

Claims

1. A method for constructing a medical knowledge graph based on hybrid intelligent technology, characterized in that: include: A hybrid intelligent model is used to identify medical entities, extract relationships, and build a knowledge graph. The hybrid intelligent model includes a large language model, a multimodal model, a time series model, and a large image model. It identifies key knowledge entities in medical knowledge materials and extracts the semantic relationships between them to automatically build a knowledge graph. The method of using a hybrid intelligent model to perform medical entity recognition and relationship extraction includes: Combining large language models with multimodal models, key medical entities, including disease names, medication information, onset sites, and abnormality descriptions, can be extracted from text and imaging data. Design a mechanism based on combining time series models with text data to achieve entity extraction and alignment of key medical knowledge concepts. Combine the descriptive relationships between medical entities mined from text to achieve alignment of medical entities and association of their related relationships. The constructed knowledge graph is proofread and optimized through an expert interactive proofreading mechanism driven by a large language model. The large language model generates preliminary proofreading suggestions, and experts make interactive adjustments and provide feedback through a visual proofreading interface. Meanwhile, the proofreading process is recorded and versioned. The preliminary proofreading suggestions generated by the large language model include: The large language model provides preliminary proofreading suggestions for each entity and relationship based on semantic analysis and relational reasoning, and marks low-confidence areas; The large language model outputs the logical basis for each proofreading suggestion, including the text source, reasoning path, and referenced knowledge base, providing decision-making support information for experts; The record proofreading process and version management include: Keep detailed records of every step and decision of the proofreading process, including the operation time, the knowledge entities or relationships involved, the revisions made by the experts and their basis; After each proofreading is completed, a new version of the knowledge graph is generated and compared with the previous version, marking the change points; The graph information proofread by experts is accumulated and re-imported into the training set, the hybrid intelligent model is updated, and the above steps are repeated. The performance of the hybrid intelligent model is continuously improved through human-computer collaboration, and the expert experience knowledge is absorbed into the knowledge graph.

2. The method for constructing a medical knowledge graph based on hybrid intelligent technology according to claim 1, characterized in that: The experts make interactive adjustments and provide feedback through a visual proofreading interface, including: Experts can quickly understand the recommendations generated by the model and the basis behind them through a visual proofreading interface that displays nodes and relationships in a graphical knowledge graph. Experts directly adjust graph nodes and relationships, including adding, modifying, or deleting entities and relationships, and provide feedback on each adjustment operation so that the hybrid intelligent model can absorb expert experience in subsequent training.

3. The method for constructing a medical knowledge graph based on hybrid intelligent technology according to claim 1, characterized in that: The accumulation of expert-verified atlas information and re-importing it into the training set to update the hybrid intelligent model includes: The expert-proofread graph information accumulated during the proofreading process is stored in the knowledge base as important data for further training the hybrid intelligent model; As feedback information accumulates, the hybrid intelligent model gradually optimizes its entity recognition and relationship extraction capabilities, reducing errors in subsequent construction and improving efficiency.

4. A medical knowledge graph construction device based on hybrid intelligent technology, characterized in that: The device is used to implement the medical knowledge graph construction method based on hybrid intelligence technology according to any one of claims 1 to 3, specifically comprising: A hybrid intelligent model construction module is used to use a hybrid intelligent model to perform medical entity recognition and relationship extraction, as well as knowledge graph construction. The hybrid intelligent model includes a large language model, a multimodal model, a time series model, and a large image model. It identifies key knowledge entities in medical knowledge materials and extracts the semantic relationships between them to achieve automatic construction of the knowledge graph. The expert interactive proofreading module is used to proofread and optimize the constructed knowledge graph through an expert interactive proofreading mechanism driven by a large language model. The large language model generates preliminary proofreading suggestions, and experts make interactive adjustments and provide feedback through a visual proofreading interface. At the same time, the proofreading process is recorded and versioned. The model update module is used to accumulate the graph information proofread by experts and re-import it into the training set, update the hybrid intelligent model, repeat the above steps in a loop, continuously improve the performance of the hybrid intelligent model through human-computer collaboration, and absorb expert experience knowledge into the knowledge graph.

5. The medical knowledge graph construction device based on hybrid intelligent technology according to claim 4 is characterized in that: The hybrid intelligent model building module includes: The entity recognition and relationship extraction unit is used to combine large language models with multimodal models to extract key medical entities from text and imaging data. It also designs a mechanism based on the combination of time series models and text data to achieve the entity extraction and alignment of key medical knowledge concepts. The knowledge graph construction unit is used to generate a structured knowledge graph after semantic classification and standardization of the extracted entities and relationships.

6. The medical knowledge graph construction device based on hybrid intelligent technology according to claim 4 is characterized in that: The expert interactive proofreading module includes: The proofreading suggestion generation unit is used by the large language model to provide preliminary proofreading suggestions for each entity and relationship based on semantic analysis and relational reasoning, mark low-confidence areas, and output the logical basis for each proofreading suggestion; A visual proofreading interface unit is used to provide experts with a graphical representation of knowledge graph nodes and relationships, enabling them to quickly understand the recommendations generated by the model and the basis behind them; An interactive adjustment and feedback unit, which allows experts to directly adjust graph nodes and relationships and provide feedback on each adjustment operation; The proofreading record and version management unit is used to record every step and decision of the proofreading in detail, and after each proofreading is completed, a new version of the knowledge graph is generated, compared with the previous version, and the change points are marked.

7. The medical knowledge graph construction device based on hybrid intelligent technology according to claim 4 is characterized in that: The model updating module includes: The knowledge accumulation unit is used to store the atlas information proofread by experts accumulated during the proofreading process; The model optimization unit is used to gradually optimize the entity recognition and relationship extraction capabilities of the hybrid intelligent model as feedback information accumulates, reduce errors in subsequent construction, and improve efficiency.

Images