A clinical term knowledge graph construction method

By constructing a clinical knowledge graph, the problems of fragmentation and heterogeneity of multi-source medical data are solved, achieving deep data collaboration and the integrity of the knowledge system, and providing high-quality data support for intelligent healthcare and medical research.

CN122314433APending Publication Date: 2026-06-30JINAN BEISEN ELECTRONIC INFORMATION CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610199864.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-02-11
Publication Date
2026-06-30

Smart Images

  • Figure CN122314433A_ABST
    Figure CN122314433A_ABST
Patent Text Reader

Abstract

This invention relates to the field of data processing technology and specifically discloses a method for constructing a clinical terminology knowledge graph. The method includes: acquiring multi-source clinical data and medical domain dictionary data; determining the clinical entity set and data block dictionary set for each clinical data block of each medical clinical data from each data source; determining multiple business domain clustering data for each business domain label and the clustered medical entity set for each business domain clustering data; calculating the statistical entity relationship set and graph entity relationship set between every two medical entities in the clustered medical entity set of each business domain label; and constructing a clinical knowledge graph. This method enables deep collaboration between multi-source clinical data and medical dictionaries, improving the professionalism and reliability of medical entity associations, balancing the independence of business domain knowledge with the interoperability of global knowledge, and providing high-quality core support for intelligent medical assistance, medical research, and clinical knowledge transmission.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method for constructing a clinical terminology knowledge graph. Background Technology

[0002] With the development of medical informatization, a large amount of clinical data is digitized and stored in the form of electronic medical records, laboratory reports, and imaging records. This data is characterized by complex formats, diverse sources, and inconsistent semantics, making it difficult to directly use for knowledge computing and intelligent analysis. Current technologies typically employ medical named entity recognition and relation extraction methods to construct medical knowledge bases, but these often rely on single data sources or local domains, lacking cross-domain semantic fusion capabilities, resulting in fragmented knowledge structures and insufficient logical connections. Furthermore, traditional methods often struggle to achieve standardized terminology matching and relation mapping when processing heterogeneous medical data, failing to meet the in-depth application needs of diverse clinical scenarios.

[0003] Therefore, this invention proposes a method for constructing a clinical terminology knowledge graph. Summary of the Invention

[0004] This invention provides a method for constructing a clinical terminology knowledge graph. It acquires and analyzes multi-source clinical data and medical domain dictionary data, determining the clinical entity set and data block dictionary set for each clinical data block of each medical clinical data from each data source. It also determines multiple business domain clustering data for each business domain label and the clustered medical entity set for each business domain clustering data. Finally, it calculates the statistical entity relationship set and graph entity relationship set between every two medical entities in the clustered medical entity set of each business domain label, constructing a clinical knowledge graph. This method enables deep collaboration between multi-source clinical data and medical dictionaries, improving the professionalism and reliability of medical entity associations, balancing the independence of business domain knowledge with the interoperability of knowledge across the entire domain, and enhancing the integrity and consistency of the knowledge system. This provides high-quality core support for intelligent medical assistance, medical research, and the transmission of clinical knowledge.

[0005] This invention provides a method for constructing a clinical terminology knowledge graph, comprising:

[0006] S1: Obtain multi-source clinical data and medical domain dictionary data from multiple data sources. Based on the multi-source clinical data and medical domain dictionary data, determine the clinical entity set and data block dictionary set for each clinical data block of each medical clinical data from each data source.

[0007] S2: Based on the medical domain dictionary data, the clinical entity set of each clinical data block of each medical clinical data from each data source, and the data block dictionary set, determine the multiple business domain cluster data of each business domain label in the medical domain dictionary data, and the clustered medical entity set of each business domain cluster data;

[0008] S3: Based on multi-source clinical data and clustered medical entity sets of each business domain cluster data for each business domain label, calculate the statistical entity relationship set and graph entity relationship set between every two medical entities in the clustered medical entity set of each business domain cluster data for each business domain label;

[0009] S4: Construct a clinical knowledge graph based on the set of graph entity relationships between all medical entities in the clustered medical entity set of all business domain clustering data for each business domain label.

[0010] Preferably, a method for constructing a clinical terminology knowledge graph involves acquiring multi-source clinical data from multiple data sources and medical domain dictionary data, including:

[0011] Acquire medical clinical data from multiple data sources and the timestamp of each data source, and preprocess the medical clinical data from each data source.

[0012] The preprocessed medical clinical data from each data source is split into multiple clinical data blocks, and multi-source clinical data is determined based on all clinical data blocks of all medical clinical data from all data sources after preprocessing.

[0013] Obtain medical domain dictionary data, which includes a set of business medical entities for each business domain label. The business domain labels include disease-symptom, drug-dosage, surgery-anatomy, gene-mutation, nursing-operation, and adverse event. The set of business medical entities includes multiple medical entities.

[0014] Preferably, a method for constructing a clinical terminology knowledge graph, based on multi-source clinical data and medical domain dictionary data, determines the set of clinical entities for each clinical data block of each medical clinical data from each data source, including:

[0015] Extract the data sources of all medical clinical data from the multi-source clinical data to determine the set of data sources for the multi-source clinical data;

[0016] Entity identification is performed on each clinical data block of each medical clinical data from each data source in multi-source clinical data to determine the set of clinical entities for each clinical data block of each medical clinical data from each data source, wherein the set of clinical entities includes multiple clinical entities.

[0017] Preferably, a method for constructing a clinical terminology knowledge graph, which determines the data block dictionary set for each clinical data block of each medical clinical data from each data source, further includes:

[0018] For each clinical data block of each medical clinical data source, the set of clinical entities is matched with the set of business entities of each business domain label in the medical domain dictionary data to determine the set of clinical-business entities of each business domain label of each clinical data block of each medical clinical data source, as well as the medical entities in the set of clinical-business entities that match each clinical entity.

[0019] Remove empty sets or sets of clinical-business entities that contain only one entity;

[0020] Based on the clinical-business entity set of each business domain label after removing clinical data blocks of each medical clinical data from each data source, and the medical entities in the clinical-business entity set that match all clinical entities, determine the clinical-business dictionary set of each business domain label after removing clinical data blocks of each medical clinical data from each data source.

[0021] Based on the clinical-business dictionary set of each clinical data block of each medical clinical data from each data source after removing all business domain labels, the data block dictionary set of each clinical data block of each medical clinical data from each data source is determined, wherein the data block dictionary set includes multiple medical entities.

[0022] Preferably, a method for constructing a clinical terminology knowledge graph, based on medical domain dictionary data, the clinical entity set of each clinical data block of each medical clinical data from each data source, and the data block dictionary set, determines multiple business domain clustering data for each business domain label in the medical domain dictionary data, including:

[0023] Each clinical data block, clinical entity set, and data block dictionary set of each medical clinical data from each data source are input into the semantic relationship model. Based on the semantic relationship model, the entity relationship between every two medical entities in the data block dictionary set of each clinical data block of each medical clinical data from each data source is determined.

[0024] Based on each business domain label, the clinical-business dictionary set of all clinical data blocks of all medical clinical data from all data sources after removal is divided to determine the business domain clinical data of each business domain label. The business domain clinical data includes the clinical-business dictionary set of multiple clinical data blocks of multiple medical clinical data from multiple data sources.

[0025] Cluster analysis is performed on the clinical data of each business domain label based on all clinical-business dictionary sets in the clinical data of each business domain label to determine multiple business domain cluster data for each business domain label. Each business domain cluster data includes a clinical-business dictionary set of multiple clinical data blocks from multiple medical clinical data sources.

[0026] Preferably, a method for constructing a clinical terminology knowledge graph determines the clustered medical entity set for each business domain clustering data for each business domain label in the medical domain dictionary data, including:

[0027] Based on the clinical-business dictionary set of each clinical data block of each medical clinical data from each data source for each business domain clustering data of each business domain label and the entity relationships between all medical entities in the data block dictionary set of each clinical data block of each medical clinical data from each data source, determine the clinical relationship set of the clinical-business dictionary set of each clinical data block of each medical clinical data from each data source for each business domain clustering data of each business domain label, wherein the clinical relationship set includes entity relationships between multiple medical entities;

[0028] Extract all medical entities from the clinical-business dictionary set of all clinical data blocks of all medical clinical data from all data sources in each business domain cluster data of each business domain label, and determine the clustered medical entity set of each business domain cluster data of each business domain label.

[0029] Preferably, a method for constructing a clinical terminology knowledge graph, based on multi-source clinical data and clustered medical entity sets of clustered data for each business domain label, calculates the statistical entity relationship set and the graph entity relationship set between every two medical entities in the clustered medical entity sets of each business domain label, including:

[0030] Based on the clinical relationship set of the data source set of multi-source clinical data, the clustered medical entity set of each business domain clustered data of each business domain label, and the clinical relationship set of all clinical data blocks of all medical clinical data from all data sources of each business domain label, calculate the statistical entity relationship set between every two medical entities in the clustered medical entity set of each business domain clustered data of each business domain label, and the confidence value of each entity relationship in the statistical entity relationship set.

[0031] Based on the clustered medical entity set of each business domain clustered data for each business domain label, the statistical entity relationship set between all medical entities in the clustered medical entity set, and the confidence value of each entity relationship in the statistical entity relationship set, calculate the graph entity relationship set between every two medical entities in the clustered medical entity set of each business domain clustered data for each business domain label.

[0032] Preferably, a method for constructing a clinical terminology knowledge graph constructs a clinical knowledge graph based on the set of graph entity relationships between all medical entities in the clustered medical entity set of all business domain clustering data for each business domain label, including:

[0033] Based on the set of graph entity relationships between every two medical entities in the clustered medical entity set of each business domain clustered data for each business domain label, multiple triplet knowledge units are determined for every two medical entities in the clustered medical entity set of each business domain clustered data for each business domain label.

[0034] Based on all triplet knowledge units of all medical entities in the clustered medical entity set of each business domain label, a clustered knowledge graph of each business domain label is constructed.

[0035] Based on the clustering knowledge graph of all business domain clustering data for each business domain label, construct a business domain knowledge graph for each business domain label.

[0036] Construct a clinical knowledge graph based on the business domain knowledge graph with all business domain labels.

[0037] The beneficial effects of this invention compared to existing technologies are as follows: It acquires and analyzes multi-source clinical data and medical domain dictionary data, determines the clinical entity set and data block dictionary set for each clinical data block of each medical clinical data from each data source, determines multiple business domain clustering data for each business domain label, and the clustered medical entity set for each business domain clustering data, calculates the statistical entity relationship set and graph entity relationship set between every two medical entities in the clustered medical entity set of each business domain clustering data for each business domain label, and constructs a clinical knowledge graph. This enables deep collaboration between multi-source clinical data and medical dictionaries, improves the professionalism and reliability of medical entity associations, balances the independence of business domain knowledge with the interoperability of global knowledge, enhances the integrity and consistency of the knowledge system, and provides high-quality core support for intelligent medical assistance, medical research, and clinical knowledge inheritance.

[0038] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0039] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0040] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0041] Figure 1 This is a flowchart of a method for constructing a clinical terminology knowledge graph according to an embodiment of the present invention. Detailed Implementation

[0042] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0043] Example 1:

[0044] This invention provides a method for constructing a clinical terminology knowledge graph, referencing Figure 1 ,include:

[0045] S1: Obtain multi-source clinical data and medical domain dictionary data from multiple data sources. Based on the multi-source clinical data and medical domain dictionary data, determine the clinical entity set and data block dictionary set for each clinical data block of each medical clinical data from each data source.

[0046] S2: Based on the medical domain dictionary data, the clinical entity set of each clinical data block of each medical clinical data from each data source, and the data block dictionary set, determine the multiple business domain cluster data of each business domain label in the medical domain dictionary data, and the clustered medical entity set of each business domain cluster data;

[0047] S3: Based on multi-source clinical data and clustered medical entity sets of each business domain cluster data for each business domain label, calculate the statistical entity relationship set and graph entity relationship set between every two medical entities in the clustered medical entity set of each business domain cluster data for each business domain label;

[0048] S4: Construct a clinical knowledge graph based on the set of graph entity relationships between all medical entities in the clustered medical entity set of all business domain clustering data for each business domain label.

[0049] In this embodiment, two types of core data are acquired: multi-source clinical data from multiple data sources and medical domain dictionary data. These two types of data are combined for subsequent processing. For each piece of medical clinical data from each data source, corresponding clinical data blocks are extracted. Then, based on the two types of core data, each clinical data block is analyzed one by one, ultimately determining the set of clinical entities corresponding to each clinical data block, and simultaneously determining the set of data block dictionaries corresponding to each clinical data block.

[0050] In this embodiment, the medical domain dictionary data is combined with the clinical entity set of each clinical data block for each medical clinical data from each data source, and the data block dictionary set of each clinical data block for each medical clinical data from each data source. First, all business domain tags in the medical domain dictionary data are sorted out. Then, each business domain tag is sorted and categorized one by one to determine the multiple business domain cluster data corresponding to each business domain tag. Simultaneously, the medical entities contained in each business domain cluster data are sorted out, and finally, the exclusive clustered medical entity set for each business domain cluster data is determined.

[0051] In this embodiment, calculations are performed based on multi-source clinical data and the clustered medical entity sets corresponding to each business domain clustering data for each business domain label. For each business domain clustering data for each business domain label, the focus is on its clustered medical entity set. Every two medical entities in the set are analyzed, and the statistical entity relationship set between every two medical entities is first calculated. Then, based on the statistical entity relationship set, further analysis and refinement are performed to obtain the corresponding graph entity relationship set between every two medical entities.

[0052] In this embodiment, all business domain clustering data under all business domain labels are aggregated, the clustered medical entity sets of each business domain clustering data are sorted out, and the corresponding graph entity relationship sets between all medical entities in the sets are extracted. All business domain labels and the graph entity relationship sets corresponding to all business domain clustering data are comprehensively integrated, establishing knowledge connections between different business domains and different clustering data, ultimately completing the overall construction of the clinical knowledge graph.

[0053] The beneficial effects of the above technologies are as follows: They acquire and analyze multi-source clinical data and medical domain dictionary data; determine the clinical entity set and data block dictionary set for each clinical data block of each medical clinical data from each data source; determine multiple business domain clustering data for each business domain label and the clustered medical entity set for each business domain clustering data; calculate the statistical entity relationship set and graph entity relationship set between every two medical entities in the clustered medical entity set of each business domain label; and construct a clinical knowledge graph. This enables deep collaboration between multi-source clinical data and medical dictionaries, improving the professionalism and reliability of medical entity associations, balancing the independence of business domain knowledge with the interoperability of global knowledge, and enhancing the integrity and consistency of the knowledge system. This provides high-quality core support for intelligent medical assistance, medical research, and the transmission of clinical knowledge.

[0054] Example 2:

[0055] Based on Example 1, a method for constructing a clinical terminology knowledge graph acquires multi-source clinical data from multiple data sources and medical domain dictionary data, including:

[0056] Acquire medical clinical data from multiple data sources and the timestamp of each data source, and preprocess the medical clinical data from each data source.

[0057] The preprocessed medical clinical data from each data source is split into multiple clinical data blocks, and multi-source clinical data is determined based on all clinical data blocks of all medical clinical data from all data sources after preprocessing.

[0058] Obtain medical domain dictionary data, which includes a set of business medical entities for each business domain label. The business domain labels include disease-symptom, drug-dosage, surgery-anatomy, gene-mutation, nursing-operation, and adverse event. The set of business medical entities includes multiple medical entities.

[0059] In this embodiment, medical clinical data from multiple different sources is acquired, and the timestamp of each record is simultaneously obtained to ensure the complete retention of the time information for each data point. After acquisition, preprocessing operations are performed on all medical clinical data corresponding to each data source. Preprocessing includes several core aspects, specifically: removing redundant data with duplicate entries, correcting spelling errors in terminology, deleting meaningless garbled and invalid information, standardizing the data format of different data sources, and cleaning up missing and outlier values. Through these operations, the raw data is brought to a standardized and usable state.

[0060] In this embodiment, the preprocessed medical clinical data from each data source is split based on clinical semantic themes and information relevance. Independent information modules are divided according to clinical business logic; for example, a complete medical record is split into symptom descriptions, medication records, examination results, surgical records, and nursing records. This ensures that each split clinical data block has a single, semantically coherent theme and does not contain cross-theme mixed information. After splitting, all clinical data blocks split from all preprocessed medical clinical data from all data sources are integrated to determine multi-source clinical data.

[0061] In this embodiment, medical domain dictionary data is acquired, which serves as the core standard for subsequent clinical entity matching and business domain classification. The medical domain dictionary data contains a fixed core structure, encompassing pre-defined sets of business medical entities corresponding to various business domain labels. These business domain labels explicitly include six categories: disease-symptom, drug-dosage, surgery-anatomy, gene-mutation, nursing-procedure, and adverse events. Each business domain label corresponds to a unique set of business medical entities, and each set contains all core medical entities within that business domain. All medical entities conform to authoritative clinical standards.

[0062] The beneficial effects of the above technologies are: acquiring multi-source clinical data and medical domain dictionary data from multiple data sources can organize multi-source heterogeneous clinical data and provide data support for the clinical entity set of each clinical data block of each medical clinical data from each data source.

[0063] Example 3:

[0064] Based on Example 2, a method for constructing a clinical terminology knowledge graph, based on multi-source clinical data and medical domain dictionary data, determines the set of clinical entities for each clinical data block of each medical clinical data from each data source, including:

[0065] Extract the data sources of all medical clinical data from the multi-source clinical data to determine the set of data sources for the multi-source clinical data;

[0066] Entity identification is performed on each clinical data block of each medical clinical data from each data source in multi-source clinical data to determine the set of clinical entities for each clinical data block of each medical clinical data from each data source, wherein the set of clinical entities includes multiple clinical entities.

[0067] In this embodiment, data source extraction is performed on the integrated multi-source clinical data. The extraction target is all medical clinical data in the multi-source clinical data, and the core is to extract the original data source information corresponding to each medical clinical data. After extraction, all extracted data source information is integrated, duplicate data source names are removed, and a complete data source set of the multi-source clinical data is formed. The set clearly presents all the data sources included in the multi-source clinical data.

[0068] In this embodiment, entity recognition is performed on each clinical data block after the medical clinical data from each data source is split. The core of the recognition is to screen out key information with actual medical significance within the data block, extract it, and transform it into corresponding clinical entities. The recognition process must adhere to clinical professional standards to ensure that the extracted entities conform to medical terminology standards. After the recognition is completed, a corresponding set of clinical entities is determined for each clinical data block of each medical clinical data from each data source. Each set summarizes all the extracted clinical entities from that data block, and each set of clinical entities contains multiple interrelated clinical entities.

[0069] The beneficial effects of the above technologies are as follows: Based on multi-source clinical data and medical domain dictionary data, the set of clinical entities for each clinical data block of each medical clinical data from each data source can be determined, which can improve the accuracy of entity extraction, avoid entity confusion across data scenarios, and further provide accurate data support for determining the dictionary set of data blocks.

[0070] Example 4:

[0071] Building upon Example 3, a method for constructing a clinical terminology knowledge graph, which determines the data block dictionary set for each clinical data block of each medical clinical data from each data source, further includes:

[0072] For each clinical data block of each medical clinical data source, the set of clinical entities is matched with the set of business entities of each business domain label in the medical domain dictionary data to determine the set of clinical-business entities of each business domain label of each clinical data block of each medical clinical data source, as well as the medical entities in the set of clinical-business entities that match each clinical entity.

[0073] Remove empty sets or sets of clinical-business entities that contain only one entity;

[0074] Based on the clinical-business entity set of each business domain label after removing clinical data blocks of each medical clinical data from each data source, and the medical entities in the clinical-business entity set that match all clinical entities, determine the clinical-business dictionary set of each business domain label after removing clinical data blocks of each medical clinical data from each data source.

[0075] Based on the clinical-business dictionary set of each clinical data block of each medical clinical data from each data source after removing all business domain labels, the data block dictionary set of each clinical data block of each medical clinical data from each data source is determined, wherein the data block dictionary set includes multiple medical entities.

[0076] In this embodiment, clinical entities and business domain entities are matched. The core objects of the matching are divided into two categories: one is the set of clinical entities corresponding to each clinical data block of each medical clinical data from each data source, and the other is the set of business entities corresponding to each business domain label in the medical domain dictionary data. During the matching, the entities in the two sets need to be accurately compared one by one. After the matching is completed, two core results need to be clarified: first, the set of clinical-business entities corresponding to each business domain label for each clinical data block of each medical clinical data from each data source; second, the corresponding medical entity matched by each clinical entity in the set of clinical-business entities, so as to ensure that each clinical entity can find a standardized business domain medical entity reference.

[0077] In this embodiment, two types of clinical-business entity sets are eliminated: one is an empty set with no entities at all, and the other is a set containing only one entity. These sets do not have value for subsequent entity relationship mining, and the elimination operation does not involve any other sets that meet the requirements. This operation filters out meaningless data.

[0078] In this embodiment, for each clinical data block of each medical clinical data from each data source, after the removal operation, all the medical entities matched by the clinical entities in the clinical-business entity set corresponding to each business domain label are summarized in the order of all clinical entities in the clinical-business entity set to determine the clinical-business dictionary set for each clinical data block of each medical clinical data from each data source after the removal operation.

[0079] In this embodiment, all medical entities in the clinical-business dictionary set of all business domain labels after removing all clinical data blocks of each medical clinical data from each data source are summarized. After removing duplicate medical entities, the data block dictionary set corresponding to each clinical data block of each medical clinical data from each data source is determined. This set is a summary of standardized medical entities after matching all business domains under the data block, which includes multiple medical entities that conform to medical standards and have completed business domain matching.

[0080] The beneficial effects of the above technologies are as follows: by determining the data block dictionary set for each clinical data block of each medical clinical data from each data source, it is possible to achieve accurate matching between clinical entities and standardized business domain entities, providing a high-quality core carrier for entity relationship mining, business domain clustering and knowledge graph construction, and improving the efficiency and accuracy of medical knowledge mining.

[0081] Example 5:

[0082] Based on Example 4, a method for constructing a clinical terminology knowledge graph, based on medical domain dictionary data, the clinical entity set of each clinical data block of each medical clinical data from each data source, and the data block dictionary set, determines multiple business domain clustering data for each business domain label in the medical domain dictionary data, including:

[0083] Each clinical data block, clinical entity set, and data block dictionary set of each medical clinical data from each data source are input into the semantic relationship model. Based on the semantic relationship model, the entity relationship between every two medical entities in the data block dictionary set of each clinical data block of each medical clinical data from each data source is determined.

[0084] Based on each business domain label, the clinical-business dictionary set of all clinical data blocks of all medical clinical data from all data sources after removal is divided to determine the business domain clinical data of each business domain label. The business domain clinical data includes the clinical-business dictionary set of multiple clinical data blocks of multiple medical clinical data from multiple data sources.

[0085] Cluster analysis is performed on the clinical data of each business domain label based on all clinical-business dictionary sets in the clinical data of each business domain label to determine multiple business domain cluster data for each business domain label. Each business domain cluster data includes a clinical-business dictionary set of multiple clinical data blocks from multiple medical clinical data sources.

[0086] In this embodiment, each clinical data block from each medical clinical data source, the corresponding set of clinical entities for each clinical data block, and the corresponding data block dictionary set are input into the semantic relationship model. The semantic relationship model identifies potential semantic connections between different medical entities. By analyzing the contextual meaning of the data blocks corresponding to medical entities, the model determines the type of semantic relationship between them. The model analyzes each input data combination, sorting out the associations between every two medical entities in the data block dictionary set corresponding to each clinical data block, and determining the corresponding entity relationships. For example, if the data block dictionary set contains two entities, aspirin and coronary heart disease, the relationship between them is determined to be treatment based on the description of the data block.

[0087] In this embodiment, the removed multi-source clinical data is segmented based on each business domain label to determine the business domain clinical data corresponding to each business domain label. The segmentation process involves integrating clinical business dictionary sets belonging to the same business domain together to form structured business domain clinical data.

[0088] In this embodiment, cluster analysis is performed on all clinical business dictionary sets based on the clinical data corresponding to each business domain label. The purpose of clustering is to discover potential patterns and semantic similarities between medical entities within the same business domain. Clinical business dictionary sets with similar semantics and close information association are grouped into the same category, while clinical business dictionary sets with large semantic differences and low information association are grouped into different categories. Based on the classification results, multiple business domain cluster data are determined for each business domain label. Each business domain cluster data contains clinical business dictionary sets corresponding to multiple clinical data blocks from multiple medical clinical data sources.

[0089] The beneficial effects of the above technologies are as follows: Based on medical domain dictionary data, the clinical entity set of each clinical data block of each medical clinical data from each data source, and the data block dictionary set, multiple business domain clustering data of each business domain label in the medical domain dictionary data can be determined, which can improve the accuracy of entity relationship identification and the consistency of medical logic, and provide high-quality data support for entity relationship integration and knowledge graph construction.

[0090] Example 6:

[0091] Based on Example 5, a method for constructing a clinical terminology knowledge graph determines the clustered medical entity set of each business domain clustering data for each business domain label in the medical domain dictionary data, including:

[0092] Based on the clinical-business dictionary set of each clinical data block of each medical clinical data from each data source for each business domain clustering data of each business domain label and the entity relationships between all medical entities in the data block dictionary set of each clinical data block of each medical clinical data from each data source, determine the clinical relationship set of the clinical-business dictionary set of each clinical data block of each medical clinical data from each data source for each business domain clustering data of each business domain label, wherein the clinical relationship set includes entity relationships between multiple medical entities;

[0093] Extract all medical entities from the clinical-business dictionary set of all clinical data blocks of all medical clinical data from all data sources in each business domain cluster data of each business domain label, and determine the clustered medical entity set of each business domain cluster data of each business domain label.

[0094] In this embodiment, two types of data are combined. The first type is the clinical business dictionary set corresponding to each clinical data block of each medical clinical data from each data source under each business domain label and each business domain cluster data. The second type is the data block dictionary set of each clinical data block of each medical clinical data from each data source, containing the established entity relationships between all medical entities. The clinical business dictionary set of each clinical data block of each medical clinical data from each data source under each business domain label and each business domain cluster data is systematically reviewed, and the corresponding entity relationships are matched. Finally, a unique set of clinical relationships is determined for each set of clinical business dictionaries, which contains entity relationships between multiple medical entities.

[0095] In this embodiment, cluster data for each business domain label is extracted. Specifically, this involves extracting the clinical business dictionary set corresponding to all clinical data blocks from all data sources within that cluster. All medical entities in these clinical business dictionary sets are extracted one by one, and duplicate entities are removed during the extraction process to avoid multiple entries of the same entity. All extracted and deduplicated medical entities are then uniformly summarized and integrated to ultimately determine the exclusive clustered medical entity set for each business domain label and its cluster data.

[0096] The beneficial effects of the above technologies are as follows: determining the clustered medical entity set of each business domain clustering data for each business domain label in the medical domain dictionary data can achieve accurate adaptation between clinical relationships and business scenarios, ensure the professionalism and logic of medical entity associations, and improve the usability of relationship data.

[0097] Example 7:

[0098] Based on Example 6, a method for constructing a clinical terminology knowledge graph, based on multi-source clinical data and clustered medical entity sets of clustered data for each business domain label, calculates the statistical entity relationship set and the graph entity relationship set between every two medical entities in the clustered medical entity sets of each business domain label, including:

[0099] Based on the clinical relationship set of the data source set of multi-source clinical data, the clustered medical entity set of each business domain clustered data of each business domain label, and the clinical relationship set of all clinical data blocks of all medical clinical data from all data sources of each business domain label, calculate the statistical entity relationship set between every two medical entities in the clustered medical entity set of each business domain clustered data of each business domain label, and the confidence value of each entity relationship in the statistical entity relationship set.

[0100] Based on the clustered medical entity set of each business domain clustered data for each business domain label, the statistical entity relationship set between all medical entities in the clustered medical entity set, and the confidence value of each entity relationship in the statistical entity relationship set, calculate the graph entity relationship set between every two medical entities in the clustered medical entity set of each business domain clustered data for each business domain label.

[0101] In this embodiment, based on the data source set of multi-source clinical data, the clustered medical entity set of each business domain clustered data for each business domain label, and the clinical relationship set of all clinical data blocks of all medical clinical data from all data sources for each business domain clustered data for each business domain label, the statistical entity relationship set between every two medical entities in the clustered medical entity set of each business domain clustered data for each business domain label, and the confidence value of each entity relationship in the statistical entity relationship set are calculated. The calculation formula is expressed as follows:

[0102] ;

[0103] ;

[0104] ;

[0105] ;

[0106] in, This represents the m-th and n-th medical entities in the clustered medical entity set of the j-th business domain clustering data labeled with the i-th business domain. Let Nij represent the r-th and s-th medical entities in the clinical-business dictionary set of the p-th medical clinical data block of the k-th data source of the j-th business domain cluster data of the i-th business domain label. Nij represents the number of data sources in the data source set, and NijN2 represents the number of medical clinical data in the k-th data source of the j-th business domain cluster data of the i-th business domain label. The number of clinical data blocks of the p-th medical clinical data from the k-th data source of the j-th business domain cluster data of the i-th business domain label. This represents the set of statistical entity relationships between the m-th and n-th medical entities in the clustered medical entity set of the j-th business domain clustering data with the i-th business domain label. This represents the a-th data source in the set of multi-source clinical data sources. This represents the k-th data source for the cluster data of the j-th business domain labeled the i-th business domain. This represents the data source weight of the a-th data source in the data source set. This represents the data source weight of the k-th data source in the cluster data of the j-th business domain labeled the i-th business domain. Indicates the time decay factor. This represents the timestamp of the p-th medical clinical data record from the k-th data source of the j-th business domain cluster data with the i-th business domain label. Indicates the current time. This represents the entity relationship between the r-th and s-th medical entities in the clinical-business dictionary set of the p-th medical clinical data block of the k-th data source of the j-th business domain cluster data labeled i-th business domain. This represents the b-th entity relationship in the statistical entity relationship set between the m-th and n-th medical entities in the clustered medical entity set of the j-th business domain clustering data with the i-th business domain label, and the second indicator function based on the clinical-business dictionary set of the p-th medical clinical data from the k-th data source. Let represent the confidence value of the b-th entity relationship in the statistical entity relationship set between the m-th and n-th medical entities in the j-th business domain cluster data of the i-th business domain label, within the business domain cluster data. This represents the b-th entity relationship in the statistical entity relationship set between the m-th and n-th medical entities in the clustered medical entity set of the j-th business domain label of the i-th business domain. This represents the set of clinical relationships in the set of clinical-business dictionaries for the qth clinical data block of the qth medical clinical data from the kth data source of the jth business domain cluster data of the ith business domain label.

[0107] In this embodiment, based on the clustered medical entity set of each business domain clustered data for each business domain label, the statistical entity relationship set between all medical entities in the clustered medical entity set, and the confidence value of each entity relationship in the statistical entity relationship set, the graph entity relationship set between every two medical entities in the clustered medical entity set of each business domain clustered data for each business domain label is calculated. The calculation formula is expressed as follows:

[0108] ;

[0109] ;

[0110] ;

[0111] ;

[0112] ;

[0113] in, Let represent the set of graph entity relationships between the m-th and n-th medical entities in the clustered medical entity set of the j-th business domain clustered data with the i-th business domain label. Let ijmnN4 represent the number of entity relationships in the statistical entity relationship set between the m-th and n-th medical entities in the clustered medical entity set of the j-th business domain clustered data with the i-th business domain label. This represents the c-th entity relationship in the statistical entity relationship set between the m-th and n-th medical entities in the clustered medical entity set of the j-th business domain clustering data with the i-th business domain label. Let represent the confidence value of the c-th entity relationship in the statistical entity relationship set between the m-th and n-th medical entities in the j-th business domain cluster data of the i-th business domain label, within the business domain cluster data. Let f(x) represent the conflict indicator function for the relationship between the m-th and n-th medical entities in the set of statistical entity relationships between the m-th and n-th medical entities in the set of clustered medical entities in the j-th business domain of the i-th business domain label, and the relationship between the b-th and c-th entity relationships in the set of statistical entity relationships between them. Indicates the conflict penalty coefficient. This represents the relationship strength value of the b-th entity relationship in the statistical entity relationship set between the m-th and n-th medical entities in the medical entity set of the j-th business domain cluster data with the i-th business domain label. These represent the relationship conflict value and relationship similarity value, respectively, of the m-th medical entity in the medical entity set and the b-th entity in the statistical entity relationship set between the j-th medical entity in the j-th business domain cluster data of the i-th business domain label. This represents the semantic similarity value of the m-th and n-th entity relationships in the statistical entity relationship set between the m-th and n-th medical entities in the j-th business domain cluster data of the i-th business domain label, and the b-th and c-th entity relationships in the statistical entity relationship set. This represents the threshold for relation strength.

[0114] In this embodiment, the relationship strength threshold The value range is 0.3-0.7, and it can be 0.4.

[0115] In this embodiment, the conflict penalty coefficient The value range is 0-1, and it can be 0.5.

[0116] In this embodiment, The conflict factor represents the conflict factor between the m-th and n-th medical entities in the set of statistical entity relationships between the m-th and n-th medical entities in the set of clustered data of the j-th business domain labeled i-th business domain.

[0117] In this embodiment, This represents the weighted data weight of all medical clinical data from all data sources for the clustering data of the j-th business domain with the i-th business domain label.

[0118] In this embodiment, The weighted data weight represents the set of clinical relationships of all clinical data blocks from all data sources, including the statistical entity relationship set between the m-th and n-th medical entities in the clustered medical entity set of the i-th business domain label.

[0119] The beneficial effects of the above technologies are as follows: Based on multi-source clinical data and clustered medical entity sets of each business domain clustered data for each business domain label, the statistical entity relationship set and graph entity relationship set between every two medical entities in the clustered medical entity set of each business domain clustered data for each business domain label can be calculated. This can accurately screen high-value associations, filter low-confidence invalid relationships, and improve the quality of relational data.

[0120] Example 8:

[0121] Based on Example 7, a method for constructing a clinical terminology knowledge graph constructs a clinical knowledge graph based on the set of graph entity relationships between all medical entities in the clustered medical entity set of all business domain clustering data for each business domain label, including:

[0122] Based on the set of graph entity relationships between every two medical entities in the clustered medical entity set of each business domain clustered data for each business domain label, multiple triplet knowledge units are determined for every two medical entities in the clustered medical entity set of each business domain clustered data for each business domain label.

[0123] Based on all triplet knowledge units of all medical entities in the clustered medical entity set of each business domain label, a clustered knowledge graph of each business domain label is constructed.

[0124] Based on the clustering knowledge graph of all business domain clustering data for each business domain label, construct a business domain knowledge graph for each business domain label.

[0125] Construct a clinical knowledge graph based on the business domain knowledge graph with all business domain labels.

[0126] In this embodiment, based on each entity relationship in the graph entity relationship set between every two medical entities in the clustered medical entity set of each business domain label, a triplet knowledge unit is determined for every two medical entities. A triplet knowledge unit is a knowledge representation structure composed of a medical subject, an entity relationship, and another medical subject. For example, if two medical entities are pneumonia and cough, and the corresponding graph entity relationship set includes clinical manifestations and comorbidities, then the triplet knowledge units are determined as pneumonia-clinical manifestations-cough and pneumonia-comorbidities-cough. This process is repeated to determine multiple triplet knowledge units for every two medical entities.

[0127] In this embodiment, the cluster data of each business domain for each business domain label is sorted out, and all triple knowledge units corresponding to all medical entities in the clustered medical entity set of the cluster data are summarized. These triple knowledge units are then connected and integrated according to the entity association logic to construct an organically related knowledge network, and finally a clustering knowledge graph exclusive to each business domain cluster data for each business domain label is formed.

[0128] In this embodiment, all clustered knowledge graphs under the same business domain label are aggregated and integrated, the inherent connections between each clustered knowledge graph are made up, the knowledge barriers between clusters are eliminated, and a unified and complete knowledge network system is formed. Finally, a business domain knowledge graph exclusive to each business domain label is constructed, which covers all the core clinical knowledge of the corresponding business domain.

[0129] In this embodiment, the knowledge graphs of all business domains are fully integrated to connect the knowledge between different business domains, realize the interoperability and fusion of knowledge from multiple business domains such as disease symptoms, drug dosage, surgery and anatomy, and form a hierarchical and logically coherent full-scale knowledge network. Finally, the overall construction of the clinical knowledge graph is completed. This graph covers the core clinical knowledge of the entire field and realizes the unified carrying of knowledge from multiple business domains.

[0130] The beneficial effects of the above technologies are as follows: Based on the set of graph entity relationships between all medical entities in the clustered medical entity set of all business domain clustering data for each business domain label, a clinical knowledge graph can be constructed, which can achieve semantic unity of multi-source data and take into account both knowledge association within the business domain and cross-domain knowledge fusion.

[0131] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for constructing a clinical terminology knowledge graph, characterized in that, include: S1: Obtain multi-source clinical data and medical domain dictionary data from multiple data sources. Based on the multi-source clinical data and medical domain dictionary data, determine the clinical entity set and data block dictionary set for each clinical data block of each medical clinical data from each data source. S2: Based on the medical domain dictionary data, the clinical entity set of each clinical data block of each medical clinical data from each data source, and the data block dictionary set, determine the multiple business domain cluster data of each business domain label in the medical domain dictionary data, and the clustered medical entity set of each business domain cluster data; S3: Based on multi-source clinical data and clustered medical entity sets of each business domain cluster data for each business domain label, calculate the statistical entity relationship set and graph entity relationship set between every two medical entities in the clustered medical entity set of each business domain cluster data for each business domain label; S4: Construct a clinical knowledge graph based on the set of graph entity relationships between all medical entities in the clustered medical entity set of all business domain clustering data for each business domain label.

2. The method for constructing a clinical terminology knowledge graph according to claim 1, characterized in that, Acquire multi-source clinical data and medical domain dictionary data from multiple data sources, including: Acquire medical clinical data from multiple data sources and the timestamp of each data source, and preprocess the medical clinical data from each data source. The preprocessed medical clinical data from each data source is split into multiple clinical data blocks, and multi-source clinical data is determined based on all clinical data blocks of all medical clinical data from all data sources after preprocessing. Obtain medical domain dictionary data, which includes a set of business medical entities for each business domain label. The business domain labels include disease-symptom, drug-dosage, surgery-anatomy, gene-mutation, nursing-operation, and adverse event. The set of business medical entities includes multiple medical entities.

3. The method for constructing a clinical terminology knowledge graph according to claim 2, characterized in that, Based on multi-source clinical data and medical domain dictionary data, the set of clinical entities for each clinical data block of each medical clinical data from each data source is determined, including: Extract the data sources of all medical clinical data from the multi-source clinical data to determine the set of data sources for the multi-source clinical data; Entity identification is performed on each clinical data block of each medical clinical data from each data source in multi-source clinical data to determine the set of clinical entities for each clinical data block of each medical clinical data from each data source, wherein the set of clinical entities includes multiple clinical entities.

4. The method for constructing a clinical terminology knowledge graph according to claim 3, characterized in that, The data block dictionary set that determines each clinical data block of each medical clinical data from each data source also includes: For each clinical data block of each medical clinical data source, the set of clinical entities is matched with the set of business entities of each business domain label in the medical domain dictionary data to determine the set of clinical-business entities of each business domain label of each clinical data block of each medical clinical data source, as well as the medical entities in the set of clinical-business entities that match each clinical entity. Remove empty sets or sets of clinical-business entities that contain only one entity; Based on the clinical-business entity set of each business domain label after removing clinical data blocks of each medical clinical data from each data source, and the medical entities in the clinical-business entity set that match all clinical entities, determine the clinical-business dictionary set of each business domain label after removing clinical data blocks of each medical clinical data from each data source. Based on the clinical-business dictionary set of each clinical data block of each medical clinical data from each data source after removing all business domain labels, the data block dictionary set of each clinical data block of each medical clinical data from each data source is determined, wherein the data block dictionary set includes multiple medical entities.

5. The method for constructing a clinical terminology knowledge graph according to claim 4, characterized in that, Based on medical domain dictionary data, the clinical entity set of each clinical data block for each medical clinical data source, and the data block dictionary set, multiple business domain clustering data for each business domain label in the medical domain dictionary data are determined, including: Each clinical data block, clinical entity set, and data block dictionary set of each medical clinical data from each data source are input into the semantic relationship model. Based on the semantic relationship model, the entity relationship between every two medical entities in the data block dictionary set of each clinical data block of each medical clinical data from each data source is determined. Based on each business domain label, the clinical-business dictionary set of all clinical data blocks of all medical clinical data from all data sources after removal is divided to determine the business domain clinical data of each business domain label. The business domain clinical data includes the clinical-business dictionary set of multiple clinical data blocks of multiple medical clinical data from multiple data sources. Cluster analysis is performed on the clinical data of each business domain label based on all clinical-business dictionary sets in the clinical data of each business domain label to determine multiple business domain cluster data for each business domain label. Each business domain cluster data includes a clinical-business dictionary set of multiple clinical data blocks from multiple medical clinical data sources.

6. The method for constructing a clinical terminology knowledge graph according to claim 5, characterized in that, Determine the clustered medical entity set for each business domain cluster data for each business domain label in the medical domain dictionary data, including: Based on the clinical-business dictionary set of each clinical data block of each medical clinical data from each data source for each business domain clustering data of each business domain label and the entity relationships between all medical entities in the data block dictionary set of each clinical data block of each medical clinical data from each data source, determine the clinical relationship set of the clinical-business dictionary set of each clinical data block of each medical clinical data from each data source for each business domain clustering data of each business domain label, wherein the clinical relationship set includes entity relationships between multiple medical entities; Extract all medical entities from the clinical-business dictionary set of all clinical data blocks of all medical clinical data from all data sources in each business domain cluster data of each business domain label, and determine the clustered medical entity set of each business domain cluster data of each business domain label.

7. The method for constructing a clinical terminology knowledge graph according to claim 6, characterized in that, Based on multi-source clinical data and clustered medical entity sets for each business domain of each business domain label, the statistical entity relationship set and graph entity relationship set between every two medical entities in the clustered medical entity set for each business domain of each business domain label are calculated, including: Based on the clinical relationship set of the data source set of multi-source clinical data, the clustered medical entity set of each business domain clustered data of each business domain label, and the clinical relationship set of all clinical data blocks of all medical clinical data from all data sources of each business domain label, calculate the statistical entity relationship set between every two medical entities in the clustered medical entity set of each business domain clustered data of each business domain label, and the confidence value of each entity relationship in the statistical entity relationship set. Based on the clustered medical entity set of each business domain clustered data for each business domain label, the statistical entity relationship set between all medical entities in the clustered medical entity set, and the confidence value of each entity relationship in the statistical entity relationship set, calculate the graph entity relationship set between every two medical entities in the clustered medical entity set of each business domain clustered data for each business domain label.

8. The method for constructing a clinical terminology knowledge graph according to claim 7, characterized in that, Based on the set of graph entity relationships between all medical entities in the clustered medical entity set for each business domain label, a clinical knowledge graph is constructed, including: Based on the set of graph entity relationships between every two medical entities in the clustered medical entity set of each business domain clustered data for each business domain label, multiple triplet knowledge units are determined for every two medical entities in the clustered medical entity set of each business domain clustered data for each business domain label. Based on all triplet knowledge units of all medical entities in the clustered medical entity set of each business domain label, a clustered knowledge graph of each business domain label is constructed. Based on the clustering knowledge graph of all business domain clustering data for each business domain label, construct a business domain knowledge graph for each business domain label. Construct a clinical knowledge graph based on the business domain knowledge graph with all business domain labels.