A knowledge graph dynamic integration method, device, equipment and medium

Abstract

The application discloses a knowledge graph dynamic integration method, device and equipment and a medium. The method comprises the following steps: performing cleaning, integration and desensitization preprocessing on multi-source heterogeneous original data, performing medical named entity recognition and semantic relationship extraction between entities according to a preset recognition rule, performing entity alignment, relationship alignment and knowledge fusion on medical entity relationship extraction data obtained through extraction and a preset medical knowledge graph, and performing dynamic evolution processing on fused medical knowledge graph data. Through the technical means, dynamic integration and evolution update of continuously generated multi-source heterogeneous original data are realized, technical defects that a medical knowledge graph constructed statically in the prior art is difficult to dynamically integrate and evolve and that medical knowledge in the graph lags behind and is missing are overcome, the dynamically updated medical knowledge graph can accurately reflect the latest medical knowledge state in a timely manner, and through medical knowledge query reasoning and decision explanation processing, the accuracy of intelligent decision support data is improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, apparatus, device, and medium for dynamic integration of knowledge graphs. Background Technology

[0002] With the continuous advancement of medical informatization, a large amount of multi-source heterogeneous raw data is constantly generated during ophthalmological diagnosis and treatment. This includes structured data such as basic patient information and diagnostic results, as well as unstructured data such as patient complaints and consultation texts. To organize and utilize the medical knowledge in this data, existing technologies typically employ knowledge graph technology to express dispersed medical knowledge in a structured form of entities, relationships, and attributes, thereby supporting the querying and decision-making applications of medical knowledge.

[0003] Currently, medical knowledge graphs are mostly statically constructed by manual organization or one-time knowledge extraction. That is, after the graph is built, the medical entities, semantic relationships and their strength weights in the graph remain fixed for a long time, and are only supplemented and maintained manually in a limited period of time.

[0004] However, medical knowledge is constantly evolving, and new diagnostic and treatment data and research results are constantly generated in clinical practice. Medical knowledge graphs constructed in a static way are difficult to dynamically integrate and update the continuously generated multi-source heterogeneous raw data. This results in the medical knowledge in the graph being outdated and missing, and unable to reflect the latest medical knowledge status in a timely and accurate manner. Consequently, the decision support provided based on the medical knowledge graph lacks accuracy and is difficult to meet the actual needs of intelligent decision support in ophthalmology diagnosis and treatment scenarios such as optometry. Summary of the Invention

[0005] This invention provides a method, apparatus, device, and medium for dynamic integration of knowledge graphs, aiming to solve the technical problem that statically constructed medical knowledge graphs are difficult to dynamically integrate and evolve, resulting in low accuracy of decision support.

[0006] In a first aspect, embodiments of the present invention provide a method for dynamic integration of knowledge graphs, including: Multi-source heterogeneous raw data is acquired, cleaned and integrated, and then desensitized to obtain standardized preprocessed data. The multi-source heterogeneous raw data includes structured and unstructured data of ophthalmic patients. The structured data includes basic information about ophthalmic patients, diagnostic results, laboratory reports, medication records, and treatment plans. The unstructured data includes ward round records, nursing records, patient complaints, consultation transcripts, medical literature, clinical guidelines, and expert consensus. The standardized preprocessed data is subjected to medical named entity recognition according to preset recognition rules. After recognition, semantic relationship extraction processing between entities is performed to obtain medical entity relationship extraction data. The extracted medical entity relationship data is then subjected to entity alignment, relationship alignment, and knowledge fusion processing with a preset medical knowledge graph to obtain fused medical knowledge graph data. The fused medical knowledge graph data is dynamically evolved to obtain a dynamically updated medical knowledge graph; The dynamically updated medical knowledge graph is integrated into a pre-defined optometrist intelligent agent decision-making architecture for medical knowledge query, reasoning, and decision interpretation processing to obtain intelligent decision support data.

[0007] Secondly, embodiments of the present invention provide a knowledge graph dynamic integration device, comprising: The data processing unit is used to acquire multi-source heterogeneous raw data, clean and integrate the multi-source heterogeneous raw data, and then perform desensitization preprocessing to obtain standardized preprocessed data. The multi-source heterogeneous raw data includes structured and unstructured data of ophthalmic patients. The structured data includes basic information of ophthalmic patients, diagnostic results, laboratory reports, medication records, and treatment plans. The unstructured data includes ward round records, nursing records, patient complaints, consultation dialogue texts, medical literature, clinical guidelines, and expert consensus. The relation extraction unit is used to perform medical named entity recognition on the standardized preprocessed data according to preset recognition rules, and then perform semantic relation extraction processing between entities to obtain medical entity relation extraction data. The graph fusion unit is used to perform entity alignment, relationship alignment and knowledge fusion processing on the extracted medical entity relationship data and the preset medical knowledge graph respectively to obtain fused medical knowledge graph data. The graph update unit is used to perform dynamic evolution processing on the fused medical knowledge graph data to obtain a dynamically updated medical knowledge graph. The decision reasoning unit is used to connect the dynamically updated medical knowledge graph to a preset optometrist intelligent agent decision architecture to perform medical knowledge query reasoning and decision interpretation processing to obtain intelligent decision support data.

[0008] Thirdly, embodiments of the present invention provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the knowledge graph dynamic integration method of the first aspect.

[0009] Fourthly, embodiments of the present invention provide a computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, it implements the knowledge graph dynamic integration method of the first aspect.

[0010] This invention provides a method for dynamic integration of knowledge graphs, including acquiring multi-source heterogeneous raw data, cleaning and integrating the multi-source heterogeneous raw data, and then performing desensitization preprocessing to obtain standardized preprocessed data. The multi-source heterogeneous raw data includes structured and unstructured data of ophthalmic patients. The structured data includes basic information of ophthalmic patients, diagnostic results, laboratory reports, medication records, and treatment plans. The unstructured data includes ward round records, nursing records, patient complaints, consultation dialogue texts, medical literature, clinical guidelines, and expert consensus. The standardized preprocessing... The data undergoes medical named entity recognition according to preset recognition rules, followed by semantic relationship extraction processing between entities to obtain medical entity relationship extraction data. This data is then subjected to entity alignment, relationship alignment, and knowledge fusion processing with a preset medical knowledge graph to obtain fused medical knowledge graph data. The fused medical knowledge graph data undergoes dynamic evolution processing to obtain a dynamically updated medical knowledge graph. Finally, the dynamically updated medical knowledge graph is integrated into a preset optometrist intelligent agent decision-making architecture for medical knowledge query reasoning and decision interpretation processing to obtain intelligent decision support data. This invention achieves dynamic integration and evolution of continuously generated multi-source heterogeneous raw data through cleaning, integration, and desensitization preprocessing of multi-source heterogeneous raw data; medical named entity recognition and semantic relationship extraction between entities according to preset recognition rules; entity alignment, relationship alignment, and knowledge fusion of the extracted medical entity relationship data with a preset medical knowledge graph; and dynamic evolution processing of the fused medical knowledge graph data. This overcomes the technical defects of existing statically constructed medical knowledge graphs, which are difficult to dynamically integrate and evolve, resulting in lag and missing medical knowledge in the graph. The dynamically updated medical knowledge graph can reflect the latest medical knowledge status in a timely and accurate manner, thereby improving the accuracy of intelligent decision support data through medical knowledge query reasoning and decision interpretation processing.

[0011] This invention also provides a knowledge graph dynamic integration device, a computer device, and a storage medium, which have the same beneficial effects as described above. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 A flowchart illustrating a dynamic knowledge graph integration method provided in an embodiment of the present invention; Figure 2 This is a schematic block diagram of a knowledge graph dynamic integration device provided in an embodiment of the present invention.

[0014] Explanation of reference numerals in the attached figures: 200. Knowledge Graph Dynamic Integration Device; 201. Data Processing Unit; 202. Relationship Extraction Unit; 203. Knowledge Graph Fusion Unit; 204. Knowledge Graph Update Unit; 205. Decision Reasoning Unit. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0017] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0018] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0019] Please see below. Figure 1 , Figure 1 The flowchart of a dynamic knowledge graph integration method provided in this embodiment of the invention specifically includes steps S101 to S105.

[0020] S101. Obtain multi-source heterogeneous raw data, clean and integrate the multi-source heterogeneous raw data, and then perform desensitization preprocessing to obtain standardized preprocessed data; wherein, the multi-source heterogeneous raw data includes structured data and unstructured data of ophthalmic patients, the structured data includes basic information of ophthalmic patients, diagnostic results, test reports, medication records and treatment plans; the unstructured data includes ward round records, nursing records, patient complaints, consultation dialogue texts, medical literature, clinical guidelines and expert consensus; S102. The standardized preprocessed data is subjected to medical named entity recognition according to preset recognition rules. After recognition, the semantic relationship between entities is extracted to obtain medical entity relationship extraction data. S103. Perform entity alignment, relationship alignment and knowledge fusion processing on the extracted medical entity relationship data and the preset medical knowledge graph respectively to obtain fused medical knowledge graph data; S104. Perform dynamic evolution processing on the fused medical knowledge graph data to obtain a dynamically updated medical knowledge graph; S105. The dynamically updated medical knowledge graph is connected to the preset optometrist intelligent agent decision-making architecture for medical knowledge query reasoning and decision interpretation processing to obtain intelligent decision support data.

[0021] In step S101, the raw data continuously generated during ophthalmological diagnosis and treatment comes from diverse sources and has various forms. It includes structured data such as basic patient information, diagnostic results, laboratory reports, medication records, and treatment plans, as well as unstructured data such as ward round records, nursing records, patient complaints, consultation transcripts, medical literature, clinical guidelines, and expert consensus. Because these multi-source heterogeneous raw data differ in data format and quality, and contain sensitive information involving patient privacy, after acquiring the multi-source heterogeneous raw data, it is first cleaned and integrated to eliminate noise and inconsistencies. Then, the cleaned and integrated data undergoes desensitization preprocessing to remove or conceal sensitive information, resulting in unified, standardized preprocessed data that meets privacy protection requirements. This provides a reliable data foundation for subsequent medical named entity recognition and semantic relationship extraction.

[0022] In one embodiment, step S101 includes: Structured and unstructured data are collected from multiple data source systems to obtain multi-source heterogeneous raw data; wherein, the data source systems include hospital information systems, electronic medical record systems, medical settlement systems, medical imaging systems, and wearable devices; The multi-source heterogeneous raw data are cleaned, deduplicated, standardized, and format converted to obtain standardized cleaned data. The scattered records belonging to the same object in the standardized cleaned data are linked and integrated in chronological order to obtain object timeline integrated data; wherein, the scattered records include medical records of the same eye patient at different times and in different departments, and the object timeline integrated data is patient medical timeline data linked and integrated in chronological order of medical visits; The unstructured text in the timeline integrated data of the object is processed by word segmentation, part-of-speech tagging and syntactic analysis to obtain text parsing and integrated data; The parsed and integrated text data is anonymized and / or pseudonymous to obtain standardized preprocessed data.

[0023] In this embodiment, considering that the medical data of eye patients are stored in different business systems, in order to obtain multi-source heterogeneous raw data as comprehensively as possible, structured data and unstructured data can be collected from multiple data source systems respectively. For example, the data source systems include a Hospital Information System (HIS), an Electronic Medical Record (EMR) system, a medical settlement system, a Picture Archiving and Communication System (PACS), and wearable devices. The Hospital Information System mainly provides management data such as registration, consultation, and settlement for eye patients; the Electronic Medical Record System mainly provides medical record documents; the Medical Settlement System mainly provides cost settlement data; the Medical Archiving and Communication System mainly provides images and reports such as fundus photography and optical coherence tomography; and the wearable devices can provide relevant monitoring data continuously collected from eye patients outside the hospital.

[0024] Structured data collected from the aforementioned data source systems may include, for example, basic information about ophthalmic patients (such as age, gender, medical history, etc.), diagnostic results (such as diagnostic conclusions expressed using the International Classification of Diseases (ICD) code), laboratory reports (such as complete blood count, biochemistry, and imaging reports, etc.), medication records, surgical records, treatment plans, and follow-up records, etc.; unstructured data collected may include, for example, ward round records, nursing records, patient complaints, consultation transcripts, medical literature, clinical guidelines, and expert consensus, etc. Combining the above structured and unstructured data yields multi-source heterogeneous raw data.

[0025] After obtaining multi-source heterogeneous raw data, which originates from different data source systems and generally suffers from inconsistent data formats, duplicate records, and inconsistent field naming and value standards, the multi-source heterogeneous raw data is cleaned, deduplicated, standardized, and format converted to obtain standardized cleaned data. For example, the cleaning process removes or corrects noisy, missing, and erroneous data; the deduplication process eliminates redundant records collected repeatedly; the standardization process unifies field names, units of measurement, and coded values ​​from different sources into a standardized form; and the format conversion process converts data in different formats into a unified data format, thereby providing reliable, standardized cleaned data with a consistent format for subsequent processing.

[0026] Considering that the same eye patient often visits different departments at different times, their medical records are scattered throughout the standardized cleaned data. Therefore, the scattered records belonging to the same subject in the standardized cleaned data are linked and integrated in chronological order to obtain the integrated timeline data. The scattered records include medical records of the same eye patient at different times and in different departments. By linking and integrating the scattered records belonging to the same eye patient in chronological order of visits, a patient's medical timeline data that fully reflects the entire treatment process of that eye patient, linked and integrated in chronological order, can be obtained—that is, the integrated timeline data.

[0027] Furthermore, since the unstructured text in the object timeline integrated data is difficult for computers to directly understand and process, the unstructured text in the object timeline integrated data is subjected to word segmentation, part-of-speech tagging, and syntactic analysis to obtain text parsing and integration data. For example, word segmentation is used to divide continuous text into word units with independent semantics, part-of-speech tagging is used to label the part-of-speech category of each word unit, and syntactic analysis is used to parse the syntactic structural relationships between word units, thereby converting the unstructured text into text parsing and integration data that is convenient for subsequent medical named entity recognition and semantic relationship extraction.

[0028] Finally, considering that the medical data of eye patients contains a large amount of sensitive information involving patient privacy, to ensure data security and compliant use, the text parsing and integration data is anonymized and / or pseudonymous to obtain standardized preprocessed data. Anonymization removes information from the data that can directly or indirectly identify a specific eye patient, making it impossible to reconstruct the identity of that individual. Pseudonymous replaces the aforementioned identifiable information with pseudonyms, reducing the risk of privacy leakage while preserving the value of data correlation analysis. After the above privacy anonymization processing, standardized preprocessed data that meets privacy protection requirements and can be used for subsequent processing is obtained.

[0029] In step S102, after obtaining the standardized preprocessed data, medical named entity recognition is performed on the standardized preprocessed data according to preset recognition rules to identify the various medical entities contained in the standardized preprocessed data. After completing the medical named entity recognition, the semantic relationships between the identified medical entities are further extracted to clarify the associations between the medical entities, thereby obtaining medical entity relationship extraction data. Through this step, the medical knowledge contained in the standardized preprocessed data can be transformed into medical entity relationship extraction data expressed in the form of entities and semantic relationships between entities, which facilitates subsequent alignment and fusion processing with the medical knowledge graph.

[0030] In one embodiment, step S102 includes: The standardized preprocessed data is used to perform medical named entity recognition according to the sequence labeling model in the preset recognition rules to obtain medical candidate named entity data. The medical candidate named entity data is categorized and labeled according to preset medical entity types to obtain medical entity category labeling data; Based on the context of the medical entity category annotation data, the semantic relationships between each medical entity are identified and extracted to obtain medical entity semantic relationship data. The attribute information associated with each medical entity in the medical entity semantic relationship data is extracted and mounted to obtain medical entity attribute mounting data; wherein, the attribute information includes the dosage attribute and usage attribute of the drug entity; The medical entity attribute mounting data is organized and encapsulated according to the triple structure of entity, relation, and attribute to obtain medical entity relation extraction data.

[0031] In this embodiment, to accurately identify the medical entities contained in the standardized preprocessed data, the standardized preprocessed data is subjected to Named Entity Recognition (NER) according to the sequence labeling model in the preset recognition rules to obtain medical candidate named entity data. NER is used to automatically identify and locate entities with specific medical meanings from medical text. For example, the sequence labeling model can be a deep learning-based Bidirectional Long Short-Term Memory - Conditional Random Field (Bi-LSTM-CRF) model, or it can be implemented using a combination of rules and dictionaries; no limitation is made here. Taking the text fragment "A patient sought medical attention due to blurred vision, was diagnosed with myopia, and was prescribed eyeglasses" from the standardized preprocessed data as an example, after NER, medical candidate named entities such as "blurred vision," "myopia," and "eyeglasses" can be identified, thus obtaining medical candidate named entity data.

[0032] After identifying the candidate named entity data for medical applications, to facilitate subsequent differentiation and processing of each medical entity, the candidate named entity data is categorized and labeled according to preset medical entity types, resulting in medical entity category labeled data. For example, the preset medical entity types include diseases, symptoms, drugs, examination items, anatomical locations, surgeries, treatment methods, and complications. Continuing the previous example, after categorizing and labeling the identified candidate named entities, "blurred vision" can be labeled as a symptom entity, "myopia" as a disease entity, and "glasses" as a treatment method entity, thus obtaining medical entity category labeled data with the corresponding entity categories.

[0033] Since medical entities often possess specific semantic relationships within the context of text, semantic relationships between them are identified and extracted based on the context of the medical entity category annotation data, resulting in medical entity semantic relationship data. For example, semantic relationship identification and extraction can be achieved using rule-based, supervised learning, or remote supervision methods. Continuing the previous example, by analyzing the context, a "cause" relationship can be identified and extracted between "myopia" and "blurred vision," and a "treatment" relationship can be extracted between "myopia" and "glasses," thus obtaining medical entity semantic relationship data that records the semantic relationships between various medical entities.

[0034] Furthermore, considering that some medical entities also have associated attribute information to supplement their descriptions, the attribute information associated with each medical entity in the semantic relationship data of the medical entities is extracted and attached to obtain medical entity attribute attachment data. The attribute information includes the dosage attribute and usage attribute of the drug entity; for example, for a drug entity appearing in the text, its corresponding dosage attribute (such as dosage per dose) and usage attribute (such as daily dosage frequency and route of administration) can be extracted together and attached to the drug entity, thereby obtaining medical entity attribute attachment data.

[0035] Finally, to facilitate subsequent alignment and integration with the medical knowledge graph, the medical entity attribute data is organized and encapsulated according to a triple structure of entity, relation, and attribute to obtain medical entity relation extraction data. For example, each medical entity, the semantic relationships between medical entities, and the attribute information associated with each medical entity are uniformly organized and encapsulated in the form of interrelated triples of entity, relation, and attribute, thereby obtaining structurally standardized medical entity relation extraction data that is easy to process later.

[0036] In step S103, after obtaining the medical entity relation extraction data, the medical entity relation extraction data is aligned with a preset medical knowledge graph to determine the correspondence between the medical entities in the medical entity relation extraction data and the existing content in the medical knowledge graph; based on the entity alignment, relation alignment is performed to determine the correspondence between the semantic relations in the medical entity relation extraction data and the existing relations in the medical knowledge graph; then, based on the results of entity alignment and relation alignment, knowledge fusion processing is performed to incorporate the medical knowledge carried by the medical entity relation extraction data into the medical knowledge graph, thereby obtaining fused medical knowledge graph data, realizing the effective connection and integration of newly extracted medical knowledge with the existing medical knowledge graph.

[0037] In one embodiment, step S103 includes: The medical entities in the extracted medical entity relationship data are matched with the existing concepts in the medical knowledge graph according to string similarity, vector embedding similarity or ontology reasoning to obtain medical entity similarity matching data. The medical entities whose similarity in the medical entity similarity matching data meets a preset threshold are mapped and bound to the corresponding existing concepts to obtain medical entity alignment mapping data. Based on the medical entity alignment mapping data, the semantic relationships in the extracted medical entity relationship data are compared and aligned with the existing relationships in the medical knowledge graph to obtain medical relationship alignment comparison data. The confidence of existing relationships in the medical relationship alignment ratio data is updated, and non-existent relationships are marked as candidate new relationships to obtain medical relationship fusion labeling data; The medical relationship fusion marker data is arbitrated for knowledge conflict according to preset arbitration rules and then incorporated into the medical knowledge graph to obtain fused medical knowledge graph data; wherein, the preset arbitration rules include voting mechanism arbitration rules, expert review arbitration rules, or confidence-based arbitration rules.

[0038] In this embodiment, to determine the correspondence between each medical entity in the extracted medical entity relationship data and existing concepts in the medical knowledge graph, the medical entities in the extracted medical entity relationship data and existing concepts in the medical knowledge graph are matched using string similarity, vector embedding similarity, or ontology reasoning methods to obtain medical entity similarity matching data. The medical knowledge graph can be constructed using existing systems nomenclature of medicine - clinical terms (SNOMED CT), the Unified Medical Language System (UMLS), and the International Classification of Diseases (ICD), among others. String similarity matching is based on the similarity between medical entities and existing concepts at the character level; vector embedding similarity matching is based on the semantic similarity between the two in the vector space; and ontology reasoning matching is based on the logical relationships between concepts in the medical ontology. For example, the entity “hypertension” extracted from the text can be matched with the corresponding concept “hypertension” in the medical knowledge graph through the above similarity matching calculation, thereby obtaining medical entity similarity matching data that reflects the degree of similarity between each medical entity and the existing concept.

[0039] After obtaining medical entity similarity matching data, medical entities whose similarity meets a preset threshold are mapped and bound to their corresponding existing concepts to obtain medical entity alignment mapping data. For example, when the similarity between a medical entity and an existing concept reaches a preset threshold, they are considered to point to the same medical concept. The medical entity is then mapped and bound to its corresponding existing concept, thereby achieving alignment between the medical entity and the existing concept and obtaining medical entity alignment mapping data.

[0040] After completing the alignment mapping of medical entities, the semantic relationships in the extracted medical entity relationship data are compared and aligned with the existing relationships in the medical knowledge graph based on the medical entity alignment mapping data, resulting in medical relationship alignment comparison data. That is, based on the already aligned medical entities, the semantic relationships between each medical entity in the extracted medical entity relationship data are further compared and aligned one by one with the existing relationships between the corresponding medical concepts in the medical knowledge graph, thereby obtaining medical relationship alignment comparison data reflecting the correspondence between the two relationships.

[0041] Furthermore, the confidence levels of existing relationships in the medical relationship alignment comparison data are updated, and non-existent relationships are marked as candidate new relationships, resulting in medical relationship fusion labeled data. For example, for relationships already existing in the medical knowledge graph, their confidence levels are updated through this alignment comparison to reflect the support provided by newly extracted medical knowledge for that relationship; for relationships not yet existing in the medical knowledge graph, they are marked as candidate new relationships for subsequent evaluation and processing, thus obtaining the corresponding labeled medical relationship fusion labeled data.

[0042] Finally, considering the potential conflicts between the newly extracted medical knowledge and existing knowledge in the medical knowledge graph, the medical relationship fusion marker data is arbitrated according to preset arbitration rules before being incorporated into the medical knowledge graph, resulting in fused medical knowledge graph data. The preset arbitration rules include voting mechanism arbitration rules, expert review arbitration rules, or confidence-based arbitration rules. Voting mechanism arbitration rules determine which knowledge to adopt by voting on knowledge from multiple sources; expert review arbitration rules involve experts reviewing and judging conflicting knowledge; and confidence-based arbitration rules compare the confidence levels of conflicting knowledge to select the best option. After the above-mentioned knowledge conflict arbitration and the incorporation of the corresponding knowledge into the medical knowledge graph, fused medical knowledge graph data that effectively connects and integrates newly extracted medical knowledge with existing medical knowledge is obtained.

[0043] In step S104, considering that medical knowledge is constantly evolving, after obtaining the fused medical knowledge graph data, the data undergoes further dynamic evolution processing. This allows the medical knowledge in the graph to be continuously updated and improved with the continuously generated multi-source heterogeneous original data, thus obtaining a dynamically updated medical knowledge graph. This step overcomes the shortcomings of statically constructed medical knowledge graphs, which are difficult to dynamically integrate and evolve, enabling the dynamically updated medical knowledge graph to reflect the latest state of medical knowledge in a timely and accurate manner.

[0044] In one embodiment, step S104 includes: Statistical analysis was performed on the frequency of occurrence and co-occurrence patterns of each relationship in the fused medical knowledge graph data, as well as the corresponding clinical outcomes, to obtain statistical characteristic data of medical relationships; Based on the statistical feature data of medical relationships, the strength weights of corresponding relationships in the fused medical knowledge graph data are dynamically adjusted to obtain medical relationship strength adjustment graph data; The newly added medical entities and relationships in the medical relationship strength adjustment graph data are verified and confirmed, and the new knowledge that has passed the verification is incorporated into the graph, resulting in new medical knowledge incorporated into the graph data; The new medical knowledge is incorporated into the atlas data for knowledge conflict and inconsistency detection, and the detected conflicting knowledge is resolved to obtain medical conflict resolution atlas data; The medical conflict resolution graph data is managed by version recording based on update content, update time, and update source to obtain a dynamically updated medical knowledge graph.

[0045] In this embodiment, to enable the fused medical knowledge graph data to reflect the actual situation of each medical relationship based on real-world medical data, statistical analysis is performed on the occurrence frequency and co-occurrence pattern of each relationship in the fused medical knowledge graph data, as well as the corresponding clinical results, to obtain statistical characteristic data of medical relationships. For example, by statistically analyzing the occurrence frequency of each relationship in a large amount of medical data, the co-occurrence pattern between related medical entities, and the clinical results corresponding to the relationship, statistical characteristic data of medical relationships that can characterize the actual performance of each medical relationship can be obtained.

[0046] After obtaining the statistical feature data of medical relationships, the strength weights of the corresponding relationships in the fused medical knowledge graph data are dynamically adjusted based on the statistical feature data of medical relationships to obtain medical relationship strength adjustment graph data. For example, the dynamic adjustment of the strength weights can be achieved through statistical analysis, machine learning models (such as Graph Neural Network (GNN)) or Reinforcement Learning (RL); for instance, when a large amount of clinical data shows that a certain drug has a significant therapeutic effect on a specific disease, the strength weight of the "treatment" relationship between the drug entity and the disease entity can be increased accordingly, thereby obtaining the medical relationship strength adjustment graph data after the corresponding relationship strength has been dynamically adjusted.

[0047] Furthermore, the newly added medical entities and relationships in the medical relationship strength adjustment graph data are verified, and the verified new knowledge is incorporated into the graph, resulting in medical new knowledge incorporated into the graph data. For example, for newly added medical entities and relationships discovered during the aforementioned processing that were not originally present in the medical knowledge graph, they are first verified, and then the verified new knowledge is incorporated into the graph. This allows the graph to promptly capture the latest advancements and clinical practice experience in the medical field, resulting in medical new knowledge incorporated into the graph data.

[0048] Considering that new knowledge incorporated into the atlas may conflict with or be inconsistent with existing knowledge, the atlas data containing new medical knowledge is subjected to knowledge conflict and inconsistency detection. Detected conflicting knowledge is then resolved to obtain medical conflict resolution atlas data. For example, when newly incorporated knowledge is detected to contradict existing knowledge in the atlas, the corresponding conflict is marked, and experts are prompted to review and judge it, thereby resolving the conflicting knowledge and obtaining medical conflict resolution atlas data that has eliminated knowledge conflicts and inconsistencies.

[0049] Finally, to facilitate the tracing and retrospection of each update to the medical conflict resolution graph, the data is managed through version recording based on update content, update time, and update source, resulting in a dynamically updated medical knowledge graph. That is, for each update to the graph, the corresponding update content, update time, and update source are recorded and version managed. This ensures the dynamic evolution of the graph while maintaining the traceability of its update process, ultimately yielding a dynamically updated medical knowledge graph that can timely and accurately reflect the latest state of medical knowledge.

[0050] In step S105, after obtaining the dynamically updated medical knowledge graph, it is connected to a preset optometrist intelligent agent decision-making architecture. The optometrist intelligent agent decision-making architecture performs medical knowledge query and reasoning based on the dynamically updated medical knowledge graph, and performs decision interpretation processing on the reasoned decision content, thereby obtaining intelligent decision support data. Since the intelligent decision support data is generated based on a dynamically updated medical knowledge graph that can reflect the latest medical knowledge status in a timely manner, and includes corresponding decision interpretations, the accuracy and interpretability of decision support are improved, meeting the actual needs of intelligent decision support in ophthalmological diagnosis and treatment scenarios such as optometry. The preset optometrist intelligent agent decision-making architecture refers to a pre-constructed intelligent decision-making architecture with an optometrist intelligent agent as the decision-making subject. The optometrist intelligent agent is an artificial intelligence (AI) agent used to simulate optometrists making diagnosis and treatment decisions in ophthalmology diagnosis and treatment scenarios such as optometry. For example, the optometrist intelligent agent decision-making architecture can be implemented using an architecture based on multi-agent reinforcement learning, which is used to initiate knowledge queries to the accessed knowledge sources during the diagnosis and treatment decision-making process, and generate corresponding decision schemes based on the query reasoning results, thereby providing intelligent decision support for ophthalmology diagnosis and treatment scenarios such as optometry.

[0051] In one embodiment, step S105 includes: The dynamically updated medical knowledge graph is connected to the preset optometrist intelligent agent decision architecture, and the knowledge query request initiated by the optometrist intelligent agent decision architecture is received to obtain the knowledge query request data. Based on the knowledge query request data, related knowledge retrieval and medical knowledge reasoning operations are performed in the dynamically updated knowledge graph to obtain knowledge reasoning result data; wherein, the medical knowledge reasoning operation includes reasoning corresponding candidate diseases based on patient symptoms, and / or recommending corresponding examination items and treatment plans based on diseases; The knowledge reasoning results are fed back to the optometrist's intelligent agent decision-making architecture to generate a decision scheme and obtain candidate decision scheme data. The decision path is traced to the supporting knowledge corresponding to the candidate decision scheme data in the dynamically updated medical knowledge graph to obtain decision interpretation path data; wherein, the supporting knowledge includes medical evidence and clinical guidelines that support the candidate decision scheme data in the dynamically updated medical knowledge graph. The candidate decision scheme data and the decision interpretation path data are associated, encapsulated, and output to obtain intelligent decision support data.

[0052] In this embodiment, the dynamically updated medical knowledge graph is connected to a preset optometrist intelligent agent decision-making architecture, and knowledge query requests initiated by the optometrist intelligent agent decision-making architecture are received to obtain knowledge query request data. That is, when the optometrist intelligent agent decision-making architecture needs to obtain relevant medical knowledge for diagnosis and treatment decisions, it initiates a knowledge query request to the connected dynamically updated medical knowledge graph, thereby obtaining the corresponding knowledge query request data.

[0053] Upon receiving a knowledge query request, the system performs related knowledge retrieval and medical knowledge reasoning operations on the dynamically updated knowledge graph based on the request data to obtain knowledge reasoning result data. The medical knowledge reasoning operations include inferring candidate diseases based on patient symptoms and / or recommending corresponding examination items and treatment plans based on the diseases. For example, based on the symptoms exhibited by an eye patient, related knowledge can be retrieved from the dynamically updated medical knowledge graph to infer candidate diseases corresponding to those symptoms; alternatively, for a specific disease, corresponding examination items and treatment plans can be recommended, thereby obtaining the corresponding knowledge reasoning result data.

[0054] After obtaining the knowledge reasoning result data, the knowledge reasoning result data is sent back to the optometrist's intelligent agent decision-making architecture to generate a decision scheme, thus obtaining candidate decision scheme data. That is, the optometrist's intelligent agent decision-making architecture generates corresponding candidate decision schemes based on the sent-back knowledge reasoning result data, thereby obtaining candidate decision scheme data.

[0055] Furthermore, to ensure the interpretability of the generated candidate decision schemes, decision path tracing is performed on the supporting knowledge corresponding to the candidate decision scheme data in the dynamically updated medical knowledge graph to obtain decision interpretation path data. The supporting knowledge includes medical evidence and clinical guidelines in the dynamically updated medical knowledge graph that support the candidate decision scheme data. For example, when the optometrist intelligent agent decision architecture generates a candidate decision scheme, it can trace back to the medical evidence and clinical guidelines supporting that candidate decision scheme in the dynamically updated medical knowledge graph, thereby obtaining decision interpretation path data that explains the basis for the generation of the candidate decision scheme.

[0056] Finally, the candidate decision scheme data and the decision interpretation path data are associated, encapsulated, and output to obtain intelligent decision support data. This involves associating the candidate decision scheme data and its corresponding decision interpretation path data, encapsulating them together, and outputting the resulting intelligent decision support data that includes both decision schemes and corresponding decision interpretations, thereby improving the accuracy and interpretability of decision support.

[0057] In one embodiment, after obtaining the intelligent decision support data, the method further includes: The execution results and feedback information generated by the intelligent decision support data in the actual application process are collected to obtain decision execution feedback data; The decision execution feedback data is quantified according to a preset reward mechanism to obtain feedback reward quantification data; Based on the feedback reward quantification data, the relationship strength and confidence of the corresponding knowledge in the dynamically updated medical knowledge graph are retrospectively adjusted to obtain a closed-loop optimized medical knowledge graph; wherein, when the decision execution feedback data represents the application effect of the corresponding decision scheme and meets the preset effect conditions, the relationship strength and confidence of the corresponding knowledge in the dynamically updated medical knowledge graph are enhanced. The closed-loop optimized medical knowledge graph is synchronously replaced with the dynamically updated medical knowledge graph and connected to the optometrist intelligent agent decision-making architecture to obtain continuously evolving knowledge graph service data.

[0058] In this embodiment, to enable the dynamically updated medical knowledge graph to be continuously optimized based on the actual application of intelligent decision support data, after obtaining the intelligent decision support data, the execution results and feedback information generated by the intelligent decision support data in actual application are collected to obtain decision execution feedback data. For example, the decisions made by the optometrist's intelligent agent decision-making architecture based on the intelligent decision support data during actual diagnosis and treatment, their execution results, and the resulting patient feedback can all be collected as feedback information to obtain decision execution feedback data.

[0059] After obtaining the decision execution feedback data, the decision execution feedback data is quantified according to a preset reward mechanism to obtain feedback reward quantification data. For example, the preset reward mechanism can be implemented based on the reward mechanism of reinforcement learning (RL). By quantifying the reward of the decision execution feedback data, the actual application effect of the decision scheme can be quantified and represented as a corresponding reward value, thereby obtaining feedback reward quantification data.

[0060] Furthermore, based on the quantified feedback reward data, the relationship strength and confidence of corresponding knowledge in the dynamically updated medical knowledge graph are retrospectively adjusted to obtain a closed-loop optimized medical knowledge graph. Specifically, when the decision execution feedback data indicates that the application effect of the corresponding decision scheme meets preset effect conditions, the relationship strength and confidence of the corresponding knowledge in the dynamically updated medical knowledge graph are enhanced. For example, when a decision scheme performs well in practical application and meets preset effect conditions, the relationship strength and confidence of the corresponding knowledge supporting that decision scheme in the dynamically updated medical knowledge graph are correspondingly enhanced, thereby obtaining a feedback-driven optimized closed-loop medical knowledge graph.

[0061] Finally, the closed-loop optimized medical knowledge graph is synchronously replaced with the dynamically updated medical knowledge graph and integrated into the optometrist's intelligent agent decision-making architecture to obtain continuously evolving knowledge graph service data. In other words, the closed-loop optimized medical knowledge graph synchronously replaces the original dynamically updated medical knowledge graph and is re-integrated into the optometrist's intelligent agent decision-making architecture to continuously provide knowledge support, thus forming a closed loop of "decision-feedback-optimization." This allows the medical knowledge graph to continuously learn and evolve, maintaining the real-time nature and accuracy of its knowledge, ultimately resulting in continuously evolving knowledge graph service data.

[0062] In summary, this application obtains multi-source heterogeneous raw data, cleans and integrates it, and then performs de-identification preprocessing to obtain standardized preprocessed data. The standardized preprocessed data is then subjected to medical named entity recognition according to preset recognition rules, followed by semantic relationship extraction processing between entities to obtain medical entity relationship extraction data. This medical entity relationship extraction data is then subjected to entity alignment, relationship alignment, and knowledge fusion processing with a preset medical knowledge graph to obtain fused medical knowledge graph data. The fused medical knowledge graph data undergoes dynamic evolution processing to obtain a dynamically updated medical knowledge graph. Finally, the dynamically updated medical knowledge graph is integrated into a preset optometrist intelligent agent decision-making architecture for medical knowledge query reasoning and decision interpretation processing to obtain intelligent decision support data.

[0063] This application ensures the standardization, integrity, and privacy of multi-source heterogeneous raw data, including structured and unstructured data from eye patients, through cleaning, integration, and desensitization preprocessing, providing a data foundation for knowledge extraction. By using medical named entity recognition and semantic relation extraction between entities, medical knowledge scattered throughout the multi-source heterogeneous raw data is transformed into structured medical entity relation extraction data, achieving automated extraction of medical knowledge contained in real-world medical data. Through entity alignment, relation alignment, and knowledge fusion processing, the newly extracted medical knowledge is effectively connected and integrated with a pre-set medical knowledge graph, ensuring the consistency and accuracy of the graph knowledge. Furthermore, by dynamically evolving the fused medical knowledge graph data, the graph... The medical knowledge in the graph can be continuously updated and improved with the continuous generation of multi-source heterogeneous raw data. This overcomes the technical defects of existing statically constructed medical knowledge graphs, which are difficult to dynamically integrate and evolve, resulting in lag and missing medical knowledge in the graph. This allows the dynamically updated medical knowledge graph to reflect the latest medical knowledge status in a timely and accurate manner. On this basis, the dynamically updated medical knowledge graph is integrated into a pre-set optometrist intelligent agent decision-making architecture for medical knowledge query reasoning and decision interpretation processing. This ensures that the obtained intelligent decision support data is generated based on the latest medical knowledge and has traceable decision interpretation, thereby improving the accuracy, reliability and interpretability of intelligent decision support. This can meet the actual needs of intelligent decision support in ophthalmology diagnosis and treatment scenarios such as optometry.

[0064] Combination Figure 2 As shown, Figure 2 This is a schematic block diagram of a dynamic knowledge graph integration device provided in an embodiment of the present invention. The dynamic knowledge graph integration device 200 includes: Data processing unit 201 is used to acquire multi-source heterogeneous raw data, clean and integrate the multi-source heterogeneous raw data, and then perform desensitization preprocessing to obtain standardized preprocessed data; wherein, the multi-source heterogeneous raw data includes structured data and unstructured data of ophthalmic patients, the structured data includes basic information of ophthalmic patients, diagnostic results, test reports, medication records and treatment plans; the unstructured data includes ward round records, nursing records, patient complaints, consultation dialogue texts, medical literature, clinical guidelines and expert consensus; The relation extraction unit 202 is used to perform medical named entity recognition on the standardized preprocessed data according to preset recognition rules, and then perform semantic relation extraction processing between entities to obtain medical entity relation extraction data. The graph fusion unit 203 is used to perform entity alignment, relationship alignment and knowledge fusion processing on the extracted medical entity relationship data and the preset medical knowledge graph respectively to obtain fused medical knowledge graph data. The graph update unit 204 is used to perform dynamic evolution processing on the fused medical knowledge graph data to obtain a dynamically updated medical knowledge graph. The decision reasoning unit 205 is used to connect the dynamically updated medical knowledge graph to a preset optometrist intelligent agent decision architecture to perform medical knowledge query reasoning and decision interpretation processing to obtain intelligent decision support data.

[0065] In this embodiment, the data processing unit 201 acquires multi-source heterogeneous raw data, cleans and integrates the multi-source heterogeneous raw data, and then performs desensitization preprocessing to obtain standardized preprocessed data. The multi-source heterogeneous raw data includes structured and unstructured data of ophthalmic patients. The structured data includes basic information of ophthalmic patients, diagnostic results, test reports, medication records, and treatment plans. The unstructured data includes ward round records, nursing records, patient complaints, consultation dialogue texts, medical literature, clinical guidelines, and expert consensus. The relationship extraction unit 202 extracts the standardized preprocessed data according to preset identification rules. Named entity recognition is performed, followed by semantic relationship extraction to obtain medical entity relationship extraction data. A graph fusion unit 203 performs entity alignment, relationship alignment, and knowledge fusion processing on the extracted medical entity relationship data and a preset medical knowledge graph to obtain fused medical knowledge graph data. A graph update unit 204 performs dynamic evolution processing on the fused medical knowledge graph data to obtain a dynamically updated medical knowledge graph. A decision reasoning unit 205 connects the dynamically updated medical knowledge graph to a preset optometrist intelligent agent decision-making architecture for medical knowledge query reasoning and decision interpretation processing to obtain intelligent decision support data.

[0066] In one embodiment, the data processing unit 201 is specifically used for: Structured and unstructured data are collected from multiple data source systems to obtain multi-source heterogeneous raw data; wherein, the data source systems include hospital information systems, electronic medical record systems, medical settlement systems, medical imaging systems, and wearable devices; The multi-source heterogeneous raw data are cleaned, deduplicated, standardized, and format converted to obtain standardized cleaned data. The scattered records belonging to the same object in the standardized cleaned data are linked and integrated in chronological order to obtain object timeline integrated data; wherein, the scattered records include medical records of the same eye patient at different times and in different departments, and the object timeline integrated data is patient medical timeline data linked and integrated in chronological order of medical visits; The unstructured text in the timeline integrated data of the object is processed by word segmentation, part-of-speech tagging and syntactic analysis to obtain text parsing and integrated data; The parsed and integrated text data is anonymized and / or pseudonymous to obtain standardized preprocessed data.

[0067] In one embodiment, the relation extraction unit 202 is specifically used for: The standardized preprocessed data is used to perform medical named entity recognition according to the sequence labeling model in the preset recognition rules to obtain medical candidate named entity data. The medical candidate named entity data is categorized and labeled according to preset medical entity types to obtain medical entity category labeling data; Based on the context of the medical entity category annotation data, the semantic relationships between each medical entity are identified and extracted to obtain medical entity semantic relationship data. The attribute information associated with each medical entity in the medical entity semantic relationship data is extracted and mounted to obtain medical entity attribute mounting data; wherein, the attribute information includes the dosage attribute and usage attribute of the drug entity; The medical entity attribute mounting data is organized and encapsulated according to the triple structure of entity, relation, and attribute to obtain medical entity relation extraction data.

[0068] In one embodiment, the map fusion unit 203 is specifically used for: The medical entities in the extracted medical entity relationship data are matched with the existing concepts in the medical knowledge graph according to string similarity, vector embedding similarity or ontology reasoning to obtain medical entity similarity matching data. The medical entities whose similarity in the medical entity similarity matching data meets a preset threshold are mapped and bound to the corresponding existing concepts to obtain medical entity alignment mapping data. Based on the medical entity alignment mapping data, the semantic relationships in the extracted medical entity relationship data are compared and aligned with the existing relationships in the medical knowledge graph to obtain medical relationship alignment comparison data. The confidence of existing relationships in the medical relationship alignment ratio data is updated, and non-existent relationships are marked as candidate new relationships to obtain medical relationship fusion labeling data; The medical relationship fusion marker data is arbitrated for knowledge conflict according to preset arbitration rules and then incorporated into the medical knowledge graph to obtain fused medical knowledge graph data; wherein, the preset arbitration rules include voting mechanism arbitration rules, expert review arbitration rules, or confidence-based arbitration rules.

[0069] In one embodiment, the map update unit 204 is specifically used for: Statistical analysis was performed on the frequency of occurrence and co-occurrence patterns of each relationship in the fused medical knowledge graph data, as well as the corresponding clinical outcomes, to obtain statistical characteristic data of medical relationships; Based on the statistical feature data of medical relationships, the strength weights of corresponding relationships in the fused medical knowledge graph data are dynamically adjusted to obtain medical relationship strength adjustment graph data; The newly added medical entities and relationships in the medical relationship strength adjustment graph data are verified and confirmed, and the new knowledge that has passed the verification is incorporated into the graph, resulting in new medical knowledge incorporated into the graph data; The new medical knowledge is incorporated into the atlas data for knowledge conflict and inconsistency detection, and the detected conflicting knowledge is resolved to obtain medical conflict resolution atlas data; The medical conflict resolution graph data is managed by version recording based on update content, update time, and update source to obtain a dynamically updated medical knowledge graph.

[0070] In one embodiment, the decision reasoning unit 205 is specifically used for: The dynamically updated medical knowledge graph is connected to the preset optometrist intelligent agent decision architecture, and the knowledge query request initiated by the optometrist intelligent agent decision architecture is received to obtain the knowledge query request data. Based on the knowledge query request data, related knowledge retrieval and medical knowledge reasoning operations are performed in the dynamically updated knowledge graph to obtain knowledge reasoning result data; wherein, the medical knowledge reasoning operation includes reasoning corresponding candidate diseases based on patient symptoms, and / or recommending corresponding examination items and treatment plans based on diseases; The knowledge reasoning results are fed back to the optometrist's intelligent agent decision-making architecture to generate a decision scheme and obtain candidate decision scheme data. The decision path is traced to the supporting knowledge corresponding to the candidate decision scheme data in the dynamically updated medical knowledge graph to obtain decision interpretation path data; wherein, the supporting knowledge includes medical evidence and clinical guidelines that support the candidate decision scheme data in the dynamically updated medical knowledge graph. The candidate decision scheme data and the decision interpretation path data are associated, encapsulated, and output to obtain intelligent decision support data.

[0071] In one embodiment, the knowledge graph dynamic integration device 200 is further configured to: The execution results and feedback information generated by the intelligent decision support data in the actual application process are collected to obtain decision execution feedback data; The decision execution feedback data is quantified according to a preset reward mechanism to obtain feedback reward quantification data; Based on the feedback reward quantification data, the relationship strength and confidence of the corresponding knowledge in the dynamically updated medical knowledge graph are retrospectively adjusted to obtain a closed-loop optimized medical knowledge graph; wherein, when the decision execution feedback data represents the application effect of the corresponding decision scheme and meets the preset effect conditions, the relationship strength and confidence of the corresponding knowledge in the dynamically updated medical knowledge graph are enhanced. The closed-loop optimized medical knowledge graph is synchronously replaced with the dynamically updated medical knowledge graph and connected to the optometrist intelligent agent decision-making architecture to obtain continuously evolving knowledge graph service data.

[0072] Since the embodiments of the apparatus and the embodiments of the method correspond to each other, please refer to the description of the embodiments of the method for the embodiments of the apparatus, which will not be repeated here.

[0073] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed, can perform the steps provided in the above embodiments. The storage medium may include various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0074] This invention also provides a computer device, which may include a memory and a processor. The memory stores a computer program, and when the processor calls the computer program in the memory, it can implement the steps provided in the above embodiments. Of course, the computer device may also include various network interfaces, a power supply, a graphics card, etc., to utilize the graphics card's performance to operate the model, such as for inference and training.

[0075] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to in the method section. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of the claims of this application.

[0076] It should also be noted that, in this specification, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A method for dynamic integration of knowledge graphs, characterized in that, include: Multi-source heterogeneous raw data is acquired, cleaned and integrated, and then desensitized to obtain standardized preprocessed data. The multi-source heterogeneous raw data includes structured and unstructured data of ophthalmic patients. The structured data includes basic information about ophthalmic patients, diagnostic results, laboratory reports, medication records, and treatment plans. The unstructured data includes ward round records, nursing records, patient complaints, consultation transcripts, medical literature, clinical guidelines, and expert consensus. The standardized preprocessed data is subjected to medical named entity recognition according to preset recognition rules. After recognition, semantic relationship extraction processing between entities is performed to obtain medical entity relationship extraction data. The extracted medical entity relationship data is then subjected to entity alignment, relationship alignment, and knowledge fusion processing with a preset medical knowledge graph to obtain fused medical knowledge graph data. The fused medical knowledge graph data is dynamically evolved to obtain a dynamically updated medical knowledge graph; The dynamically updated medical knowledge graph is integrated into a pre-defined optometrist intelligent agent decision-making architecture for medical knowledge query, reasoning, and decision interpretation processing to obtain intelligent decision support data.

2. The knowledge graph dynamic integration method according to claim 1, characterized in that, The process of acquiring multi-source heterogeneous raw data, cleaning and integrating the multi-source heterogeneous raw data, and then performing de-identification preprocessing to obtain standardized preprocessed data includes: Structured and unstructured data are collected from multiple data source systems to obtain multi-source heterogeneous raw data; wherein, the data source systems include hospital information systems, electronic medical record systems, medical settlement systems, medical imaging systems, and wearable devices; The multi-source heterogeneous raw data are cleaned, deduplicated, standardized, and format converted to obtain standardized cleaned data. The scattered records belonging to the same object in the standardized cleaned data are linked and integrated in chronological order to obtain object timeline integrated data; wherein, the scattered records include medical records of the same eye patient at different times and in different departments, and the object timeline integrated data is patient medical timeline data linked and integrated in chronological order of medical visits; The unstructured text in the timeline integrated data of the object is processed by word segmentation, part-of-speech tagging and syntactic analysis to obtain text parsing and integrated data; The parsed and integrated text data is anonymized and / or pseudonymous to obtain standardized preprocessed data.

3. The knowledge graph dynamic integration method according to claim 1, characterized in that, The process involves performing medical named entity recognition on the standardized preprocessed data according to preset recognition rules, followed by semantic relationship extraction processing between entities to obtain medical entity relationship extraction data, including: The standardized preprocessed data is used to perform medical named entity recognition according to the sequence labeling model in the preset recognition rules to obtain medical candidate named entity data. The medical candidate named entity data is categorized and labeled according to preset medical entity types to obtain medical entity category labeling data; Based on the context of the medical entity category annotation data, the semantic relationships between each medical entity are identified and extracted to obtain medical entity semantic relationship data. The attribute information associated with each medical entity in the medical entity semantic relationship data is extracted and mounted to obtain medical entity attribute mounting data; wherein, the attribute information includes the dosage attribute and usage attribute of the drug entity; The medical entity attribute mounting data is organized and encapsulated according to the triple structure of entity, relation, and attribute to obtain medical entity relation extraction data.

4. The knowledge graph dynamic integration method according to claim 1, characterized in that, The step of performing entity alignment, relationship alignment, and knowledge fusion processing on the extracted medical entity relationship data and the preset medical knowledge graph to obtain fused medical knowledge graph data includes: The medical entities in the extracted medical entity relationship data are matched with the existing concepts in the medical knowledge graph according to string similarity, vector embedding similarity or ontology reasoning to obtain medical entity similarity matching data. The medical entities whose similarity in the medical entity similarity matching data meets a preset threshold are mapped and bound to the corresponding existing concepts to obtain medical entity alignment mapping data. Based on the medical entity alignment mapping data, the semantic relationships in the extracted medical entity relationship data are compared and aligned with the existing relationships in the medical knowledge graph to obtain medical relationship alignment comparison data. The confidence of existing relationships in the medical relationship alignment ratio data is updated, and non-existent relationships are marked as candidate new relationships to obtain medical relationship fusion labeling data; The medical relationship fusion marker data is arbitrated for knowledge conflict according to preset arbitration rules and then incorporated into the medical knowledge graph to obtain fused medical knowledge graph data; wherein, the preset arbitration rules include voting mechanism arbitration rules, expert review arbitration rules, or confidence-based arbitration rules.

5. The knowledge graph dynamic integration method according to claim 1, characterized in that, The step of dynamically evolving the fused medical knowledge graph data to obtain a dynamically updated medical knowledge graph includes: Statistical analysis was performed on the frequency of occurrence and co-occurrence patterns of each relationship in the fused medical knowledge graph data, as well as the corresponding clinical outcomes, to obtain statistical characteristic data of medical relationships; Based on the statistical feature data of medical relationships, the strength weights of corresponding relationships in the fused medical knowledge graph data are dynamically adjusted to obtain medical relationship strength adjustment graph data; The newly added medical entities and relationships in the medical relationship strength adjustment graph data are verified and confirmed, and the new knowledge that has passed the verification is incorporated into the graph, resulting in new medical knowledge incorporated into the graph data; The new medical knowledge is incorporated into the atlas data for knowledge conflict and inconsistency detection, and the detected conflicting knowledge is resolved to obtain medical conflict resolution atlas data; The medical conflict resolution graph data is managed by version recording based on update content, update time, and update source to obtain a dynamically updated medical knowledge graph.

6. The knowledge graph dynamic integration method according to claim 1, characterized in that, The process involves integrating the dynamically updated medical knowledge graph into a pre-defined optometrist intelligent agent decision-making architecture for medical knowledge query, reasoning, and decision interpretation, resulting in intelligent decision support data, including: The dynamically updated medical knowledge graph is connected to the preset optometrist intelligent agent decision architecture, and the knowledge query request initiated by the optometrist intelligent agent decision architecture is received to obtain the knowledge query request data. Based on the knowledge query request data, related knowledge retrieval and medical knowledge reasoning operations are performed in the dynamically updated knowledge graph to obtain knowledge reasoning result data; wherein, the medical knowledge reasoning operation includes reasoning corresponding candidate diseases based on patient symptoms, and / or recommending corresponding examination items and treatment plans based on diseases; The knowledge reasoning results are fed back to the optometrist's intelligent agent decision-making architecture to generate a decision scheme and obtain candidate decision scheme data. The decision path is traced to the supporting knowledge corresponding to the candidate decision scheme data in the dynamically updated medical knowledge graph to obtain decision interpretation path data; wherein, the supporting knowledge includes medical evidence and clinical guidelines that support the candidate decision scheme data in the dynamically updated medical knowledge graph. The candidate decision scheme data and the decision interpretation path data are associated, encapsulated, and output to obtain intelligent decision support data.

7. The knowledge graph dynamic integration method according to claim 1, characterized in that, After obtaining the intelligent decision support data, the process also includes: The execution results and feedback information generated by the intelligent decision support data in the actual application process are collected to obtain decision execution feedback data; The decision execution feedback data is quantified according to a preset reward mechanism to obtain feedback reward quantification data; Based on the feedback reward quantification data, the relationship strength and confidence of the corresponding knowledge in the dynamically updated medical knowledge graph are retrospectively adjusted to obtain a closed-loop optimized medical knowledge graph; wherein, when the decision execution feedback data represents the application effect of the corresponding decision scheme and meets the preset effect conditions, the relationship strength and confidence of the corresponding knowledge in the dynamically updated medical knowledge graph are enhanced. The closed-loop optimized medical knowledge graph is synchronously replaced with the dynamically updated medical knowledge graph and connected to the optometrist intelligent agent decision-making architecture to obtain continuously evolving knowledge graph service data.

8. A dynamic integration device for knowledge graphs, characterized in that, include: The data processing unit is used to acquire multi-source heterogeneous raw data, clean and integrate the multi-source heterogeneous raw data, and then perform desensitization preprocessing to obtain standardized preprocessed data. The multi-source heterogeneous raw data includes structured and unstructured data of ophthalmic patients. The structured data includes basic information of ophthalmic patients, diagnostic results, laboratory reports, medication records, and treatment plans. The unstructured data includes ward round records, nursing records, patient complaints, consultation dialogue texts, medical literature, clinical guidelines, and expert consensus. The relation extraction unit is used to perform medical named entity recognition on the standardized preprocessed data according to preset recognition rules, and then perform semantic relation extraction processing between entities to obtain medical entity relation extraction data. The graph fusion unit is used to perform entity alignment, relationship alignment and knowledge fusion processing on the extracted medical entity relationship data and the preset medical knowledge graph respectively to obtain fused medical knowledge graph data. The graph update unit is used to perform dynamic evolution processing on the fused medical knowledge graph data to obtain a dynamically updated medical knowledge graph. The decision reasoning unit is used to connect the dynamically updated medical knowledge graph to a preset optometrist intelligent agent decision architecture to perform medical knowledge query reasoning and decision interpretation processing to obtain intelligent decision support data.

9. A computer device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the knowledge graph dynamic integration method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the knowledge graph dynamic integration method as described in any one of claims 1 to 7.

Images