AI medical information accurate retrieval method based on knowledge graph

Abstract

The invention discloses an AI medical information accurate retrieval method based on a knowledge graph, and belongs to the field of medical information, and the method comprises the steps: collecting and integrating multi-source medical data, employing an entity recognition and relation extraction technology, accurately extracting medical entities and semantic relations, and constructing a complete and accurate medical knowledge graph; according to the method, the medical knowledge graph is constructed, query is converted into a query sub-graph structure in combination with a natural language processing technology, and the FFGR algorithm is applied to fuse attributes, structures and semantic features of nodes to calculate a comprehensive score, so that the correlation between the nodes and the query can be accurately evaluated; the potential relationship can be deeply mined when the complex disease query is processed; the DKU algorithm utilizes a deep learning model to carry out feature extraction and relation prediction on new data, and the structure and content of the knowledge graph are updated in real time. And new diseases, medicines and relationships can be timely reflected in the knowledge graph, so that the retrieval result is always based on the latest medical knowledge.

Description

Technical Field

[0001] The present invention relates to the field of medical information technology, and in particular to an AI medical information precision retrieval method based on knowledge graphs. Background Art

[0002] Medical information retrieval currently faces many difficulties. Traditional retrieval methods rely on keyword matching and are unable to understand the complex semantic associations of medical terms. For example, when searching for "eye complications caused by diabetes and treatment drugs," it is difficult to accurately locate relevant information. Although existing knowledge graph-based retrieval algorithms have made some progress, they have reasoning limitations, are insufficient in mining complex relationships, and lack adaptability to dynamic knowledge, making them unable to meet the growing demand for accurate information acquisition in the medical field.

[0003] Furthermore, Publication No. CN114898831A discloses a medical data retrieval system and method, which implements basic retrieval through identity authentication and data classification. However, it has the following core flaws:

[0004] Insufficient semantic reasoning capabilities: Relying on "type labels" for coarse-grained matching, it is unable to handle complex comorbidity relationships such as "hypertension combined with diabetes" and polysemous terms such as "ibuprofen", and the accuracy rate of complex queries is less than 50%.

[0005] Inefficient multi-hop reasoning: Traditional algorithms (such as PathRank) use fixed rules or random walks, which severely attenuate semantic information in long paths (>3 hops). The reasoning accuracy of more than 4 hops is only 38%.

[0006] Lack of personalized services: Search results are not dynamically optimized based on user roles (doctor / patient), doctors are inefficient in obtaining clinical guidelines, and patients find it difficult to obtain popular medication guidance.

[0007] Weak rule generalization ability: Inference rules need to be manually defined, and cannot adapt to the dynamic updates of semantic relationships of new diseases (such as COVID-19) or new drugs (such as new hypoglycemic drugs).

[0008] To address the above problems, the present invention proposes a dynamic adaptive retrieval method based on knowledge graph, which significantly improves the accuracy and efficiency of medical information retrieval by enhancing semantic matching through hierarchical attention, driving multi-hop reasoning through reinforcement learning, and achieving rule generalization through meta-learning. Summary of the Invention

[0009] In response to the shortcomings of the existing technology, the present invention provides an AI medical information precision retrieval method based on knowledge graph, which solves the problems raised in the above background technology.

[0010] Technical Solution: To solve the above technical problems, according to one aspect of the present invention, more specifically, an AI medical information precision retrieval method based on knowledge graphs, comprising the following steps:

[0011] S1. Knowledge Graph Construction and Data Integration: Collect and integrate multi-source medical data (electronic medical records, medical literature, clinical guidelines, etc.), apply entity recognition and relationship extraction techniques to accurately extract medical entities (such as diseases, drugs, symptoms, examination items, etc.) and semantic relationships (such as "treatment," "initiate," "accompany," etc.), and build a complete and accurate medical knowledge graph;

[0012] And use a distributed graph database to store knowledge graphs to ensure efficient management and fast access to data;

[0013] S2. Query Semantic Understanding and Conversion: Leveraging natural language processing technology, we conduct in-depth analysis of user-entered queries. Through word segmentation, part-of-speech tagging, and named entity recognition, we accurately grasp the medical entities and semantic relationships within the query.

[0014] Convert natural language queries into query subgraph structures in the knowledge graph, aligning the query with the structure of the knowledge graph and providing support for accurate retrieval;

[0015] S3, Inference Calculation:

[0016] a. Feature-Fused Graph Reasoning (FFGR): This algorithm combines the attribute features, structural features, and semantic features related to the query to calculate a comprehensive score for each node, thereby accurately assessing the relevance of the node to the query.

[0017] The core formula of the FFGR algorithm is:

[0018] S(n)=α·F a (n)+β·F s (n)+γ·F q (n)

[0019] Among them, S(n) represents the comprehensive score of node n, F a (n) represents the attribute feature score of node n, F s (n) is the structural feature score of node n, F q (n) is the semantic feature score of node n related to the query, and α, β, and γ are weight coefficients optimized by machine learning methods (which can adapt to different query scenarios and knowledge graph structures; it breaks through the limitations of existing algorithms that rely only on single features or simple relationship calculations, and significantly improves retrieval accuracy. For example, when processing complex disease queries, existing algorithms may miss important related information, while this algorithm can deeply explore potential relationships, making the retrieval results more comprehensive and accurate);

[0020] b. Dynamic Knowledge Update Algorithm (DKU Algorithm): The Dynamic Knowledge Update Algorithm (DKU Algorithm) is used to update the knowledge graph in real time (such as adding new diseases, drugs, or relationships). The deep learning model is used to extract features and predict relationships on new data, and the structure and content of the knowledge graph are updated in a timely manner. The core formula of the DKU algorithm is:

[0021] ΔG=f(N,E,G old )

[0022] Among them, ΔG represents the update amount of the knowledge graph, N represents the entity set in the new data, E is the relationship set in the new data, G old The function f is implemented using a deep learning model to update the knowledge graph before the update. This model can automatically learn how to integrate new data with the original knowledge graph. Compared with traditional knowledge graph update methods, the dynamic knowledge update algorithm does not require extensive manual intervention and can achieve dynamic and real-time updates of the knowledge graph, ensuring that search results are always based on the latest medical knowledge.

[0023] S4. Search result optimization and personalized recommendation: Based on the node comprehensive score calculated by the FFGR algorithm, the search results are sorted and filtered, and information highly relevant to the query is displayed first;

[0024] Combine the user's historical query records, browsing behavior, disease concerns and other data to build an accurate user profile and achieve personalized search result recommendations;

[0025] For doctor users, priority is given to pushing the latest clinical research results, diagnosis and treatment guidelines and case analyses; for patient users, easy-to-understand disease explanations, symptom relief methods and medication recommendations are provided.

[0026] The beneficial effects of the AI ​​medical information precision retrieval method based on knowledge graph of the present invention are:

[0027] (1) This invention constructs a medical knowledge graph, combines natural language processing technology to transform queries into query subgraph structures, and uses the FFGR algorithm to fuse the attributes, structure, and semantic features of nodes to calculate a comprehensive score, which can accurately assess the relevance between nodes and queries. This allows for deep mining of potential relationships when processing complex disease queries, avoiding missing important related information. Compared with existing algorithms that rely on single features or simple relationship calculations, this method significantly improves retrieval accuracy.

[0028] (2) The DKU algorithm of the present invention uses a deep learning model to extract features and predict relationships from new data, updating the structure and content of the knowledge graph in real time. New diseases, drugs, and relationships are reflected in the knowledge graph in a timely manner, ensuring that search results are always based on the latest medical knowledge.

[0029] (3) The present invention can construct accurate user portraits based on the user's historical query records, browsing behavior, disease concerns and other data to achieve personalized search result recommendations. For doctor users, the latest clinical research results, diagnosis and treatment guidelines and case analysis are pushed first to help doctors quickly obtain valuable information; for patient users, easy-to-understand disease explanations, symptom relief methods and medication recommendations are provided to facilitate patient understanding and application. In the patient's self-health query scenario, the system presents the search results in simple and clear language based on the patient's historical query data to meet the patient's needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The present invention will be further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0031] Figure 1 It is a structural schematic diagram of the present invention. DETAILED DESCRIPTION

[0032] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that, unless there is a conflict, the embodiments and features in the embodiments of the present application can be combined with each other.

[0033] In order to make the technical solution of the present invention clearer, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0034] Reference Figure 1 , an AI medical information accurate retrieval method based on knowledge graph, characterized by comprising the following steps:

[0035] S1. Knowledge Graph Construction and Data Integration: Collect and integrate multi-source medical data (electronic medical records, medical literature, clinical guidelines, etc.), apply entity recognition and relationship extraction techniques to accurately extract medical entities (such as diseases, drugs, symptoms, examination items, etc.) and semantic relationships (such as "treatment," "initiate," "accompany," etc.), and build a complete and accurate medical knowledge graph;

[0036] And use a distributed graph database to store knowledge graphs to ensure efficient management and fast access to data;

[0037] S2. Query Semantic Understanding and Conversion: Leveraging natural language processing technology, we conduct in-depth analysis of user-entered queries. Through word segmentation, part-of-speech tagging, and named entity recognition, we accurately grasp the medical entities and semantic relationships within the query.

[0038] Convert natural language queries into query subgraph structures in the knowledge graph, aligning the query with the structure of the knowledge graph and providing support for accurate retrieval;

[0039] S3, Inference Calculation:

[0040] a. Feature-Fused Graph Reasoning (FFGR): This algorithm combines the attribute features, structural features, and semantic features related to the query to calculate a comprehensive score for each node, thereby accurately assessing the relevance of the node to the query.

[0041] The core formula of the FFGR algorithm is:

[0042] S(n)=α·F a (n)+β·F s (n)+γ·F q (n)

[0043] Among them, S(n) represents the comprehensive score of node n, F a (n) represents the attribute feature score of node n, F s (n) is the structural feature score of node n, F q (n) is the semantic feature score of node n related to the query, and α, β, and γ are weight coefficients optimized by machine learning methods (which can adapt to different query scenarios and knowledge graph structures; it breaks through the limitations of existing algorithms that rely only on single features or simple relationship calculations, and significantly improves retrieval accuracy. For example, when processing complex disease queries, existing algorithms may miss important related information, while this algorithm can deeply explore potential relationships, making the retrieval results more comprehensive and accurate);

[0044] b. Dynamic Knowledge Update Algorithm (DKU Algorithm): The Dynamic Knowledge Update Algorithm (DKU Algorithm) is used to update the knowledge graph in real time (such as adding new diseases, drugs, or relationships). The deep learning model is used to extract features and predict relationships on new data, and the structure and content of the knowledge graph are updated in a timely manner. The core formula of the DKU algorithm is:

[0045] ΔG=f(N,E,G old )

[0046] Among them, ΔG represents the update amount of the knowledge graph, N represents the entity set in the new data, E is the relationship set in the new data, G old The function f is implemented using a deep learning model to update the knowledge graph before the update. This model can automatically learn how to integrate new data with the original knowledge graph. Compared with traditional knowledge graph update methods, the dynamic knowledge update algorithm does not require extensive manual intervention and can achieve dynamic and real-time updates of the knowledge graph, ensuring that search results are always based on the latest medical knowledge.

[0047] S4. Search result optimization and personalized recommendation: Based on the node comprehensive score calculated by the FFGR algorithm, the search results are sorted and filtered, and information highly relevant to the query is displayed first;

[0048] Combine the user's historical query records, browsing behavior, disease concerns and other data to build an accurate user profile and achieve personalized search result recommendations;

[0049] For doctor users, priority is given to pushing the latest clinical research results, diagnosis and treatment guidelines and case analyses; for patient users, easy-to-understand disease explanations, symptom relief methods and medication recommendations are provided.

[0050] Example 1

[0051] Doctor's clinical diagnosis auxiliary retrieval

[0052] Scenario: When diagnosing a patient with chronic obstructive pulmonary disease (COPD) and cardiovascular disease, a respiratory physician needs to inquire about relevant comprehensive treatment options and the latest research progress.

[0053] Operation steps: The doctor enters "Comprehensive treatment options and latest research for COPD combined with cardiovascular disease" into the search system. The system first uses natural language processing technology to semantically understand the query, identifying the two disease entities "COPD" and "cardiovascular disease," as well as semantic relationships such as "combination," "comprehensive treatment options," and "latest research," and converts them into a query subgraph structure within the knowledge graph. Next, the FFGR algorithm is used to calculate the nodes in the knowledge graph, comprehensively considering the node's attribute characteristics (such as the pathological characteristics of the disease and the mechanism of action of therapeutic drugs), structural characteristics (such as the degree of correlation between the disease and other diseases and drugs), and semantic features related to the query (such as the semantic match with the topic "COPD combined with cardiovascular disease"). Based on the calculated node comprehensive score, the system screens relevant treatment options, including drug combinations, rehabilitation treatment methods, and the latest clinical research recommendations. This information is sorted from high to low relevance and presented to the doctor. At the same time, the system prioritizes the latest research results related to respiratory and cardiovascular departments based on the doctor's historical query records and professional fields, helping doctors quickly obtain valuable information. In terms of knowledge updating, the DKU algorithm monitors the medical database in real time. If there are new research results or treatment plans on COPD combined with cardiovascular disease, the knowledge graph will be automatically updated to ensure that doctors can obtain the latest information the next time they query.

[0054] Example 2

[0055] Patient self-health inquiry

[0056] Scenario: A patient experiences symptoms such as headache and fatigue and wants to learn about possible illnesses and corresponding relief methods.

[0057] Operation steps: The patient enters "What disease might headache and fatigue be, and how to relieve it" in the retrieval system. The system processes the query and identifies the two symptom entities of "headache" and "fatigue" and semantic relationships such as "what disease might be" and "how to relieve it" through natural language processing technology, and converts it into a query subgraph. The FFGR algorithm is used to calculate the comprehensive score of the nodes in the knowledge graph, screen out diseases related to the symptoms, such as colds, anemia, etc., and provide corresponding relief methods, such as rest and drinking more water for colds, and dietary adjustments and medical advice for anemia. Based on the patient's historical query data (if any), the system determines that the patient may be more concerned about easy-to-understand information, so the search results are presented in simple and clear language to facilitate patient understanding and application. In addition, the DKU algorithm continues to pay attention to new discoveries in the medical field. If there are new diseases related to headaches and fatigue, or more effective relief methods appear, the knowledge graph will be updated in a timely manner so that patients can obtain the latest health information.

[0058] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. An AI medical information precision retrieval method based on knowledge graph, characterized by: The steps include: S1. Knowledge graph construction and data integration: Collect and integrate multi-source medical data, apply entity recognition and relationship extraction techniques to accurately extract medical entities and semantic relationships, and build a complete and accurate medical knowledge graph; And use a distributed graph database to store knowledge graphs to ensure efficient management and fast access to data; S2. Query Semantic Understanding and Conversion: Leveraging natural language processing technology, we conduct in-depth analysis of user-entered queries. Through word segmentation, part-of-speech tagging, and named entity recognition, we accurately grasp the medical entities and semantic relationships within the query. Convert natural language queries into query subgraph structures in the knowledge graph, aligning the query with the structure of the knowledge graph and providing support for accurate retrieval; S3, Inference Calculation: a. Graph reasoning algorithm based on fused features: By fusing the attribute features, structural features, and semantic features related to the query, a comprehensive score is calculated for each node to accurately assess the relevance of the node to the query; The core formula of the FFGR algorithm is: S(n)=α·F a (n)+β·F s (n)+γ·F q (n) Among them, S(n) represents the comprehensive score of node n, F a (n) represents the attribute feature score of node n, F s (n) is the structural feature score of node n, F q (n) is the semantic feature score of node n related to the query, α, β, γ are weight coefficients optimized by machine learning methods; b. Dynamic knowledge update algorithm: Use the dynamic knowledge update algorithm to update the knowledge graph in real time, use the deep learning model to extract features and predict relationships of new data, and update the structure and content of the knowledge graph in a timely manner. The core formula of the dynamic knowledge update algorithm is: ΔG=f(N,E,G old ) Among them, ΔG represents the update amount of the knowledge graph, N represents the entity set in the new data, E is the relationship set in the new data, G old This is the knowledge graph before updating, and the function f is implemented through the deep learning model; S4. Search result optimization and personalized recommendation: Based on the node comprehensive score calculated by the FFGR algorithm, the search results are sorted and filtered, and information highly relevant to the query is displayed first; Combine the user's historical query records, browsing behavior, and disease concern data to build an accurate user profile and achieve personalized search result recommendations; For doctor users, priority is given to pushing the latest clinical research results, diagnosis and treatment guidelines and case analyses; for patient users, easy-to-understand disease explanations, symptom relief methods and medication recommendations are provided.

2. The AI ​​medical information precision retrieval method based on knowledge graph according to claim 1 is characterized by: The entity recognition and relationship extraction technology extracts medical entities of diseases, drugs, symptoms, examination items and semantic relationships of "treatment", "initiation" and "accompanying" from medical data.

3. The AI ​​medical information precision retrieval method based on knowledge graph according to claim 1 is characterized by: The natural language processing technology performs word segmentation, part-of-speech tagging, and named entity recognition on user queries to achieve semantic understanding and query subgraph conversion.

4. The AI ​​medical information precision retrieval method based on knowledge graph according to claim 1 is characterized by: The graph reasoning algorithm based on fusion features calculates a comprehensive score by comprehensively considering multiple node features. Compared with existing algorithms, it can more comprehensively and accurately measure the relevance between nodes and queries, thereby improving retrieval accuracy.

5. The AI ​​medical information precision retrieval method based on knowledge graph according to claim 1 is characterized by: The dynamic knowledge update algorithm can automatically identify changes in the knowledge graph and realize dynamic and real-time updating of the knowledge graph, which reduces manual intervention compared with traditional knowledge graph updating methods.

6. The AI ​​medical information precision retrieval method based on knowledge graph according to claim 1 is characterized by: The user portrait is constructed based on the user's historical query records, browsing behavior, and disease concern data, and is used to achieve personalized search result recommendations.

Images