Drug recommendation method and device based on three-layer hyper-relational knowledge graph model

Abstract

The application provides a drug recommendation method and device based on a three-layer super-relation knowledge graph model, and the method comprises the following steps: obtaining user attribute information of a target user, wherein the user attribute information comprises disease conditions, clinical symptoms, physical examination and physiological state information; constructing a three-layer super-relation knowledge model based on a super-relation knowledge graph, inputting the user attribute information into the three-layer super-relation knowledge model to convert the user attribute information into search conditions of a graph query statement; translating the search conditions into a correct graph database query statement, obtaining a reasoning result by calling a graph database search engine according to preset reasoning conditions; adding the reasoning result to a drug use scheme, obtaining a target drug according to the drug use scheme, and pushing the target drug to the target user. The application realizes a medical rule-based drug recommendation auxiliary decision application.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of knowledge graph. BACKGROUND

[0002] Knowledge graph is a large-scale semantic network composed of nodes and directed edges, and the nodes are used to represent entities or concepts, and the edges represent various semantic relationships between entities. Generally, the knowledge graph takes SPO (Subject-Predication-Object, subject-predicate-object) triple as the basic unit, and SPO can be divided into <entity-relation-entity> or <entity-attribute-attribute value> triples, which are mainly represented by RDF model or property graph model.

[0003] Due to the strong professionalism and high complexity of medical knowledge, the SPO model structure is relatively simple and difficult to accurately express medical knowledge. As one of the most common complex chronic diseases in the world, the complexity and professionalism of hypertension knowledge are representative in the medical field. The following takes hypertension knowledge as an example to specifically illustrate the difficulties in constructing a knowledge graph with triples as the basic unit:

[0004] Difficulty 1: Simple binary relationship is difficult to represent complex medical knowledge. Generally, the representation method of SPO-based triples simplifies the complexity of data stored in the knowledge graph too much. Studies have shown that many knowledge in the real world cannot be completely represented by simple triples. In particular, the loss of high-order information by simple triples reduces the expression ability of the knowledge graph. Related studies have confirmed that more than 33.3% of entities and more than 61% of relationships in Freebase cannot be simply represented by binary relationships. Medicine is a complex discipline and will face more severe challenges in knowledge modeling. For example, a hypertension knowledge k is: "The initial antihypertensive treatment of patients with chronic kidney disease complicated by hypertension should include an ACEI or ARB", which involves 5 entities, 1 explicit binary relationship initial antihypertensive treatment and 1 implicit relationship disease. Obviously, this knowledge cannot be simply expressed by 2 binary relationships, and the solution is to represent multiple binary relationships (such as Figure 2 (a)), but this leads to the non-independence of a single binary relationship, which is easy to cause ambiguity in knowledge expression. And through the multi-relation (such as Figure 2 (b)) can properly represent the knowledge k.

[0005] Difficulty 2: The general knowledge graph architecture cannot represent logical rules. Since hypertension treatment needs to follow the principles of evidence-based medicine, its core idea is that medical decisions should be based on objective clinical research results as much as possible. Therefore, the modeling of the hypertension knowledge graph must consider the logical expression of diagnosis, treatment rules, and other clinical basis, otherwise the knowledge lacks interpretability. Most of the current knowledge graphs are based on ontology for knowledge organization, and the core description is static facts, with limited ability to express logical judgments. Continue to quote knowledge k as an example, the text contains the logical relationship "ACEI or ARB" (choose one of them in clinical use), and the graph must explicitly represent the or relationship to avoid ambiguity and reduce errors in actual application decisions.

[0006] Difficulty 3: Fine-grained medical text knowledge cannot be completely extracted by automated methods. On the one hand, the medical field is different from other professional fields, and has very high requirements for the accuracy and granularity of knowledge extraction. On the other hand, the current research on relationship extraction mainly stays in simple binary relationship extraction, which is not enough to model the complex semantics of unstructured text in the medical scenario, and the related research on multi-element relationship extraction is not mature enough to extract accurate multi-element relationships (such as Figure 2 ) from knowledge k. SUMMARY

[0007] The present application aims to at least partially solve one of the problems of the related art.

[0008] To this end, the present application proposes a drug recommendation method based on a three-layer super relationship knowledge graph model for realizing a medical rule-based drug recommendation auxiliary decision application.

[0009] To achieve the above-mentioned purpose, the first aspect of the present application proposes a drug recommendation method based on a three-layer super relationship knowledge graph model, comprising:

[0010] Obtaining user attribute information of a target user, the user attribute information including disease conditions, clinical symptoms, physical examinations, and physiological state information;

[0011] Constructing a three-layer super relationship knowledge model based on a super relationship knowledge graph, and inputting the user attribute information into the three-layer super relationship knowledge model to convert it into a search condition of a graph query statement;

[0012] Translating the search condition into a correct graph database query statement, and obtaining a reasoning result by calling a graph database search engine according to a preset reasoning condition;

[0013] Adding the reasoning result to a drug use scheme, obtaining a target drug according to the drug use scheme, and pushing the target drug to the target user.

[0014] In addition, a drug recommendation method based on a three-layer hyper-relationship knowledge graph model according to the above embodiments of the present invention may also have the following additional technical features:

[0015] Furthermore, in one embodiment of the present invention, the three-layer hyper-relationship knowledge model includes a computational layer, a conceptual layer, and an instance layer;

[0016] The instance layer is used to describe factual knowledge and store instance data.

[0017] The conceptual layer is used to describe abstract and conceptual static knowledge, storing disease types, clinical manifestations, physiological states, examination data, drug types, the applicability and contraindications of diseases and drugs, and the interactions between drugs;

[0018] The computational layer is used to describe complete medical rules by logically combining the static knowledge described in the conceptual layer.

[0019] Furthermore, in one embodiment of the present invention, inputting the user attribute information into the three-layer hyper-relationship knowledge model includes:

[0020] User attribute information is obtained, and patient data is stored in 5 subgraph spaces through the three-layer hyper-relationship knowledge model. The 5 subgraphs include an instance layer subgraph, a concept layer subgraph, a computation layer subgraph, an instance-concept relationship subgraph, and a concept-computation relationship subgraph.

[0021] Furthermore, in one embodiment of the present invention, storing patient data in five subgraph spaces through the three-layer hyper-relationship knowledge model includes:

[0022] The instance layer, concept layer, and computation layer subgraphs are responsible for storing the entities and relationships and their attributes within each layer. The instance-concept relationship and concept-computation relationship subgraphs are responsible for storing instance_of, abstract_of, and the entities connected by these two relationships, respectively.

[0023] Furthermore, in one embodiment of the present invention, the step of translating the correct graph database query statement using the search conditions and obtaining the reasoning result by calling the graph database search engine according to preset reasoning conditions includes:

[0024] Transform patient data into a concept layer diagram pattern;

[0025] Based on the user attribute information, write Cypher statements according to predetermined rules;

[0026] The Cypher statement is input into a Neo4j query for pre-filtering to obtain all data that meets the pre-filtering conditions. The data is then grouped by the hyperrelation facts of the computation layer, and each group is considered a complete medical rule.

[0027] Iterate through the rules that meet the conditions;

[0028] Take a rule and determine whether each inference condition matches the user attribute information. If they match, take the true value; otherwise, take the false value.

[0029] The inference result is obtained by calculating the values ​​of the inference conditions and logical operators.

[0030] Furthermore, in one embodiment of the present invention, adding the reasoning result to the medication plan, obtaining the target medication according to the medication plan, and pushing the target medication to the target user includes:

[0031] If the reasoning result is true, add the result to the medication regimen list; otherwise, skip it.

[0032] The list of medication regimens is merged according to the recommendation level, which is divided into: applicable, possibly applicable, insufficient evidence or not applicable, and contraindicated.

[0033] Output drug recommendation results.

[0034] To achieve the above objectives, a second aspect of the present invention proposes a drug recommendation device based on a three-layer hyperrelationship knowledge graph model, comprising:

[0035] The acquisition module is used to acquire user attribute information of the target user, including disease status, clinical symptoms, physical examination, and physiological status information.

[0036] The construction module is used to build a three-layer hyper-relationship knowledge model based on the hyper-relationship knowledge graph, and to input the user attribute information into the three-layer hyper-relationship knowledge model to transform it into search conditions for graph query statements;

[0037] The search module is used to translate the correct graph database query statement using the search conditions, and to obtain the reasoning result by calling the graph database search engine according to the preset reasoning conditions.

[0038] The output module is used to add the reasoning results to the medication plan, obtain the target drug according to the medication plan, and push the target drug to the target user.

[0039] Furthermore, in one embodiment of the present invention, the output module is further configured to:

[0040] Transform patient data into a concept layer diagram pattern;

[0041] Based on the user attribute information, write Cypher statements according to predetermined rules;

[0042] The Cypher statement is input into a Neo4j query for pre-filtering to obtain all data that meets the pre-filtering conditions. The data is then grouped by the hyperrelation facts of the computation layer, and each group is considered a complete medical rule.

[0043] Iterate through the rules that meet the conditions;

[0044] Take a rule and determine whether each inference condition matches the user attribute information. If they match, take the true value; otherwise, take the false value.

[0045] The inference results are calculated based on the values ​​of the inference conditions and logical operators. If the inference result is true, the result is added to the medication plan list; otherwise, it is skipped.

[0046] The list of medication regimens was merged according to the recommendation level, which was divided into: applicable, possibly applicable, insufficient evidence or not applicable, and contraindicated.

[0047] Output drug recommendation results.

[0048] To achieve the above objectives, a third aspect of the present invention provides a computer device, characterized in that it includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the drug recommendation method based on a three-layer hyperrelationship knowledge graph model as described above.

[0049] To achieve the above objectives, a fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the drug recommendation method based on a three-layer hyperrelationship knowledge graph model as described above.

[0050] The drug recommendation method based on a three-layer hyperrelationship knowledge graph model in this invention first introduces the concept representation of multi-relationship knowledge in a hyperrelationship knowledge graph. Second, it decomposes the hypertension knowledge graph into a three-layer architecture: an instance layer, a concept layer, and a computational layer. Compared to a general knowledge graph, a computational layer is added above the concept layer, and a logical representation structure is introduced within the computational layer. Finally, based on unstructured texts such as authoritative hypertension guidelines and clinical studies, and following the principles of evidence-based medicine, a refined and accurate hypertension knowledge graph is constructed using a semi-automatic method. Attached Figure Description

[0051] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0052] Figure 1 This is a flowchart illustrating the drug recommendation method based on a three-layer hyperrelationship knowledge graph model provided in an embodiment of the present invention.

[0053] Figure 2 This is a schematic diagram comparing binary and multivariate relationships provided in an embodiment of the present invention.

[0054] Figure 3 This is a schematic diagram of the super-relationship fact structure provided for an embodiment of the present invention.

[0055] Figure 4 This is a schematic diagram of the main structure of the hypertension superrelationship diagram provided in an embodiment of the present invention.

[0056] Figure 5 This is a schematic diagram of an instance layer provided in an embodiment of the present invention.

[0057] Figure 6 This is a factual illustration of the conceptual layer "patient profile" provided for embodiments of the present invention.

[0058] Figure 7 This is a schematic diagram illustrating the conceptual layer of a "medication regimen" provided in an embodiment of the present invention.

[0059] Figure 8 This is a schematic diagram of the computational layer "medical rule" hyperrelationship fact provided in an embodiment of the present invention.

[0060] Figure 9 This is a schematic diagram of super-relationship fact transformation provided in an embodiment of the present invention.

[0061] Figure 10 This is a schematic diagram of the drug recommendation device based on a three-layer hyperrelationship knowledge graph model provided in an embodiment of the present invention. Detailed Implementation

[0062] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0063] The following describes a drug recommendation method based on a three-layer hyperrelationship knowledge graph model according to an embodiment of the present invention, with reference to the accompanying drawings.

[0064] Figure 1 This is a schematic flowchart of a drug recommendation method based on a three-layer hyperrelationship knowledge graph model provided in an embodiment of the present invention.

[0065] like Figure 1As shown, the drug recommendation method based on a three-layer hyper-relationship knowledge graph model includes the following steps:

[0066] S101: Obtain the user attribute information of the target user, including disease status, clinical symptoms, physical examination, and physiological status information;

[0067] S102: Construct a three-layer hyper-relationship knowledge model based on hyper-relationship knowledge graph, and input user attribute information into the three-layer hyper-relationship knowledge model to transform it into search conditions for graph query statements;

[0068] S103: Translate the correct graph database query statement using the search conditions, and obtain the reasoning result by calling the graph database search engine based on the preset reasoning conditions;

[0069] S104: Add the reasoning results to the medication plan, obtain the target drug according to the medication plan, and push the target drug to the target user.

[0070] Furthermore, in one embodiment of the present invention, the three-layer hyper-relationship knowledge model includes a computational layer, a conceptual layer, and an instance layer;

[0071] The instance layer is used to describe factual knowledge and store instance data.

[0072] The concept layer is used to describe abstract and conceptual static knowledge, storing disease types, clinical manifestations, physiological states, examination data, drug types, the applicability and contraindications of diseases and drugs, and drug interactions;

[0073] The computational layer is used to describe complete medical rules by logically combining the static knowledge described in the conceptual layer.

[0074] Specifically, a hyperrelation knowledge graph refers to a knowledge graph composed of multiple hyperrelation facts. Each hyperrelation fact can be represented by a principal triple plus n additional information, as shown in the specific structure below. Figure 3 As shown. This invention models a three-layer Hypertension Hyper-relational Knowledge Graph (THH-KG), including a computational layer, a conceptual layer, an instance layer, and mapping relationships between adjacent layers. Each layer expresses multi-relational facts through a hyper-relational graph model. The specific structure is as follows: Figure 4 As shown, each layer of the hyperrelationship graph consists of a principal triple (<head entity, principal relation, tail entity>) plus n additional key-value pairs (<additional relation, additional entity>).

[0075] In the three-tier architecture, the instance layer describes factual knowledge and stores instance data, including basic patient information, diseases, examination data, and medication information; it can also be understood as the patient layer, primarily storing case information for hypertensive patients. The instance layer structure is as follows: Figure 5 As shown, the main ternary set represents the disease diagnosed in the patient's case, and the additional information includes: other diseases the patient has, clinical manifestations, physiological status, examination data, and medications previously taken.

[0076] The concept layer describes abstract and conceptual static knowledge, storing information such as disease types, drug types, applicability and contraindications between diseases and drugs, drug interactions, etc. It also describes static knowledge about various types of hypertension. Based on the super-relationship fact representation, knowledge can be mainly classified into three categories: 1) "is a subclass of...", representing the subordinate relationship between concepts. It mainly stores the relationship between concrete entities such as diseases, symptoms, and drugs, abstracted into higher-level concepts, for example, <primary hypertension, is a subclass of..., hypertension>. 2) "Patient profile", describing the characteristics of different types of patients, which is the conceptual expression of patient instance data. The main triple is <patient, diagnosis, disease>, and the additional key-value pairs include: co-existing diseases, clinical manifestations, physiological state, examination data, etc. The structure is similar to the "diagnosis" super-relationship fact in the instance layer. A specific diagram and differences from the instance layer are shown below. Figure 6 As shown in the diagram. 3) "Medication regimen" refers to a recommended medication regimen for a specific patient group, extracted from clinical guidelines, authoritative literature, and other medical texts. The medication regimen may include entity types such as drug classification, drug type, drug chemical name, drug brand name, and combination therapy. A specific illustration is shown below. Figure 7 As shown.

[0077] The computational layer describes the static knowledge described in the conceptual layer into a complete medical rule through logical combination. The main purpose of designing the computational layer is to explicitly express the logical judgments involved in the medical rule, reducing ambiguity and increasing the intuitiveness and interpretability of the knowledge graph. Each rule of hypertension is represented by a computational layer hyperrelational fact. First, the medical logic rule can be abstracted as n inference conditions (con) undergoing logical operations (denoted as *) to derive (->) the inference result (res), i.e., con1*con2*…*conn->res, where con is represented by the head entity and appended entity of the main triple, res is represented by the tail entity of the main triple, "->" is represented by the main relation, and logical operations are represented by appended keys, such as... Figure 8 As shown. When all the reasoning conditions are true after logical operations, the reasoning relation of the principal triple is valid, and the reasoning result is true.

[0078] The connection between the instance layer and the concept layer is represented by a triple (hI, instance_of, tC), where hI is an instance-layer entity and tC is a concept-layer entity. The instance_of triple as a whole expresses the mapping relationship between a concept and its instantiated concrete data, for example (Zhang San, instance_of, patient). It should be noted that some entities are both instance entities and concept entities. For example, hypertension is both a disease concept in the concept layer and the specific disease type suffered by a patient in the instance layer.

[0079] The connection between the conceptual layer and the computational layer is represented by a triple (hC, abstract_of, tR), where hC is the super-relational fact in the conceptual layer and tR is the entity in the computational layer. The abstract_of triple as a whole represents the abstraction of the knowledge of hypertension expressed by the super-relational facts in the conceptual layer into an entity in the computational layer, which then participates in logical operations and the expression of medical rules in the computational layer.

[0080] Furthermore, in one embodiment of the present invention, inputting user attribute information into a three-layer hyper-relationship knowledge model includes:

[0081] User attribute information is obtained, and patient data is stored in 5 subgraph spaces through a three-layer hyper-relationship knowledge model. The 5 subgraphs include instance layer subgraph, concept layer subgraph, computation layer subgraph, instance-concept relationship subgraph, and concept-computation relationship subgraph.

[0082] To address the current lack of hyper-relationship graph database products on the market, this invention proposes a universal storage method applicable to all existing graph database products. While the specific storage requirements for each graph database may need to be adjusted accordingly, the modeling logic remains entirely consistent.

[0083] This invention proposes that the hypertensive super-relationship graph is logically divided into five subgraph spaces for storage: instance layer subgraph, concept layer subgraph, computation layer subgraph, instance-concept relationship subgraph, and concept-computation relationship subgraph. The instance layer, concept layer, and computation layer subgraphs are responsible for storing the entities and relationships within each layer, along with their attributes. The instance-concept relationship and concept-computation relationship subgraphs are responsible for storing `instance_of`, `abstract_of`, and the entities connected to these two relationships, respectively.

[0084] For hyperrelational facts, the method of transforming them into binary relations is as follows: Figure 9 As shown: The super-relation fact is abstracted and added as a new entity as the center. The head entity of the main triple is connected through the "subject" relation, and the tail entity of the main triple is connected through the "object" relation. The additional key-value pairs are connected to the head entity of the super-relation fact in the form of relation + tail entity. The relation of the main triple remains unchanged.

[0085] Furthermore, in one embodiment of the present invention, patient data is stored in five subgraph spaces using a three-layer hyper-relational knowledge model, including:

[0086] The instance layer, concept layer, and computation layer subgraphs are responsible for storing the entities and relationships and their attributes within each layer. The instance-concept relationship and concept-computation relationship subgraphs are responsible for storing instance_of, abstract_of, and the entities connected to these two relationships, respectively.

[0087] This invention designs a graph database-based translation engine (Hypertension Knowledge GraphReasoning Engine, HKG-RE), which transforms patient data in a hyperrelationship knowledge graph into search conditions for graph query statements, translates the correct graph data query statements, calls the graph database search engine, and finally outputs a diagnosis and treatment plan based on the hypertension decision-making process. In this way, the graph search function replaces the model training process for knowledge reasoning.

[0088] Furthermore, in one embodiment of the present invention, the correct graph database query statement is translated using search conditions, and the reasoning result is obtained by calling the graph database search engine according to preset reasoning conditions, including:

[0089] Transform patient data into a concept layer diagram pattern;

[0090] Write Cypher statements based on user attribute information and predetermined rules;

[0091] Input the Cypher statement into Neo4j query for pre-filtering to obtain all data that meet the pre-filtering criteria. Then, group the data into units of hyperrelation facts of the computational layer, and treat each group as a complete medical rule.

[0092] Iterate through the rules that meet the conditions;

[0093] Take a rule and determine whether each inference condition matches the user attribute information. If they match, take the true value; otherwise, take the false value.

[0094] The inference result is obtained by calculating the values ​​of the inference conditions and logical operators.

[0095] Furthermore, in one embodiment of the present invention, the reasoning result is added to the medication regimen, the target medication is obtained according to the medication regimen, and the target medication is pushed to the target user, including:

[0096] If the reasoning result is true, add the result to the medication regimen list; otherwise, skip it.

[0097] The list of medication regimens is merged according to the recommendation level, which is divided into: applicable, possibly applicable, insufficient evidence or not applicable, and contraindicated.

[0098] Output drug recommendation results.

[0099] Taking this rule for hypertension medication as an example: "When a patient has gestational hypertension and a systolic blood pressure ≥180 mmHg or a diastolic blood pressure ≥120 mmHg, the recommended medications are sodium nitroprusside, phentolamine, nicardipine, magnesium sulfate, esmolol, metoprolol, urapidil, diltiazem, hydralazine, and ladoprolol." In the instance layer, the patient is named Zhang San, who has hypertension and is pregnant. Her measured systolic blood pressure (SBP) is 190 mmHg and her diabetic blood pressure (DBP) is 115 mmHg. At the conceptual layer, Zhang San is abstracted as the patient entity. Hypertension and pregnancy remain unchanged. SBP is mapped to the range ≥180 mmHg, and DBP is mapped to <120 mmHg. In the conceptual layer, the conceptual entities SBP≥180 and DBP≥120 form two super-relational facts with the patient, hypertension, and pregnancy, respectively, while DBP<120 cannot form a super-relational fact with the patient, hypertension, or pregnancy. Therefore, the superrelation fact consisting of (patient, hypertension, pregnancy, SBP (mmHg) ≥ 180) is mapped to the head entity of the principal triple in the computational layer and takes a value of true. The superrelation fact consisting of (patient, hypertension, pregnancy, DBP (mmHg) ≥ 120) is mapped to the additional entity in the computational layer, and takes a value of false because the DBP value does not match the patient's condition. Since the additional relation in the computational layer is "OR", the logical operation result is true, therefore the principal triple relation is valid. The recommended medication regimen at this time is: sodium nitroprusside, phentolamine, nicardipine, etc. The doctor only needs to choose one of these drugs as the final medication regimen for patient Zhang San.

[0100] Based on the above scheme, the problem that current knowledge graphs at home and abroad cannot represent logical reasoning relationships can be solved, and breakthrough progress has been made in the structured representation of hypertension clinical pathways.

[0101] The drug recommendation method based on a three-layer hyperrelation knowledge graph model in this invention firstly introduces the concept representation of multi-relation knowledge graphs to solve the problem that ordinary binary knowledge graphs cannot accurately express complex medical knowledge (including multi-relation and logical relations). Compared with other medical knowledge graphs, THH-KG has a finer granularity of knowledge in the field of hypertension, and it is based on authoritative literature such as medical clinical guidelines, which conforms to the principles of evidence-based medicine, thus making the knowledge graph more interpretable. Secondly, the hypertension knowledge graph is decomposed into a three-layer architecture: instance layer, concept layer, and computation layer. Compared with general knowledge graphs, a computation layer is added above the concept layer, and a logical representation structure is introduced in the computation layer. This overcomes the shortcoming that hyperrelation knowledge graph data cannot be persistently stored in graph databases, making THH-KG compatible with graph database products on the market. Finally, based on authoritative hypertension guidelines, clinical research, and other unstructured texts, and following the principles of evidence-based medicine, a refined and accurate hypertension knowledge graph was constructed using a semi-automated method. This solved the problem of how to use THH-KG as a dataset for medication recommendations when large-scale datasets are lacking. The accuracy of its medication recommendations is close to that of expert physicians, providing valuable auxiliary suggestions for doctors in actual diagnosis and treatment scenarios.

[0102] In addition, to demonstrate the feasibility of the above model, a hyper-relationship graph storage method based on a binary relational graph database and a hypertension graph reasoning and interpretation engine are proposed. The hyper-relationship graph storage method addresses the current lack of hyper-relationship graph database products on the market, achieving persistent storage of a multi-level hierarchical knowledge graph of hypertension. The hypertension graph reasoning and interpretation engine transforms the graph data into graph query statements, using graph search functionality to replace the deep learning model's reasoning process for medication recommendations, thus enabling assisted decision-making for medication regimens for various types of hypertension patients.

[0103] Figure 10 This is a schematic diagram of a drug recommendation device based on a three-layer hyperrelationship knowledge graph model, provided in an embodiment of the present invention.

[0104] like Figure 10 As shown, the drug recommendation device based on a three-layer hyper-relationship knowledge graph model includes: an acquisition module 100, a construction module 200, a search module 300, and an output module 400, wherein...

[0105] The acquisition module is used to acquire user attribute information of the target user, including disease status, clinical symptoms, physical examination, and physiological status information.

[0106] The module is used to build a three-layer hyper-relationship knowledge model based on the hyper-relationship knowledge graph, and to input user attribute information into the three-layer hyper-relationship knowledge model to transform it into search conditions for graph query statements;

[0107] The search module is used to translate the correct graph database query statement using search conditions, and to obtain the reasoning result by calling the graph database search engine based on the preset reasoning conditions.

[0108] The output module is used to add the reasoning results to the medication plan, obtain the target drug according to the medication plan, and push the target drug to the target user.

[0109] Furthermore, in one embodiment of the present invention, the output module is also used for:

[0110] Transform patient data into a concept layer diagram pattern;

[0111] Write Cypher statements based on user attribute information and predetermined rules;

[0112] Input the Cypher statement into Neo4j query for pre-filtering to obtain all data that meet the pre-filtering criteria. Then, group the data into units of hyperrelation facts of the computational layer, and treat each group as a complete medical rule.

[0113] Iterate through the rules that meet the conditions;

[0114] Take a rule and determine whether each inference condition matches the user attribute information. If they match, set the value to true; otherwise, set the value to false.

[0115] The inference results are calculated based on the values ​​of the inference conditions and logical operators. If the inference result is true, the result is added to the medication plan list; otherwise, it is skipped.

[0116] The list of medication regimens was merged according to the recommendation level, which was divided into: applicable, possibly applicable, insufficient evidence or not applicable, and contraindicated.

[0117] Output drug recommendation results.

[0118] To achieve the above objectives, a third aspect of the present invention provides a computer device, characterized in that it includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the drug recommendation method based on a three-layer hyperrelationship knowledge graph model as described above.

[0119] To achieve the above objectives, a fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the drug recommendation method based on a three-layer hyperrelationship knowledge graph model as described above.

[0120] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0121] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0122] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A drug recommendation method based on a three-layer hyper-relationship knowledge graph model, characterized in that, Includes the following steps: Obtain user attribute information of the target user, including disease status, clinical symptoms, physical examination, and physiological status information; A three-layer hyper-relationship knowledge model is constructed based on a hyper-relationship knowledge graph, and the user attribute information is input into the three-layer hyper-relationship knowledge model and transformed into search conditions for graph query statements; The correct graph database query statement is translated using the search criteria, and the reasoning result is obtained by calling the graph database search engine based on the preset reasoning conditions; The reasoning results are added to the medication plan, the target drug is obtained according to the medication plan, and the target drug is pushed to the target user. The three-layer hyper-relationship knowledge model includes a computational layer, a conceptual layer, and an instance layer; The instance layer is used to describe factual knowledge and store instance data. The conceptual layer is used to describe abstract and conceptual static knowledge, storing disease types, clinical manifestations, physiological states, examination data, drug types, the applicability and contraindications of diseases and drugs, and the interactions between drugs; The computational layer is used to describe complete medical rules by logically combining the static knowledge described in the conceptual layer. The step of inputting the user attribute information into the three-layer hyper-relationship knowledge model includes: User attribute information is obtained, and patient data is stored in 5 subgraph spaces through the three-layer hyper-relationship knowledge model. The 5 subgraphs include an instance layer subgraph, a concept layer subgraph, a computation layer subgraph, an instance-concept relationship subgraph, and a concept-computation relationship subgraph.

2. The method according to claim 1, characterized in that, The process of storing patient data in five subgraph spaces using the three-layer hyper-relationship knowledge model includes: The instance layer, concept layer, and computation layer subgraphs are responsible for storing the entities and relationships and their attributes within each layer. The instance-concept relationship and concept-computation relationship subgraphs are responsible for storing instance_of, abstract_of, and the entities connected by these two relationships, respectively.

3. The method according to claim 1, characterized in that, The process of translating the correct graph database query statement using the search criteria and deriving the inference result by calling the graph database search engine based on preset inference conditions includes: Transform patient data into a concept layer diagram pattern; Based on the user attribute information, write Cypher statements according to predetermined rules; The Cypher statement is input into a Neo4j query for pre-filtering to obtain all data that meets the pre-filtering conditions. The data is then grouped by the hyperrelation facts of the computation layer, and each group is considered a complete medical rule. Iterate through the rules that meet the conditions; Take a rule and determine whether each inference condition matches the user attribute information. If they match, take the true value; otherwise, take the false value. The inference result is obtained by calculating the values ​​of the inference conditions and logical operators.

4. The method according to claim 1, characterized in that, Adding the inference result to the medication plan, obtaining the target medication according to the medication plan, and pushing the target medication to the target user includes: If the reasoning result is true, add the result to the medication regimen list; otherwise, skip it. The list of medication regimens is merged according to the recommendation level, which is divided into: applicable, possibly applicable, insufficient evidence or not applicable, and contraindicated. Output drug recommendation results.

5. A drug recommendation device based on a three-layer hyper-relationship knowledge graph model, characterized in that, The device is used to implement the drug recommendation method based on a three-layer hyperrelationship knowledge graph model as described in claim 1, and the device includes the following modules: The acquisition module is used to acquire user attribute information of the target user, including disease status, clinical symptoms, physical examination, and physiological status information. The construction module is used to build a three-layer hyper-relationship knowledge model based on the hyper-relationship knowledge graph, and to input the user attribute information into the three-layer hyper-relationship knowledge model to transform it into search conditions for graph query statements; The search module is used to translate the correct graph database query statement using the search conditions, and to obtain the reasoning result by calling the graph database search engine according to the preset reasoning conditions. The output module is used to add the reasoning results to the medication plan, obtain the target drug according to the medication plan, and push the target drug to the target user.

6. The apparatus according to claim 5, characterized in that, The output module is also used for: Transform patient data into a concept layer diagram pattern; Based on the user attribute information, write Cypher statements according to predetermined rules; The Cypher statement is input into a Neo4j query for pre-filtering to obtain all data that meets the pre-filtering criteria. The data is then grouped into units of hyperrelational facts in the computational layer, with each group considered as a complete medical rule. Iterate through the rules that meet the conditions; Take a rule and determine whether each inference condition matches the user attribute information. If they match, take the true value; otherwise, take the false value. The inference results are calculated based on the values ​​of the inference conditions and logical operators. If the inference result is true, the result is added to the medication plan list; otherwise, it is skipped. The list of medication regimens was merged according to the recommendation level, which was divided into: applicable, possibly applicable, insufficient evidence or not applicable, and contraindicated. Output drug recommendation results.

7. A computer device, characterized in that, The device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the drug recommendation method based on a three-layer hyperrelationship knowledge graph model as described in any one of claims 1-4.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the drug recommendation method based on a three-layer hyperrelationship knowledge graph model as described in any one of claims 1-4.

Images