Construction method and system of Chinese herbal medicine knowledge graph

Abstract

The invention discloses a Chinese herbal medicine knowledge graph construction method and system, and the method comprises the steps: collecting structured data and unstructured data, and generating a drug property and pharmacology data table; extracting property and flavor channel tropism attributes of the medicinal materials, generating a medicinal property triple and a pharmacological triple, and establishing preliminary cross-domain association; defining logical constraints and compatibility rules of nature and flavor channel tropism, and constructing a causal reasoning chain; performing spatial semantic mapping on the channel tropism attributes and the pharmacological entities to generate a dynamic cross-layer mapping relationship; using an attribute graph database to store node types and edge types, and constructing a hybrid index; new data streams are collected in real time, the atlas is updated, and an intelligent decision-making interface and a visual interaction platform are provided. Through multi-source fusion, dynamic association, efficient storage and closed-loop optimization, the problems of semantic segmentation, update lag and low retrieval efficiency in traditional Chinese medicine digitization are solved, and scientific and real-time technical support is provided for traditional Chinese medicine modernization research, accurate medication and new medicine research and development.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of knowledge graph construction, in particular to a Chinese herbal medicine knowledge graph construction method and system. BACKGROUND

[0002] In recent years, with the acceleration of the digitization of traditional Chinese medicine, the construction of Chinese herbal medicine knowledge graph has gradually become the core direction of promoting the modernization of traditional Chinese medicine research and application, which shows great potential in integrating Chinese herbal medicine knowledge, revealing the internal relationship of drugs and assisting clinical decision-making. However, the current construction technology of Chinese herbal medicine knowledge graph still has many significant defects, which seriously restricts its further development and wide application: (1) Insufficient data integration capability: Traditional methods mostly rely on a single data source and cannot effectively integrate structured drug property data and unstructured pharmacological data. Semantic conflicts across data sources (such as differences in the description of the nature and taste of "Huangqi" in different documents) lead to fragmented knowledge, making it difficult to construct a unified knowledge system; (2) Weak drug property-pharmacological correlation: Existing graphs mostly use static rule mapping (such as manually defining "cold nature -> anti-inflammatory"), lack of dynamic correlation mechanism based on spatial semantics and statistical rules, leading to the separation of traditional drug property theory and modern molecular mechanism, and unable to support bidirectional reasoning; (3) Lack of dynamic update mechanism: Most graphs rely on periodic batch updates, which cannot incorporate new pharmacological research findings or clinical feedback in real time, and conflict data requires manual intervention, with high maintenance costs; (4) Limited retrieval efficiency and practicality: Graph database retrieval mostly relies on basic indexing strategies, and does not design mixed indexes for meridian properties and pharmacological weights , resulting in high response delay for "property-taste-target-disease" multi-hop queries, affecting the efficiency of clinical decision support. SUMMARY

[0003] In view of the above existing problems, the present application is proposed.

[0004] Therefore, the present application provides a Chinese herbal medicine knowledge graph construction method and system to solve the problems of limited feature extraction capability, weak anti-interference ability and insufficient model generalization performance in the existing cable partial discharge pattern recognition method.

[0005] To solve the above technical problems, the present application provides the following technical solutions: In a first aspect, the present application provides a Chinese herbal medicine knowledge graph construction method, comprising: According to the ancient books of Chinese herbal medicine, the modern pharmacological database and the clinical drug records, the structured data and unstructured data are collected, and through the term standardization, conflict resolution and semantic coding, the drug property and pharmacological data table is generated; Extract the properties of medicinal materials from the medicinal pharmacological data table, generate "medicinal materials, properties, and values" medicinal property triplets, and extract the action relationship between chemical components and targets from the medicinal pharmacological data table to generate "component, inhibition, and target" pharmacological triplets. Based on the medicinal property triplets, a medicinal property ontology layer is constructed, the logical constraints and compatibility rules of properties and meridians are defined, and based on the pharmacological triplets, a biomedical ontology is integrated to construct a "target-pathway-disease" causal reasoning chain. The spatial semantic mapping of the meridian properties and the pharmacological entities is performed to obtain the meridian coordinates and target positioning, and the spatial similarity of the meridian coordinates and the target positioning is calculated to generate a dynamic cross-layer mapping relationship. An attribute graph database is used to store node types and edge types, and a hybrid index is constructed. Real-time new data streams are collected, and an incremental graph embedding algorithm is used to dynamically update the graph, and an intelligent decision-making interface and a visual interaction platform are provided, and the graph optimization is driven in reverse through user feedback.

[0006] As a preferred scheme of the method for constructing the traditional Chinese medicine knowledge graph, the generation of the medicinal pharmacological data table comprises: The four properties of cold, heat, warm, and cool, and the five flavors of pungent, sweet, sour, bitter, and salty are numerically coded based on a predefined medicinal property term library, and the meridian property is mapped to a meridian coding table. The target names in the pharmacological entities are unified as UniProt ID, and the disease names are stored using ICD-11 coding. The semantic ambiguity problem of the same name medicinal materials is solved through expert review, and a cross-source data mapping rule library is established.

[0007] As a preferred scheme of the method for constructing the traditional Chinese medicine knowledge graph, the establishment of the preliminary cross-domain association comprises: A preset mapping rule is set, and if the property and taste attributes of the medicinal materials and the function description of the pharmacological targets have common keywords, a cross-layer relationship edge is generated. The cross-layer relationship edge is spatially pre-matched, and the same name mapping of the meridian belonging viscera and target tissue positioning is performed.

[0008] As a preferred scheme of the method for constructing the traditional Chinese medicine knowledge graph, the construction of the medicinal property ontology layer comprises: The logical constraints of properties and meridians are defined. The construction of the causal reasoning chain comprises, based on the KEGG pathway database, establishing a directional association of "target-pathway-disease", and labeling the pathway enrichment degree weight.

[0009] As a preferred scheme of the method for constructing the traditional Chinese medicine knowledge graph, the generation of the dynamic cross-layer mapping relationship comprises: set the Euclidean distance threshold between the coordinates, and calculate the Euclidean distance between the meridian coordinates and the target point positioning, if the distance is less than the preset threshold, a cross-layer relationship edge is generated; Set the statistical threshold, construct the weight matrix based on the co-occurrence frequency of nature and taste-target points, and if the weight value is higher than the statistical threshold, the generalization rule is derived.

[0010] As a preferred scheme of the construction method of the traditional Chinese medicine knowledge graph of the present application, the construction of the mixed index comprises: The B+ tree index is adopted for the meridian attribute, so that the mixed index supports range query; Hash index is adopted for the pharmacological entity name to realize accurate matching acceleration; Pre-calculate the high-frequency path and cache the subgraph structure.

[0011] As a preferred scheme of the construction method of the traditional Chinese medicine knowledge graph of the present application, the graph is dynamically updated, and an intelligent decision interface and a visual interaction platform are provided, and the optimization of the graph is driven backward through user feedback, comprising: New pharmacological research data, clinical drug feedback and ancient book digitization text flow are accessed in real time through a distributed message queue, and it is verified through a logic rule engine whether the new data conforms to the drug nature compatibility taboo; According to the difference between the new data and the original data, a threshold is set, and the data below the threshold is low conflict data, which is directly inserted into the graph, and the high conflict data above the threshold triggers an artificial review branch; The intelligent decision interface includes drug nature compatibility detection and drug recommendation: The drug nature compatibility detection includes calculating the risk value of the medicinal material combination based on the meridian overlap rate and the target point interaction database, and issuing an alarm for the combination that produces risk; Drug recommendation includes inputting symptoms and patient constitution, and jointly searching the drug nature layer and the pharmacological layer to generate a multi-dimensional drug use scheme.

[0012] In the second aspect, the present application provides a construction system of a traditional Chinese medicine knowledge graph, comprising: a distributed data acquisition module for configuring a multi-threaded crawler engine and an API gateway for real-time acquisition of original data from heterogeneous data sources; A natural language processing engine is used to integrate a field adaptive pre-training model and a rule engine to complete entity recognition and relationship extraction; A knowledge fusion computing module is used to deploy a graph embedding algorithm and an ontology reasoning machine to realize entity alignment and logical rule verification; A graph database storage system is used to construct a distributed storage cluster based on an attribute graph model to support high-concurrency query and transaction processing; A dynamic update control module is used to realize real-time processing of incremental data through a streaming computing framework, and update the graph in combination with confidence evaluation; The visualization interaction platform provides a visualization rendering engine and a RESTful API interface, and supports multi-terminal access and application integration.

[0013] Compared with the prior art, the present application has the following beneficial effects: the present application unifies ancient herbal property terms and modern pharmacological entities through term library driven data cleaning and semantic encoding, eliminates the conflict of homonymy, constructs a high-quality property pharmacological data table, and solves the problem of data fragmentation; based on the co-occurrence frequency and spatial similarity calculation of property-target, confidence weighted cross-layer mapping rules are generated, which breaks through the limitations of artificial rules, supports bidirectional reasoning of "property-target" and "target-meridian", and improves the association accuracy; by constructing a "target-pathway-disease" directional reasoning chain, the transformation from traditional experience to molecular mechanism is realized, and the scientificity of compatibility suggestion is improved; the attribute graph database is used to store node types and edge types, combined with B+ tree index and hash index, the response time of "property-target-disease" multi-hop query is shortened to milliseconds; through the real-time processing of new data by the streaming computing engine, combined with the consistency verification model of property-pharmacology and the incremental graph embedding algorithm, the atlas is dynamically updated; user feedback triggers knowledge correction in reverse, forming a closed loop of "data update-service optimization-feedback collection-renewal", which greatly improves the knowledge timeliness retention rate. Through multi-source fusion, dynamic association, efficient storage and closed loop optimization, the present application constructs a Chinese herbal medicine knowledge graph with deep coupling of property-pharmacology, which not only solves the problems of semantic fragmentation, update lag and low efficiency of retrieval in traditional Chinese medicine digitalization, but also provides scientific and real-time technical support for modernization research, precise medication and new drug research and development of traditional Chinese medicine. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0015] Fig. 1 The overall flowchart of the construction method of the Chinese herbal medicine knowledge graph according to an embodiment of the present application.

[0016] Fig. 2 The TF-IDF calculation flowchart of the construction method of the Chinese herbal medicine knowledge graph according to an embodiment of the present application.

[0017] Fig. 3 The mixed index query flowchart of the construction method of the Chinese herbal medicine knowledge graph according to an embodiment of the present application.

[0018] Fig. 4The drug property compatibility detection flowchart of the construction method of the traditional Chinese medicine knowledge graph is shown in an embodiment of the present application. DETAILED DESCRIPTION

[0019] In order to make the above objectives, characteristics and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings in the specification. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should belong to the protection scope of the present application.

[0020] Embodiment 1 Reference Figs. 1-4 In an embodiment of the present application, a construction method of a traditional Chinese medicine knowledge graph is provided, comprising: S100: Collecting structured data and unstructured data according to ancient Chinese medicine books, modern pharmacology databases and clinical medication records, and generating a drug property pharmacology data table through term standardization, conflict resolution and semantic coding; S200: Extracting the property, flavor and meridian tropism attributes of medicinal materials from the drug property pharmacology data table to generate a “medicinal material, property, value” drug property triple, and extracting the action relationship between chemical components and target points from the drug property pharmacology data table to generate a “component, inhibition, target point” pharmacology triple, and establishing a preliminary cross-domain association between drug property terms and pharmacology entities; S300: Constructing a drug property ontology layer based on the drug property triple, defining the logical constraints and compatibility rules of property, flavor and meridian tropism, and integrating a biomedical ontology based on the pharmacology triple to construct a “target point-pathway-disease” causal reasoning chain; S400: Spatial semantic mapping of meridian tropism attributes and pharmacology entities to obtain meridian tropism coordinates and target point positioning, and calculating the spatial similarity of meridian tropism coordinates and target point positioning to generate a dynamic cross-layer mapping relationship; S500: Storing node types and edge types using an attribute graph database and constructing a hybrid index; S600: Real-time collection of new data streams, dynamic updating of the graph using an incremental graph embedding algorithm, and providing an intelligent decision-making interface and a visual interactive platform, and driving the graph optimization in reverse through user feedback.

[0021] It should be noted that the data forms of ancient literature, modern pharmacology databases and clinical records are significantly different, the term descriptions have semantic conflicts and ambiguities, and the traditional drug property attributes lack dynamic mapping rules with molecular targets and metabolic pathways, making it difficult to support “experience-science” bidirectional reasoning. In addition, new pharmacology research results and clinical feedback cannot be incorporated into the graph in real time, and the existing storage indexing strategy is single, with high response delay, which restricts the timeliness of clinical application, mainly in the following aspects: Firstly, the diversity and heterogeneity of data sources lead to integration difficulties: Traditional Chinese medicine knowledge involves ancient literature (e.g., "Bencao Gangmu"), modern pharmacological databases, and clinical medication records, with data forms including structured tables, free text, medicinal material morphology, and non-standardized terminology. For example, "cold nature" in ancient books may be expressed as "cool" or "mildly cold" in different literature, and the chemical components and targets of "Huangqi" in modern databases need to be aligned across sources, such as aligning astragaloside with TNF-α (tumor necrosis factor-α) across sources. The semantic conflicts and format differences of such heterogeneous data lead to knowledge fragmentation, making it difficult for traditional methods to build a unified knowledge representation framework.

[0022] Secondly, the semantic gap between medicinal property theory and modern pharmacological mechanisms: Traditional medicinal property attributes, such as four properties, five flavors, and meridian tropism, are mostly experiential descriptions, and lack direct correlation rules with modern molecular mechanisms, such as targets, pathways, and diseases. For example, the theory of "acrid flavor returning to the lung meridian" cannot be mapped to the molecular pathway of "regulating TRPV1 receptor (transient receptor potential vanilloid 1 receptor) to inhibit cough reflex", limiting the reasoning ability of the knowledge graph. Existing methods rely on artificial rules, such as defining "cold-natured medicinal materials can resist inflammation" based on medical experience, but cannot rely on dynamically capturing statistical associations between medicinal properties and targets, such as relying on high-frequency co-occurrence of cold-natured medicinal materials and inflammation pathways in databases to infer that cold-natured medicinal materials can resist inflammation, which restricts the conversion efficiency from traditional theory to scientific validation.

[0023] Finally, data dynamics and the complexity of application scenarios: New pharmacological research findings, clinical medication feedback, and digital content of ancient books continue to emerge, requiring real-time integration into the knowledge graph. However, existing systems rely on batch update mechanisms and cannot handle streaming data such as daily PubMed (Public Medical Information Retrieval System) literature, and conflict data requires human intervention, resulting in high maintenance costs. In addition, complex query scenarios, such as querying "among the medicinal materials with warm nature and returning to the spleen meridian, what components can inhibit the COX-2 pathway", lack mixed indexing and pre-computation optimization, resulting in high response delay and difficulty in meeting the real-time needs of clinical decision-making.

[0024] Therefore, the present application constructs a complete technical solution of the construction of the Chinese herbal medicine knowledge graph through the steps S100-S600, integrates multi-source data such as ancient book literature, modern pharmacology databases and clinical records, constructs a unified drug property-pharmacology data table through term standardization and semantic conflict resolution, completely solves the data fragmentation and semantic ambiguity problems in the traditional method, and automatically generates weighted cross-layer rules based on the co-occurrence frequency of property-tar get, the space similarity calculation of meridian coordinates and target positioning, thereby breaking through the limitations of artificial static rules; through the construction of the drug property ontology layer and the pharmacology ontology layer, combining the attribute graph database and the hybrid index, the retrieval efficiency in a complex scene is improved; through real-time collection of new data streams, dynamic updating of the graph is realized; user feedback triggers knowledge correction in reverse, forming a closed loop of “data updating→service optimization→feedback collection→renewal”.

[0025] Embodiment 2 Reference Figs. 1-4 For an embodiment of the present application, based on the above embodiment, a method for constructing a Chinese herbal medicine knowledge graph is provided.

[0026] In the present application, in step S100, structured data and unstructured data are collected according to ancient books of Chinese herbal medicine, modern pharmacology databases and clinical records of drug use, and a drug property-pharmacology data table is generated through term standardization, conflict resolution and semantic encoding, including the following steps A1-A3: A1: based on a pre-defined drug property term library, the four properties of cold, heat, warm and cool, and the five flavors of pungent, sweet, sour, bitter and salty are numerically encoded, and the meridian property is mapped to a meridian encoding table; Specifically, in the present application, the four properties of cold, heat, warm and cool are numerically encoded, wherein cold is -1, cool is -0.5, warm is +0.5, and hot is +1; the five flavors of pungent, sweet, sour, bitter and salty are mapped to vector encoding, wherein pungent = [1, 0, 0, 0, 0], sweet = [0, 1, 0, 0, 0], sour = [0, 0, 1, 0, 0], bitter = [0, 0, 0, 1, 0], and salty = [0, 0, 0, 0, 1]; the meridian property is based on a meridian encoding table, and the meridian encoding table encodes the twelve meridians according to English abbreviations, such as liver meridian encoding LR and lung meridian encoding LU, and in order to further refine the description of the meridian property, each meridian is associated with human anatomy coordinates, such as the liver meridian corresponding to the right costal region coordinates (x, y, z), wherein the specific coordinates are defined according to human anatomy standards.

[0027] It should be noted that the vector encoding of the five flavors of pungent, sweet, sour, bitter and salty can ensure that each property corresponds to a unique position, in addition, specific Chinese herbal medicines often have multiple flavors, and vector encoding can conveniently represent multiple flavor combinations, such as “Ba Ji Tian, pungent and sweet” which can be represented as: [1, 1, 0, 0, 0].

[0028] A2: The target name in the pharmacological entity is unified as UniProt ID, and the disease name is stored in ICD-11 code; Among them, UniProt is a comprehensive database that provides protein sequence and function information, which integrates the information of multiple databases and provides rich protein annotations including sequence, function, structure, interaction, etc. UniProt ID is the unique identifier of each protein entry in UniProt database, which is a 6-digit code composed of letters and numbers. UniProt ID does not follow a specific coding rule, but is randomly generated to ensure the uniqueness of the code, and each UniProt ID is used to identify a unique protein sequence, which will not change due to database updates, such as TNF-α named P01375, where P01375 identifies the protein Tumor necrosis factor.

[0029] The disease name uses ICD-11 code (International Classification of Diseases, 11th Revision), which is a brand new disease classification system. Its coding structure uses combined coding rules, namely main code + extension code. The main code is the core part of the code, which is used to define the main category of the disease or health problem. The main code is usually an independent code that can be used alone to provide the least but most meaningful information. The extension code is used to supplement the main code to provide more detailed descriptions, such as the severity of the disease, anatomical location, etc. The extension code cannot be used alone and must be used in conjunction with the main code.

[0030] It should be noted that the coding method of ICD-11 is divided into pre-assembly and post-assembly. The pre-assembly is a main code that pre-assembles all relevant information of a clinical concept, for example, the coding of abdominal aortic aneurysm with perforation is BD50.40. The post-assembly is to describe a complex clinical concept by combining multiple codes, for example, the coding of left hip osteoarthritis with chronic pain is FA00.Z&XK8G / MG30.30.

[0031] A3: Solve the semantic ambiguity problem of the same name medicinal materials through expert review, and establish a cross-source data mapping rule library.

[0032] Specifically, experts determine whether the same entity is the same entity through molecular fingerprint similarity for the same name medicinal materials, such as "Baimu" containing Chuanbeimu and Zhebeimu. The fingerprint similarity of the two medicinal materials is determined, and the specific steps include: First, use chemical informatics tools to extract the molecular fingerprints of the main components in Chuanbeimu and Zhebeimu, for example, for Chuanbeimu in Chuanbeimu and Zhebeimu in Zhebeimu, respectively generate its MACCS Keys fingerprint or ECFP fingerprint; Secondly, the Tanimoto coefficient (similarity coefficient) between the molecular fingerprints of beibasin and zhebeisanoside is calculated. The Tanimoto coefficient is a commonly used index for measuring the similarity of two molecular fingerprints, and its calculation formula is: ; Wherein: Tanimoto coefficient; a and b are the number of bits set to 1 in the two molecular fingerprints of fritillariae cirrhosae and fritillariae thunbergii respectively; c is the number of bits commonly set to 1 in the two molecular fingerprints of fritillariae cirrhosae and fritillariae thunbergii.

[0033] It should be noted that the value range of Tanimoto coefficient is 0 to 1, and the value closer to 1 indicates that the similarity of two molecules is higher. In the field of drug design, it is generally believed that when Tanimoto coefficient ≥ 0.85, two molecules have high similarity. According to the published data, fritillariae cirrhosae mainly contains steroidal alkaloids such as beibasin, while fritillariae thunbergii contains various steroidal saponins, and the chemical components of the two are significantly different, and the Tanimoto coefficient is less than 0.85, so it is determined that fritillariae cirrhosae and fritillariae thunbergii cannot be regarded as the same entity.

[0034] The establishment of cross-source data mapping rule library includes defining the mapping rule of "property taste→target function", and storing it as RDF triple (Resource Description Framework Triple, Resource Description Framework Triple), the format is "entity, associated entity, corresponding entity", for example "(bitter, associated target, TNF-α)".

[0035] In the embodiment of the application, the property taste meridian property of medicinal materials is extracted from the medicinal property pharmacological data table in step S200 to generate "medicinal material, property taste, value" medicinal property triple, and the action relationship between chemical components and targets is extracted from the medicinal property pharmacological data table to generate "component, inhibition, target" pharmacological triple, and a preliminary cross-domain association is established between medicinal property terms and pharmacological entities, including the following steps B1-B2: B1: Set the preset mapping rule, if the property taste attribute of medicinal materials and the function description of pharmacological targets exist co-occurring keywords, generate cross-layer relationship edge; Specifically, first, the TF-IDF algorithm (Term Frequency-Inverse Document Frequency, Term Frequency-Inverse Document Frequency) is used to extract property taste-target co-occurring keywords from literature, and when the co-occurring frequency is ≥ 5 times, the cross-layer edge is generated; Wherein, the TF-IDF algorithm is a commonly used text mining technology, which is used to evaluate the importance of a word in a document set. It combines the concepts of term frequency (TF) and inverse document frequency (IDF), which specifically includes: ; ; ;

[0036] wherein, tf(t, d) represents the term frequency of the word t in the document ; t represents a specific word; d represents a specific document; idf(t, D) represents the inverse document frequency of the word t in the document collection D; D represents the document collection; tf(t, d) represents the number of times the word t appears in the document ; N represents the total number of words in the document ; N is the total number of documents in the document collection D, is the number of documents containing the word t.

[0037] It should be noted that, tf(t, d) represents the importance of the word t in the document , while considering its universality in the entire document collection D, so as to extract the key words of the nature- flavor- target point co-occurrence.

[0038] Secondly, the calculation formula of the edge attribute annotation confidence c is as follows: ; In an optional embodiment, step S200 can also extract the key words of the nature- flavor- target point co-occurrence based on the co-occurrence of the point mutual information and the chi-square test, and the specific steps include: Firstly, text preprocessing, the literature is segmented, stop words are removed, and stem extraction or morphological reduction is performed, and the expression form is unified, such as standardizing “clearing heat and detoxifying” and “clearing heat and detoxifying” to the same word.

[0039] Secondly, the co-occurrence matrix is constructed, the co-occurrence times of nature, flavor and target point, the number of times of nature and flavor alone, and the number of times of target point alone are counted, and the co-occurrence matrix is constructed.

[0040] Thirdly, the point mutual information is calculated to measure the correlation strength of nature, flavor and target point. The greater the PMI value, the more significant the correlation. The formula of the point mutual information is as follows: ;

[0041] wherein, , , , N is the total number of word pairs; a is the nature and flavor attribute; b is the pharmacological target point; is the co-occurrence probability of nature and flavor a and target point b; is the probability of nature and flavor a alone appearing. The probability of the target point b appearing alone.

[0042] Fourthly, by chi-square test, the significance verification is carried out, a 2x2 contingency table (co-occurrence / non-co-occurrence) is constructed, and the chi-square statistic is calculated, and the formula is as follows: ;

[0043] Among them, is the actual observation frequency; is the expected frequency, .

[0044] According to the degrees of freedom and the chi-square value, the p value is calculated, and the co-occurrence pair with p<0.05p<0.05 is screened out, wherein the degrees of freedom (df) is ; the p value is determined according to the chi-square distribution table and the degrees of freedom (usually p<0.05 is considered significant).

[0045] Fifthly, threshold setting and edge generation, including: generating cross-layer edges by comprehensively considering the following conditions: PMI value>0 (positive correlation); co-occurrence times≥3 (to avoid low-frequency noise); Chi-square test is significant (p<0.05p<0.05).

[0046] Finally, the edge attribute confidence is calculated, combined with PMI and co-occurrence frequency, and the normalized confidence C is defined as: ;

[0047] Among them, is the point mutual information; is the maximum PMI value in all co-occurrence pairs.

[0048] It should be noted that PMI combined with chi-square test can identify low co-occurrence but statistically significant associations, such as special compatibility in traditional Chinese medicine. Through significance test to filter random co-occurrence, avoid the over-reliance of TF-IDF on high-frequency words, PMI directly quantifies the association strength, and chi-square test provides statistical basis, which meets the needs of traditional Chinese medicine knowledge discovery.

[0049] B2: Spatial pre-matching is performed on the cross-layer relationship edges, and the same name mapping is performed on the zang-fu organs belonging to the meridian and the target tissue positioning.

[0050] Specifically, the specific steps include: Firstly, name standardization, including: zang-fu standardization, that is, unified language standard for traditional zang-fu names; target tissue standardization, input target tissue positioning description, and standardize the target tissue name.

[0051] Further, a predefined meridian-viscera mapping table is defined, and a traditional meridian name is input, and the meridian is converted into a modern anatomical name according to the predefined meridian-viscera mapping table, such as “liver meridian→liver”.

[0052] In the embodiments of the present application, the drug property ontology layer is constructed based on the drug property triplets in step S300, the logical constraints of nature, flavor and meridian are defined, and the biomedical ontology is integrated based on the pharmacological triplets to construct the causal reasoning chain of “target point-pathway-disease”, including the following steps C1-C2: C1: defining the logical constraints of nature, flavor and meridian; Specifically, the logical constraints of nature, flavor and meridian are defined, the RDF triplets generated in step S200 are input, and the data is inferred according to the set rules. If the rule is triggered, the rule engine will automatically mark the compatibility as contraindication, and the result is stored.

[0053] C2: constructing the causal reasoning chain, including establishing the directional association of “target point-pathway-disease” based on the KEGG (Kyoto Encyclopedia of Genes and Genomes) pathway database, and labeling the pathway enrichment weight.

[0054] Specifically, the specific steps of constructing the causal reasoning chain based on the KEGG pathway database include: First, determine the directional association of target point-pathway-disease: Among them, the association between target point and pathway is to obtain the association information between target point and pathway through KEGG pathway database, for example, target point TNF-α participates in inflammation pathway; The association between pathway and disease: extract the association between pathway and disease from KEGG database, for example, inflammation pathway is related to rheumatoid arthritis.

[0055] Second, label the pathway enrichment weight, the specific process includes: Gene set enrichment analysis: use gene set enrichment analysis to calculate the enrichment weight of the pathway in a specific disease, and the formula of the pathway enrichment weight E is as follows: ;

[0056] Among them, E represents the enrichment degree of the pathway in a specific disease or condition, the larger the value, the more significant the association between the pathway and the disease or condition; represents the probability of the occurrence of the gene or protein set in the pathway, the smaller the p value, the higher the significance of the pathway.

[0057] Label the enrichment weight: label the calculated enrichment weight to the pathway for use in the causal reasoning chain.

[0058] Finally, the causal reasoning chain is constructed: (1) The data of KEGG pathway database, ICD-11 disease database and UniProt target database are integrated to construct a complete “target-pathway-disease” knowledge graph; (2) The rule-based reasoning engine is used to analyze the causal reasoning chain, and in the reasoning process, the pathway enrichment degree weight needs to be added to preferentially select the pathways with high enrichment degree for reasoning, so as to improve the accuracy and reliability of the reasoning result.

[0059] In the embodiment of the present application, the space semantic mapping of the meridian tropism attribute and the pharmacological entity is performed in step S400 to obtain the meridian tropism coordinates and target positioning, and the space similarity of the meridian tropism coordinates and the target positioning is calculated to generate a dynamic cross-layer mapping relationship, including the following steps D1-D2: D1: Set the Euclidean distance threshold between coordinates, and calculate the Euclidean distance between the meridian tropism coordinates and the target positioning. If the distance is less than the preset threshold, a cross-layer relationship edge is generated; Wherein, the threshold of the distance between the meridian tropism coordinates and the target positioning is 15mm, and the calculation formula is: ;

[0060] Wherein, is the coordinate value of the meridian tropism coordinates in the three-dimensional space, is the coordinate value of the target positioning in the three-dimensional space; If , an edge <meridian tropism, space association, target> is generated.

[0061] D2: Set a statistical threshold, construct a weight matrix based on the nature and taste-target co-occurrence frequency, and if the weight value is higher than the statistical threshold, the generalization rule is derived.

[0062] Wherein, the statistical threshold is set to be, the nature and taste-target co-occurrence frequency is ≥10 times, and the confidence is ≥70% and the support is ≥5%; Let the nature and taste set be F={F1, F2, …, F m}, and the target set be T={T1, T2, …, T n}.

[0063] The element of the weight matrix W represents the association strength between the nature and taste F i and the target T j , and the calculation formula is: ;

[0064] Wherein, represents the nature and taste F​i With target T j The number of times they co-occur; Indicates flavor F i Total number of occurrences; Indicates target point T j Total number of occurrences; like If the value is ≥0.6, then the generalization rule is derived. For example, if the flavor F i (Bitterness) and target T j Weight of (regulation of inflammatory pathways) If the value is ≥0.6, the generalization rule "bitter taste → regulation of inflammatory pathways" can be derived.

[0065] In this embodiment of the application, step S500 uses an attribute graph database to store node types and edge types, and constructs a hybrid index, including the following steps E1-E2: E1: A B+ tree index is used for the meridian attribute, enabling the hybrid index to support range queries; B+ tree indexes are self-balancing tree data structures commonly used in databases and file systems, suitable for storing large amounts of data. Each node in a B+ tree index contains multiple key values ​​and pointers to child nodes. Internal nodes do not store data, only key values ​​and pointers; leaf nodes store the actual data and pointers to the next leaf node. All leaf nodes are at the same level and linked together sequentially, forming a linked list structure. The key values ​​of internal nodes are used to partition the data, ensuring that the data in each subtree is less than or equal to the key value of that node, while the key values ​​in leaf nodes represent a portion of the actual stored data.

[0066] Specifically, the meridian attribute uses a numerical format as the key-value pair in the B+ tree, and each key-value node stores information including the meridian attribute, a disk address pointing to a leaf node, and the key-value range of the current node. Each leaf node stores a maximum of 100 key-value pairs, ensuring that the data size of a single node is adapted to the memory page size, reducing disk I / O operations. Leaf nodes are connected by a doubly linked list to support efficient range traversal.

[0067] For example, to search for medicinal materials with the range "warmth ≤ code ≤ heat" (i.e., 0.5 ≤ key ≤ 1), the specific steps for this range search include: Starting from the root node, locate the first leaf node with a key value ≥ 0.5; Traverse the leaf node linked list and collect all records whose key values ​​are in the range [0.5, 1]. Returns a list of matching medicinal materials, displaying all medicinal materials that meet the requirements.

[0068] It should be noted that when traversing the leaf nodes, the preloading of adjacent node data into the cache reduces random access delay; when the node splits, the batch write strategy is used to reduce disk fragmentation.

[0069] It should be further noted that when the number of key values of a node exceeds 100, it is split into two nodes according to the middle value, i.e. the 50th key, and the parent node is updated; when the number of key values stored in a node is less than 30, it is automatically merged with the adjacent node, triggering the parent node key value update.

[0070] E2: Hash index is used for pharmacological entity name to realize accurate matching acceleration; Specifically, the hash value of the pharmacological entity name is calculated by using a polynomial rolling hash for UniProt ID, and the formula is as follows: ； Wherein: S is the input string, i.e. UniProt ID; n is the length of the string; is the integer value of the th character in the string; 31 is a constant used to control the distribution of the hash value; is the modulus used to limit the range of the hash value.

[0071] It should be noted that the bucket size is set to 1000, and the hash value is mapped to the bucket index (0~999) after taking the modulus, wherein a linked list is maintained in each bucket to store the conflicting key-value pairs, when two different UniProt IDs are calculated by the hash function to get the same bucket index, they will be stored in the same bucket's linked list. When the length of the linked list exceeds 10, it is converted into a skip list, reducing the query complexity from O(n) to , wherein the skip list is a data structure that allows fast search, which realizes fast search by adding multiple layers of index to the linked list; when the total number of keys / buckets>0.75, trigger expansion, i.e. the number of new buckets is expanded to twice the original, and update all hash keys, this process uses gradual expansion to avoid service interruption.

[0072] Wherein, O(n) represents that the time complexity of the algorithm is linear, i.e. the running time of the algorithm is linearly related to the size of the input data, when the size of the data increases, the running time also increases proportionally. O(n) processes only one data item at each step, which is suitable for simple traversal operations such as linear search of arrays.

[0073] , which represents that the time complexity of the algorithm is logarithmic, usually because the algorithm uses a divide-and-conquer strategy, which halves the data size each time, and the running time grows much slower than the data size. Each step processes half of the data size, which is suitable for efficient search algorithms such as binary search.

[0074] In an optional embodiment, step S500 can also construct a pharmacological entity name index based on a radix tree compressed hierarchical index, including: First, the radix tree structure is optimized, and the UniProt ID is split by characters into a hierarchical structure, and each layer corresponds to a character position. If a path is unique, the nodes are combined to reduce the depth. For example, if all IDs start with "P1", the root node directly points to the combined node of "P1" instead of expanding character by character. Further, the characters at the end of the path are lightly hashed, and the hash value is used as the key of the leaf node to directly associate the storage location. Combining path compression and hashing reduces the tree depth and comparison frequency.

[0075] Secondly, a hybrid index strategy is established, in which the prefix segment index divides the ID into a fixed length prefix and the remaining part, the prefix is managed by the radix tree, the remaining part is quickly filtered by the Bloom filter to filter invalid matches, and then combined with the binary search to locate in the ordered list; the lazy node expansion dynamically expands the tree node only when a new ID is inserted, avoiding pre-allocated memory and improving space utilization.

[0076] Finally, conflict processing and expansion, when the number of child nodes of a node exceeds the threshold, the node is automatically split into a more granular level to balance query efficiency and memory usage, and the level of the tree or the segment length is increased as needed to avoid global reconstruction. For example, when the number of IDs under a certain prefix exceeds the threshold, the remaining part of the prefix is further segmented.

[0077] E3: Precompute high-frequency paths and cache subgraph structures.

[0078] The high-frequency path identification includes: recording daily query logs, and counting the top 10 paths of daily queries; The query log is weighted, and the top 10 paths with the highest weight are selected for precomputation. The path weight The calculation formula is as follows: ;

[0079] Where P is a specific query path.

[0080] The precomputation specific strategy includes: using the materialized view function of the graph database, periodically executing the predefined query, and generating a subgraph snapshot; caching the subgraph snapshot to the in-memory database, and setting the key as the path name and the value as the serialized subgraph data.

[0081] Compared with the existing scheme, the advantages of the present application are shown in Table 1: Table 1 Comparison of Herb Knowledge Graph Query Performance

[0082] In the embodiments of the present application, the new data stream is collected in real time in step S500, the incremental graph embedding algorithm is used to dynamically update the graph, and the intelligent decision interface and the visual interaction platform are provided, and the graph optimization is driven in reverse through user feedback, including the following steps F1-F2: F1: Real-time access to new pharmacological research data, clinical medication feedback and digitized text stream of ancient books through a distributed message queue, and verify whether the new data meets the drug property compatibility contraindication through a logical rule engine; According to the difference between the new data and the original data, a threshold is set, and the data below the threshold is low conflict data, which is directly inserted into the graph, and the high conflict data above the threshold triggers an artificial review branch; Specifically, the low conflict data is data with a difference value of ≤5%, which is directly inserted into the graph; The high conflict data is data with a difference value of >5%, which triggers an artificial review process, generates an audit report and emails the expert.

[0083] The difference value calculation method includes: First, calculate the meridian overlapping rate, the formula is as follows: ； Wherein, is the meridian set of medicinal material A; is the meridian set of medicinal material B; is the common meridian set of medicinal material A and medicinal material B; is the total meridian set of medicinal material A and medicinal material B.

[0084] It should be noted that if the meridian of medicinal material A is liver meridian and lung meridian, and the meridian of medicinal material B is lung meridian and heart meridian, the meridian overlapping rate is 1 / 3, i.e. common lung meridian.

[0085] Second, calculate the target point interaction risk value: calculate the combination risk value based on the edge weight of the target point interaction network.

[0086] It should be noted that if there are multiple high-weight paths in the target point interaction network of medicinal material A and medicinal material B, the risk value is higher.

[0087] For example, the meridian of a certain medicinal material combination is liver meridian and lung meridian, and the target point interaction risk value is 0.7; the meridian of the original data of the medicinal material combination is liver meridian and heart meridian, and the target point interaction risk value is 0.3.

[0088] Therefore, the difference between the original meridian (liver, heart) and the new meridian (liver, lung) is 1 / 2, and the difference percentage is 50%; the difference in target interaction risk is 0.7-0.3=0.4, and the difference percentage is 133%.

[0089] The difference percentage is calculated as follows: ;

[0090] The average of the difference values between 50% and 133% is 91.5%, which exceeds the threshold value of 5%, triggering the manual review process.

[0091] It should be noted that in the "Guidelines for the Construction of Traditional Chinese Medicine Knowledge Graph", the recommended difference threshold range is 4%-6%, of which 5% is the optimal value for balancing accuracy and efficiency, and in the weight mapping of nature, taste and meridian, and target relationship, 5% difference corresponds to the "acceptable range of drug efficacy deviation" in drug nature theory (such as the equivalence of nature of slight cold and nature of cool). In addition, in statistics, a 5% difference threshold is a common rule of thumb, often used to distinguish between low-conflict and high-conflict data. This rule of thumb can help the system automatically identify most of the data with high consistency, and submit data that may have problems to manual review.

[0092] F2: Intelligent decision-making interface includes drug nature compatibility detection and medication recommendation: For drug nature compatibility detection, based on the meridian overlap rate and target interaction database, the risk value of the medicinal material combination is calculated, and an alarm is issued for combinations that produce risks; Specifically, compatibility detection needs to calculate the meridian overlap rate of medicinal material combination. The formula for calculating the meridian overlap rate of medicinal material combination is as follows: ; Among them, is the meridian set of medicinal material A; is the meridian set of medicinal material B; is the common meridian set of medicinal material A and medicinal material B; is the total meridian set of medicinal material A and medicinal material B.

[0093] It should be noted that if then trigger the warning.

[0094] Medication recommendation includes inputting symptoms and patient constitution, and jointly searching drug nature layer and pharmacology layer to generate multi-dimensional medication scheme; Specifically, inputting the symptom "cough due to lung heat" is analyzed as nature and taste constraints, i.e. cold nature, lung meridian, and target constraints, i.e. inhibiting IL-6, to finally generate medication scheme recommendations, such as candidate medicinal materials being Huangqi, Sangbaipi, and the confidence ranking of the medication scheme under the disease.

[0095] In summary, the present application encodes four qi and five flavors by numerical value, which is convenient for computer processing and analysis, improves the data processing efficiency, and is convenient for quantitative analysis and pattern recognition; by unifying the target name in the pharmacological entity as UniProtID and storing the disease name by ICD-11 code, the data standardization is realized, which is convenient for data integration and comparative analysis; by determining whether it is the same entity through molecular fingerprint similarity, the ambiguity problem of the same medicinal material is solved, the accuracy of medicinal material identification is improved, and confusion is avoided; by setting the preset mapping rule, the cross-layer relationship edge is generated, the space pre-matching of the cross-layer relationship edge is performed, the association between the nature and flavor of medicinal materials and the pharmacological targets is established, and the basis for subsequent reasoning analysis is provided; by defining the logical constraints and compatibility rules of nature and flavor, the biomedical ontology is integrated based on the pharmacological triple, the causal reasoning chain of "target-pathway-disease" is constructed, and the scientificity and rationality of medicinal material compatibility are improved; by setting the Euclidean distance threshold between coordinates, the Euclidean distance between the meridian coordinates and the target positioning is calculated, the weight matrix is constructed based on the co-occurrence frequency of nature and flavor-target, and the recommendation of medicinal material compatibility is optimized; by using B+ tree index and hash index to improve the data storage and query efficiency, precomputing high-frequency paths and caching subgraph structures, the system performance is improved, the query delay is reduced, and the user experience is enhanced; by collecting new data streams in real time, dynamically updating the atlas after consistency verification of medicinal properties and pharmacology, providing an intelligent decision-making interface and a visual interaction platform, the dynamic updating and intelligent recommendation of knowledge are realized, and the accuracy and efficiency of decision-making are improved.

[0096] Embodiment 3 The above is an illustrative scheme of the method for constructing the Chinese herbal medicine knowledge graph. It should be noted that the technical scheme of the system for constructing the Chinese herbal medicine knowledge graph belongs to the same concept as the technical scheme of the method for constructing the Chinese herbal medicine knowledge graph described above. The technical scheme of the system for constructing the Chinese herbal medicine knowledge graph in this embodiment is not described in detail, and can be referred to the description of the technical scheme of the method for constructing the Chinese herbal medicine knowledge graph.

[0097] The present embodiment also provides a system for constructing a Chinese herbal medicine knowledge graph, comprising: A distributed data acquisition module is configured to configure a multi-threaded crawler engine and an API gateway for real-time acquisition of raw data from heterogeneous data sources; A natural language processing engine is configured to integrate a field adaptive pre-training model and a rule engine to complete entity recognition and relationship extraction; A knowledge fusion computing module is configured to deploy a graph embedding algorithm and an ontology reasoning machine to realize entity alignment and logical rule verification; A graph database storage system is configured to construct a distributed storage cluster based on an attribute graph model to support high-concurrency query and transaction processing; The dynamic update control module realizes real-time processing of incremental data through a streaming computing framework, and updates the graph in combination with confidence evaluation; The visualization interaction platform provides a visualization rendering engine and a RESTful API interface, and supports multi-terminal access and application integration.

[0098] The embodiment also provides an electronic device suitable for the construction of the Chinese herbal medicine knowledge graph, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the method for constructing the Chinese herbal medicine knowledge graph proposed in the above embodiment.

[0099] The embodiment also provides a storage medium having a computer program stored thereon, and the program is executed by a processor to realize the method for constructing the Chinese herbal medicine knowledge graph proposed in the above embodiment.

[0100] The storage medium proposed in the embodiment and the method for constructing the Chinese herbal medicine knowledge graph proposed in the above embodiment belong to the same inventive concept, and the technical details not described in detail in the embodiment can be referred to the above embodiment, and the embodiment has the same beneficial effects as the above embodiment.

[0101] Through the above description of the embodiments, those skilled in the art can clearly understand that the present application can be realized by means of software and necessary general hardware, and of course can also be realized by hardware. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, and the computer software product can be stored in a computer readable storage medium, such as a floppy disk, a read-only memory (ROM), a random access memory (RAM), a FLASH, a hard disk or an optical disk, etc., including a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods of various embodiments of the present application.

[0102] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, and all should be covered in the scope of the claims of the present application.

Claims

1. A method for constructing a traditional Chinese medicine knowledge graph, characterized in that, The application relates to a Chinese medicine property and pharmacology knowledge graph construction method and system. According to ancient Chinese herbal medicine books, modern pharmacology databases and clinical medication records, structured data and unstructured data are collected, and drug property pharmacology data tables are generated through term standardization, conflict resolution and semantic coding; The properties of medicinal materials are extracted from the drug property pharmacology data table to generate a "medicinal material, property and value" drug property three tuple, and the action relationship between chemical components and target points is extracted from the drug property pharmacology data table to generate a "component, inhibition and target point" pharmacology three tuple, and a preliminary cross-domain association is established between the drug property terms and the pharmacology entities; A drug property ontology layer is constructed based on the drug property three tuple, logical constraints and compatibility rules of properties are defined, a biomedical ontology is integrated based on the pharmacology three tuple, and a "target point-pathway-disease" causal reasoning chain is constructed; The spatial semantic mapping of the meridian property and the pharmacology entity is carried out to obtain the meridian coordinate and the target point positioning, and the spatial similarity of the meridian coordinate and the target point positioning is calculated to generate a dynamic cross-layer mapping relationship; An attribute graph database is used to store the node types and edge types of the meridian property and the pharmacology entity, and a hybrid index is constructed; New data streams are collected in real time, an incremental graph embedding algorithm is used to dynamically update the graph, an intelligent decision interface and a visual interaction platform are provided, and the graph optimization is driven in reverse through user feedback. 2.The method of claim 1, wherein: The drug property pharmacology data table is generated by: Based on the pre-defined drug property term library, the four properties of cold, heat, warm and cool and the five flavors of pungent, sweet, sour, bitter and salty are numerically coded, and the meridian property is mapped to the meridian coding table; The target point name in the pharmacology entity is unified as UniProt ID, and the disease name is stored in ICD-11 coding; The semantic ambiguity problem of the same name medicinal material is solved through expert review, and a cross-source data mapping rule library is established. 3.The method of claim 2, wherein: The preliminary cross-domain association is established by: Setting a preset mapping rule, if the property and flavor attributes of the medicinal material and the function description of the pharmacology target point have common keywords, a cross-layer relationship edge is generated; The spatial pre-matching of the cross-layer relationship edge is carried out, and the same name mapping of the meridian belonging viscera and the target point tissue positioning is carried out. 4.The method of claim 3, wherein the method comprises: The drug property ontology layer is constructed by: Defining the logical constraints of the property and flavor meridian; The causal reasoning chain is constructed by establishing the directional association of "target point-pathway-disease" based on the KEGG pathway database, and the pathway enrichment degree weight is marked. 5.The method of claim 4, wherein the method comprises: The dynamic cross-layer mapping relationship is generated by: Setting the Euclidean distance threshold value between the coordinates, calculating the Euclidean distance between the meridian coordinate and the target point positioning, and if the distance is less than the preset threshold value, a cross-layer relationship edge is generated; Setting a statistical threshold, constructing a weight matrix based on the co-occurrence frequency of property and target point, and if the weight value is higher than the statistical threshold, a generalization rule is derived. 6.The method of claim 5, wherein the method comprises: The hybrid index is constructed by: B+ tree index is adopted for the meridian property, so that the hybrid index supports range query; Hash index is adopted for the pharmacology entity name to realize accurate matching acceleration; High-frequency paths are pre-calculated and subgraph structures are cached.

7. The method for constructing a knowledge graph of traditional Chinese medicine as described in claim 6, characterized in that: The dynamic updating of the graph, the provision of the intelligent decision interface and the visual interaction platform, and the driving of the graph optimization in reverse through user feedback include: Real-time access to new pharmacological research data, clinical drug feedback and ancient book digitized text flow through distributed message queues to obtain new data, and logical rule engine to verify whether the new data meets the drug compatibility contraindications; According to the difference between the new data and the original data, the threshold is set, and the data below the threshold is low conflict data, which is directly inserted into the atlas, and the high conflict data above the threshold triggers manual review branch; Intelligent decision-making interface includes drug compatibility detection and medication recommendation: Drug compatibility detection includes calculating the risk value of herb combination based on the overlap rate of meridian and target interaction database, and issuing an alarm for the combination that produces risk; Medication recommendation includes inputting symptoms and patient constitution, and combining drug property layer and pharmacological layer to generate multi-dimensional medication plan.

8. A Chinese herbal medicine knowledge graph construction system, applying the method of any one of claims 1-7, characterized in that, Including: Distributed data acquisition module, used for configuring multi-threaded crawler engine and API gateway, for real-time acquisition of original data from heterogeneous data sources; Natural language processing engine, for integrating field adaptive pre-training model and rule engine, to complete entity recognition and relationship extraction; Knowledge fusion computing module, for deploying graph embedding algorithm and ontology reasoning machine, to realize entity alignment and logical rule verification; Graph database storage system, based on attribute graph model to build distributed storage cluster, supporting high concurrency query and transaction processing; Dynamic update control module, through the stream computing framework to realize the real-time processing of incremental data, combined with confidence evaluation to update the atlas; Visual interactive platform; provides visual rendering engine and RESTful API interface, supports multi-terminal access and application integration. 9.An electronic device, comprising: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions, which realize the steps of the construction method of the Chinese herbal medicine knowledge graph in any one of claims 1 to 7 when executed by the processor. 10.A computer readable storage medium storing computer executable instructions, which realize the steps of the construction method of the Chinese herbal medicine knowledge graph in any one of claims 1 to 7 when executed by the processor.

Images