Knowledge graph-based intelligent triage recommendation method and system for intelligent outpatient service

Abstract

The invention relates to a knowledge graph-based intelligent triage recommendation method and system for an intelligent outpatient service. The method comprises the following steps: constructing an outpatient service knowledge unit network containing medical entities, attribute features of the medical entities and association relationships; receiving symptom description information input by a patient, and extracting a corresponding symptom entity combination and attribute features of each symptom entity; performing association path analysis on the symptom entity combination and the attribute features based on an outpatient knowledge unit network, and generating a multi-level association path sequence from the symptom entity combination to the department entities; calculating recommendation priority scores of the department entities according to the association strength characteristics of the association paths, and generating a triage recommendation scheme; and feeding back the triage recommendation scheme to the outpatient triage system, receiving a triage execution result, and updating the association strength characteristics of the corresponding association paths in the outpatient knowledge unit network according to the result. According to the method, intelligent and accurate outpatient triage recommendation can be realized, and the triage efficiency and accuracy are improved.

Description

Technical Field

[0001] This application relates to the fields of data processing and data recommendation technology, and in particular to a smart clinic intelligent triage recommendation method and system based on knowledge graph. Background Technology

[0002] In the current outpatient care process, triage is a crucial link connecting patients with subsequent medical services, and its accuracy and efficiency directly impact the patient's experience and the service quality of medical institutions. However, existing outpatient triage methods have many problems that urgently need to be addressed. Traditional outpatient triage mainly relies on patients' self-assessment or manual guidance from information desk staff, which has significant limitations in practical application. On the one hand, patients often lack professional medical knowledge and find it difficult to accurately determine the department corresponding to their symptoms, easily leading to incorrect registration or going to the wrong department, which not only wastes valuable treatment time but may also worsen their condition due to delays. On the other hand, the professional level and clinical experience of information desk staff vary, and when faced with complex and diverse combinations of symptoms, inaccurate triage may occur due to incomplete information or misjudgment.

[0003] With the advancement of healthcare informatization, some medical institutions have introduced simple symptom query systems, where patients can obtain department recommendations by entering symptom keywords. However, most of these systems are based on pre-set rule bases and lack in-depth analysis of the complex relationships between symptoms and diseases, and between diseases and departments. For example, when a patient has multiple accompanying symptoms, the system cannot effectively analyze the intrinsic connections between symptoms and their multi-level associations with diseases and departments, resulting in a significant reduction in the accuracy of the recommendations. Furthermore, existing systems lag behind in knowledge updates, making it difficult to incorporate new medical research findings, clinical experience, and actual operational data from medical institutions in real time, thus preventing triage recommendations from adapting to constantly changing clinical needs.

[0004] Meanwhile, existing triage methods often only extract single symptom entities when processing symptom information, failing to effectively identify symptom attributes such as frequency, duration, and nature. These attributes are crucial for accurately diagnosing conditions and recommending departments. For example, even with the same headache symptom, persistent severe headaches and intermittent mild headaches may correspond to different diseases and departments, but existing systems struggle to effectively distinguish and utilize these differences. Furthermore, when calculating department recommendation priorities, existing methods typically only consider the direct association between symptoms and departments, ignoring the strength differences in different association paths and the real-time resource status of departments, resulting in recommendations lacking scientific rigor and practicality. Consequently, triage suffers from low accuracy, inefficiency, slow knowledge updates, and an inability to effectively utilize symptom attributes and relationships, failing to meet the demands of smart clinics for precise and intelligent triage services. Summary of the Invention

[0005] In view of the above, in order to at least partially address the shortcomings of the existing technology, in a first aspect, embodiments of this application provide a smart outpatient clinic intelligent triage and recommendation method based on knowledge graphs, the method comprising: Construct an outpatient knowledge unit network that includes medical entities, their attribute features, and relationships. The medical entities include symptom entities, department entities, disease entities, and doctor entities. The attribute features include symptom manifestation features, departmental treatment scope features, disease clinical features, and doctor professional direction features. The relationships include the causal relationship between symptoms and diseases, the affiliation relationship between diseases and departments, and the affiliation relationship between doctors and departments. The system receives symptom description information input by the patient and performs entity extraction processing to obtain the combination of symptom entities corresponding to the symptom description information and the attribute features of each symptom entity. Based on the outpatient knowledge unit network, the association path analysis is performed on the symptom entity combination and the attribute features of each symptom entity to generate a multi-level association path sequence from the symptom entity combination to the department entity. Each association path in the multi-level association path sequence includes the intermediate entity on the path, the type of association relationship between the entities, and the association strength features. The recommendation priority score of the department entity is calculated based on the association strength characteristics of each association path in the multi-level association path sequence, and a triage recommendation scheme containing the recommended department sequence is generated based on the recommendation priority score. The triage recommendation scheme is fed back to the outpatient triage system, and the triage execution result returned by the outpatient triage system is received. The association strength feature of the corresponding associated path in the outpatient knowledge unit network is updated according to the triage execution result.

[0006] Secondly, embodiments of this application also provide a smart outpatient clinic intelligent triage and recommendation system based on knowledge graphs, including a processor and a machine-readable storage medium. The machine-readable storage medium is connected to the processor, and the machine-readable storage medium is used to store programs, instructions, or code. The processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the smart outpatient clinic intelligent triage and recommendation method based on knowledge graphs.

[0007] In summary, the knowledge graph-based intelligent outpatient triage recommendation method and system provided in this application provides comprehensive and structured knowledge support for intelligent triage by constructing an outpatient knowledge unit network that includes medical entities, their attribute features, and relationships. This network covers multiple types of entities, such as symptoms, diseases, departments, and doctors, as well as various relationships between entities, including causal and attribution relationships, and assigns strength features to these relationships. This allows for clear characterization of complex connections between different entities, laying a solid foundation for subsequent path analysis. When receiving patient symptom descriptions and performing entity extraction, this method can accurately identify symptom entities and their attribute features. By combining synonym expansion, exact matching, and fuzzy matching, it improves the accuracy and completeness of symptom entity identification, avoiding information loss or misjudgment due to differences in symptom descriptions. Furthermore, contextual semantic analysis is used to verify candidate symptom entities, further ensuring the consistency between the extraction results and the patient's actual symptoms.

[0008] Furthermore, the association path analysis based on the outpatient knowledge unit network can start from symptom entity combinations and mine multi-level association paths of department entities through breadth-first search and other methods. It fully considers the multi-level associations between symptoms and diseases, diseases and departments, as well as the accompanying relationships between symptoms and the concurrent relationships between diseases. The generated multi-level association path sequence comprehensively reflects the possible paths from symptoms to departments, providing rich evidence for department recommendations. When calculating the department recommendation priority score, this method comprehensively considers the strength of the association path, the severity of symptoms, and the department's workload. Through reasonable weighted calculation and normalization, the scientific and reasonable nature of the score is ensured. Specifically, the calculation of the association path strength considers the weight differences of association relationships at different locations; the assessment of symptom severity combines multiple attribute characteristics of symptoms; and the consideration of department workload makes the recommendation results more consistent with the actual operating conditions of medical institutions, improving the practicality of triage recommendations.

[0009] Furthermore, this application achieves dynamic optimization of the knowledge network by receiving triage execution results and updating the association strength characteristics of corresponding related paths in the outpatient knowledge unit network. With the continuous accumulation of actual clinical data, the knowledge network can continuously learn and evolve, making the strength characteristics of associations more closely match actual clinical situations, thereby continuously improving the accuracy and adaptability of triage recommendations, forming a closed-loop intelligent optimization system. Overall, this application, through the combination of knowledge graph technology and intelligent algorithms, achieves intelligent processing of the entire process from symptom input to department recommendation, effectively improving the accuracy, efficiency, and intelligence level of outpatient triage, reducing the waste of medical resources and patient treatment delays caused by triage errors, and enhancing the patient's medical experience and the service quality of medical institutions.

[0010] Other features and advantages of this application will be described in detail in the following detailed description section. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on the above drawings without creative effort.

[0012] To gain a more complete understanding of this application and its beneficial effects, the following description will be provided in conjunction with the accompanying drawings, wherein the same reference numerals in the following description denote the same parts.

[0013] Figure 1 This is a flowchart illustrating a knowledge graph-based intelligent triage and recommendation method for smart clinics, as provided in an embodiment of this application.

[0014] Figure 2 This is the intended application scenario for the knowledge graph-based intelligent triage and recommendation method for smart clinics provided in the embodiments of this application.

[0015] Figure 3 This is a schematic diagram of a smart clinic intelligent triage and recommendation system based on knowledge graphs provided in an embodiment of this application. Detailed Implementation

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the protection scope of this application.

[0017] Please see Figure 1 and Figure 2 , Figure 1 This is a flowchart illustrating a knowledge graph-based intelligent triage and recommendation method for smart clinics, as provided in an embodiment of this application. Figure 2 This is a schematic diagram illustrating an application scenario of the knowledge graph-based intelligent triage recommendation method for smart clinics provided in this application embodiment. The application scenario includes an intelligent triage platform for information interaction and multiple triage terminals. In this embodiment, the intelligent triage platform can be a server, server cluster, or other device with big data processing and storage capabilities. The triage terminals can be user terminals such as laptops, mobile phones, and tablets, or medical triage terminals installed at medical institutions; this embodiment does not specifically limit their application.

[0018] like Figure 1 As shown, the method includes steps S110-S150, which will be described in detail below.

[0019] Step S110: Construct an outpatient knowledge unit network containing medical entities, their attribute features, and relationships. The medical entities include symptom entities, department entities, disease entities, and doctor entities. The attribute features include symptom manifestation features, departmental treatment scope features, disease clinical features, and doctor's professional direction features. The relationships include the causal relationship between symptoms and diseases, the affiliation relationship between diseases and departments, and the relationship between doctors and departments.

[0020] In this embodiment, S110 may include the following sub-steps S111-S118, which will be described in detail below.

[0021] Step S111: Extract symptom entities, department entities, disease entities, and doctor entities from the medical knowledge base, assign a unique identifier to each entity, and establish an entity basic information table, which includes entity name, entity type, and entity description information.

[0022] For example, a medical knowledge base can be a comprehensive database integrating authoritative medical textbooks, clinical practice guidelines, hospital information system data, etc. Symptom entities extracted from this knowledge base can include headache, cough, fever, abdominal pain, etc.; department entities can include internal medicine, surgery, pediatrics, dermatology, etc.; disease entities can include hypertension, pneumonia, appendicitis, urticaria, etc.; and doctor entities can include Dr. Zhang, Dr. Li, Dr. Wang, etc. A unique identifier is assigned to each entity; for example, the identifier Z1001 is assigned to headache, K2001 to internal medicine, B3001 to hypertension, and Y4001 to Dr. Zhang. In the basic information table of entities, for headache, the entity name is headache, the entity type is symptom entity, and the entity description is the sensation of head pain, which can manifest as throbbing pain, stabbing pain, pulsating pain, etc.; for internal medicine, the entity name is internal medicine, the entity type is department entity, and the entity description is a department that diagnoses and treats diseases of the internal organ system, covering the diagnosis and treatment of diseases of multiple systems such as cardiovascular, respiratory, and digestive; for hypertension, the entity name is hypertension, the entity type is disease entity, and the entity description is a cardiovascular disease characterized by persistently elevated systemic arterial blood pressure; for Dr. Zhang, the entity name is Dr. Zhang, the entity type is doctor entity, and the entity description is a physician with rich clinical experience in the field of cardiovascular diseases.

[0023] Step S112: Extract the symptom manifestation features of each symptom entity, including symptom location features, symptom nature features, symptom frequency features, and symptom duration features.

[0024] For example, for symptom-based headache, the location of the symptom can be a set of multiple possible locations such as the forehead, temples, top of the head, and occipital region, such as {forehead, temples}; the nature of the symptom can be a set of throbbing pain, stabbing pain, pulsating pain, and dull pain, such as {throbbing pain, pulsating pain}; the frequency of symptom attacks can be a multi-dimensional feature consisting of occasional attacks, daily attacks, and several attacks per week, represented by different level labels, such as attack frequency level F1 (occasional attacks), F2 (1-2 times per week), F3 (3-5 times per week), and F4 (daily attacks), and the frequency of headache attacks may be {F2, F3}; the duration of the symptom can be a multi-dimensional feature consisting of several minutes, several hours, and several days, represented by duration level T1 (<30 minutes), T2 (30 minutes - 2 hours), T3 (2-24 hours), and T4 (>24 hours), and the duration of headache attacks may be {T2, T3}.

[0025] Step S113: Extract the departmental treatment scope features of each departmental entity, wherein the departmental treatment scope features include the characteristics of the types of diseases treated, the characteristics of the treatment techniques, and the characteristics of the treatment equipment.

[0026] Taking the internal medicine department as an example, its disease types can be a collection of multiple diseases such as hypertension, coronary heart disease, pneumonia, and gastritis, such as {hypertension, coronary heart disease, pneumonia}; its diagnostic and treatment technology characteristics can be a collection of examinations such as electrocardiogram (ECG), gastroscopy, and respiratory function testing, such as {ECG, gastroscopy}; and its diagnostic and treatment equipment characteristics can be a collection of equipment such as ECG machine, gastroscope, and ventilator, such as {ECG machine, gastroscope}.

[0027] Step S114: Extract the clinical features of each disease entity, including clinical manifestation features, pathogenesis features, diagnostic criteria features, and treatment principle features.

[0028] For hypertension as a disease entity, clinical manifestations can be a set of symptoms such as headache, dizziness, and palpitations, such as {headache, dizziness}; pathogenesis can be a set of factors such as genetic factors, environmental factors, and abnormal neurohumoral regulation, such as {genetic factors, environmental factors}; diagnostic criteria can be multidimensional features such as the range of blood pressure values ​​measured at different times, such as systolic blood pressure ≥140 mmHg and / or diastolic blood pressure ≥90 mmHg; treatment principles can be a set of factors such as lifestyle intervention and drug therapy, such as {lifestyle intervention, drug therapy}.

[0029] Step S115: Extract the doctor's professional direction features for each doctor entity, including features of disease types of expertise, clinical experience, and academic research direction.

[0030] Taking Dr. Zhang as an example, his specialty disease type characteristics could be a set of diseases such as hypertension, coronary heart disease, and heart failure, such as {hypertension, coronary heart disease}; his clinical experience characteristics could be multi-dimensional characteristics such as the number of years he has been engaged in clinical work and the number of patients he has treated, represented by experience levels E1 (less than 5 years), E2 (5-10 years), E3 (10-20 years), and E4 (more than 20 years). Dr. Zhang's clinical experience characteristics might be E3; his academic research direction characteristics could be a set of diseases such as the pathogenesis of hypertension and the prevention and treatment of cardiovascular diseases, such as {pathogenesis of hypertension}.

[0031] Step S116: Define the types of relationships between entities, including causal relationships between symptoms and diseases, affiliation relationships between diseases and departments, affiliation relationships between doctors and departments, concurrent relationships between diseases, and accompaniment relationships between symptoms.

[0032] For example, there is a causal relationship between headache and hypertension; there is a disease-department relationship between hypertension and internal medicine; there is a doctor-department relationship between Dr. Zhang and internal medicine; there may be a disease-complication relationship between hypertension and diabetes; there may be a symptom-accompanying relationship between headache and dizziness.

[0033] Step S117: Set an initial value for the association strength feature for each association type, the initial value of which is determined based on the evidence support in the medical knowledge base.

[0034] In this embodiment, S117 may include the following sub-steps S1171-S1177, which will be described in detail below.

[0035] Step S1171: Collect the number of clinical documents and clinical cases in the medical knowledge base that support each type of association.

[0036] For example, regarding the causal relationship between headache and hypertension, the number of clinical literatures collected is L1, and the number of clinical cases is C1; regarding the relationship between hypertension and internal medicine, the number of clinical literatures is L2, and the number of clinical cases is C2.

[0037] Step S1172: Standardize the number of clinical literatures and the number of clinical cases to obtain literature support and case support.

[0038] Standardization can be achieved using the min-max normalization method, mapping the number of clinical documents to the [0,1] interval. The formula for calculating document support is: Document Support = (L - Lmin) / (Lmax - Lmin), where L is the number of clinical documents in the current association, Lmin is the minimum number of clinical documents in all associations, and Lmax is the maximum number of clinical documents in all associations. Similarly, case support = (C - Cmin) / (Cmax - Cmin), where C is the number of clinical cases in the current association, Cmin is the minimum number of clinical cases in all associations, and Cmax is the maximum number of clinical cases in all associations. After calculation, the document support for the causal relationship between headache and hypertension is W1, and the case support is W2; the document support for the relationship between hypertension and internal medicine affiliation is W3, and the case support is W4.

[0039] Step S1173: Calculate the evidence support for the association based on the literature support and the case support, wherein the evidence support is a weighted average of the literature support and the case support, and the weight of the literature support is higher than the weight of the case support.

[0040] As an example, let the weight of the literature support be 'a', and the weight of the case support be 'b', where a + b = 1 and a > b. For example, a = 0.6 and b = 0.4. Then the evidence support for the causal relationship between headache and hypertension is Z1 = a × W1 + b × W2; the evidence support for the relationship between hypertension and internal medicine affiliation is Z2 = a × W3 + b × W4.

[0041] Step S1174: Map the evidence support to the preset range of association strength feature values ​​to obtain the initial value of the association strength feature for each type of association relationship.

[0042] As an example, if the preset range of the correlation strength feature is [0,1], then the initial value of the correlation strength between headache and hypertension is Q1=Z1 (because Z1 is already in the range of [0,1]); the initial value of the correlation strength between hypertension and internal medicine is Q2=Z2.

[0043] Step S1175: For the causal relationship between symptoms and disease, if the symptoms are specific symptoms of the disease, then increase the initial value of the association strength feature of the causal relationship.

[0044] For example, if headache is one of the specific symptoms of hypertension, the initial value Q1 of the causal relationship between headache and hypertension is increased by a certain proportion, such as 10%, to obtain Q1'=Q1×1.1.

[0045] Step S1176: For the relationship between disease and department, if the disease is a designated disease for treatment by the department, then increase the initial value of the association strength feature of the relationship.

[0046] As an example, assuming that hypertension is a designated disease for internal medicine (which can be understood as a key disease for treatment), the initial value Q2 of the association strength between hypertension and internal medicine is increased by a certain proportion, such as 15%, resulting in Q2' = Q2 × 1.15.

[0047] Step S1177: For the relationship between doctors and departments, if the doctor is the designated doctor of the department, then increase the initial value of the association strength feature of the relationship.

[0048] As an example, if Dr. Zhang is a designated physician in the Department of Internal Medicine (which can be understood as a key physician), the initial association strength between him and the Department of Internal Medicine is Q3. If this is increased by a certain percentage, such as 20%, we get Q3' = Q3 × 1.2.

[0049] Step S118: Construct the topology of the outpatient knowledge unit network based on the entity basic information table, the attribute characteristics of each entity and the association relationship type. In the topology, each node corresponds to a medical entity, and each edge corresponds to an association relationship type and is labeled with the association strength feature.

[0050] As an example, the extracted symptom entities, disease entities, department entities, and doctor entities are used as nodes, and the relationships between entities are used as edges. The edges are labeled with the initial values ​​of the corresponding association strength features to construct a topological structure. For example, there is an edge between node headache (Z1001) and node hypertension (B3001), labeled with causal relationship and association strength Q1'; there is an edge between node hypertension (B3001) and node internal medicine (K2001), labeled with affiliation relationship and association strength Q2'; there is an edge between node Dr. Zhang (Y4001) and node internal medicine (K2001), labeled with affiliation relationship and association strength Q3', etc.

[0051] Step S120: Receive the symptom description information input by the patient and perform entity extraction processing to obtain the symptom entity combination corresponding to the symptom description information and the attribute features of each symptom entity.

[0052] In this embodiment, S120 may include the following sub-steps S121-S127, which will be described in detail below.

[0053] Step S121: Receive symptom description information in text form input by the patient through the outpatient triage system, and perform word segmentation on the symptom description information to obtain multiple word units.

[0054] As an example, the symptom description information input by the patient is "I have had frequent headaches in the past week. The pain is mainly a dull pain in the forehead, and occasionally there is also a little dizziness. It hurts about two or three times a day, and each time it lasts for one or two hours." Perform word segmentation on this text, and use a word segmentation tool to split it into multiple word units, such as "I", "recently", "one week", "frequently", "headache", "mainly", "is", "forehead", "dull pain", "occasionally", "also", "a little", "dizziness", "every day", "about", "hurt", "two or three times", "each time", "last", "one or two", "hours".

[0055] Step S122: Match the multiple word units with the symptom entity names in the outpatient knowledge unit network, and identify the word units that match the symptom entity names as candidate symptom entities.

[0056] In this embodiment, S122 may include the following sub-steps S1221-S1226, which will be introduced in detail below.

[0057] Step S1221: Perform synonym expansion processing on the symptom entity names in the outpatient knowledge unit network, and establish a synonym table for the symptom entity names.

[0058] For example, the synonyms of the symptom entity name "headache" can be "headache", "head pain", etc.; the synonyms of "dizziness" can be "light-headedness", "vertigo", etc., and establish the corresponding synonym table.

[0059] Step S1222: Match the multiple word units with the symptom entity names and their synonym tables respectively, and identify the word units that match exactly as the first candidate symptom entities.

[0060] As an example, among the word units obtained by word segmentation, "headache" exactly matches the symptom entity name "headache", and "dizziness" exactly matches the symptom entity name "dizziness" (exact match). Therefore, "headache" and "dizziness" are used as the first candidate symptom entities.

[0061] Step S1223: For the unmatched word units, perform fuzzy matching processing, calculate the text similarity between the word units and the symptom entity names, and use the word units corresponding to the symptom entity names with a text similarity higher than the preset threshold as the second candidate symptom entities.

[0062] For the unmatched word unit such as "hurt", calculate its text similarity with the symptom entity name "headache". Assuming that the cosine similarity is used for calculation, the similarity is 0.8, and the preset threshold is 0.7. Since 0.8>0.7, the symptom entity "headache" corresponding to "hurt" is used as the second candidate symptom entity.

[0063] Step S1224: Perform deduplication on the first candidate symptom entity and the second candidate symptom entity, and retain the unique symptom entity.

[0064] As an example, the first candidate symptom entity is "headache" and "dizziness", the second candidate symptom entity is "headache", and after removing duplicates, we get "headache" and "dizziness".

[0065] Step S1225: Combine the context of the symptom description information to perform semantic verification on the deduplicated candidate symptom entities and remove candidate symptom entities that do not conform to the context.

[0066] Based on the context "I've had frequent headaches for the past week...and occasionally a bit of dizziness," both "headache" and "dizziness" fit the context, and there are no candidate symptom entities that need to be removed.

[0067] Step S1226: Select the semantically validated candidate symptom entities as the final candidate symptom entities, namely "headache" and "dizziness".

[0068] Step S123: Perform contextual semantic analysis on the candidate symptom entities to determine the semantic role of each candidate symptom entity in the symptom description information. The semantic role includes the main symptom, accompanying symptoms, and triggering factors.

[0069] As an example, analyzing the context reveals that the patient primarily described headaches, while dizziness occurred only occasionally. Therefore, "headache" is the primary symptom, and "dizziness" is a secondary symptom.

[0070] Step S124: Based on the semantic roles, candidate symptom entities are filtered out, and candidate symptom entities whose semantic roles are inducing factors are removed to obtain a combination of symptom entities.

[0071] As an example, since none of the candidate symptom entities have a semantic role as a triggering factor, the symptom entity combination is "headache" and "dizziness".

[0072] Step S125: Extract attribute description information related to each symptom entity in the symptom entity combination from the symptom description information. The attribute description information includes text fragments describing the location of the symptom, text fragments describing the nature of the symptom, text fragments describing the frequency of symptom onset, and text fragments describing the duration of the symptom.

[0073] As an example, for "headache," the relevant attribute description information is as follows: the text fragment describing the location of the symptom is "forehead"; the text fragment describing the nature of the symptom is "throbbing pain"; the text fragment describing the frequency of the symptom is "about two or three times a day"; and the text fragment describing the duration of the symptom is "each episode lasts one or two hours." For "dizziness," since the description does not explicitly mention the location, nature, frequency, and duration of the symptom, the corresponding attribute description information cannot be extracted at this time.

[0074] Step S126: Match the attribute description information with the symptom manifestation features of the corresponding symptom entities, and determine the specific values ​​of symptom location features, symptom nature features, symptom frequency features, and symptom duration features for each symptom entity.

[0075] As an example, for "headache," the symptom location feature matches "forehead"; the symptom nature feature matches "throbbing pain"; the symptom frequency feature, based on "approximately two or three times a day," corresponds to frequency level F3 (3-5 times per week is not suitable; here, we assume daily occurrences, with F3 being 2-3 times per day); and the symptom duration feature, based on "each episode lasts one or two hours," corresponds to duration level T2 (30 minutes - 2 hours). For "dizziness," due to insufficient attribute description information, the specific value cannot be determined at this time and can be marked as pending supplementation.

[0076] Step S127: The symptom location features, symptom nature features, symptom frequency features, and symptom duration features with determined specific values ​​are used as attribute features for each symptom entity, resulting in a combination of symptom entities containing the symptom entity and its attribute features.

[0077] As an example, the final combination of symptom entities is: headache (symptom location feature: forehead; symptom nature feature: throbbing pain; symptom frequency feature: F3; symptom duration feature: T2), dizziness (attribute features to be supplemented).

[0078] Step S130: Based on the outpatient knowledge unit network, perform association path analysis on the symptom entity combination and the attribute features of each symptom entity to generate a multi-level association path sequence from the symptom entity combination to the department entity. Each association path in the multi-level association path sequence includes intermediate entities on the path, the type of association relationship between entities, and the association strength features.

[0079] In this embodiment, S130 may include the following sub-steps S131-S136, which will be described in detail below.

[0080] Step S131: Using each symptom entity in the symptom entity combination as the starting node, perform a breadth-first search in the outpatient knowledge unit network to find disease entities directly associated with the starting node, and record the causal relationship between the symptom entities and the disease entities and the corresponding association strength features.

[0081] As an example, starting with "headache," a breadth-first search is performed in the outpatient knowledge unit network, finding directly related disease entities such as hypertension, migraine, and cervical spondylosis. The causal relationship strength between headache and hypertension is Q1', between headache and migraine is Q4, and between headache and cervical spondylosis is Q5. Starting with "dizziness," directly related disease entities such as hypertension and anemia are found, with the causal relationship strength between dizziness and hypertension being Q6 and between dizziness and anemia being Q7.

[0082] Step S132: Using the found disease entity as the intermediate node, continue to perform a breadth-first search in the outpatient knowledge unit network to find the department entity directly associated with the intermediate node, and record the affiliation relationship between the disease entity and the department entity and the corresponding association strength features.

[0083] As an example, using hypertension as an intermediate node, the associated department entity is internal medicine, with a correlation strength of Q2'; using migraine as an intermediate node, the associated department entity is neurology, with a correlation strength of Q8; using cervical spondylosis as an intermediate node, the associated department entity is orthopedics, with a correlation strength of Q9; and using anemia as an intermediate node, the associated department entity is hematology, with a correlation strength of Q10.

[0084] Step S133: Connect the starting node, intermediate node and ending node (department entity) in the order of the search path to form a secondary association path from the symptom entity to the department entity. The secondary association path includes the symptom entity, the disease entity, the department entity, the causal relationship and association strength characteristics between the symptom entity and the disease entity, and the attribution relationship and association strength characteristics between the disease entity and the department entity.

[0085] As an example, the resulting second-level association path is as follows: Path 1: Headache -> Hypertension -> Internal Medicine, including causal relationship (association strength Q1') and attribution relationship (association strength Q2'). Path 2: Headache -> Migraine -> Neurology, including causal relationship (association strength Q4) and attribution relationship (association strength Q8). Path 3: Headache -> Cervical spondylosis -> Orthopedics, including causal relationship (association strength Q5) and attribution relationship (association strength Q9); Path 4: Dizziness -> Hypertension -> Internal Medicine, including causal relationship (association strength Q6) and attribution relationship (association strength Q2'). Path 5: Dizziness -> Anemia -> Hematology Department, including causal relationship (association strength Q7) and attribution relationship (association strength Q10).

[0086] Step S134: For multiple symptom entities that have a causal relationship in the symptom entity combination, search for common disease entities that have a causal relationship with the symptom entities in the outpatient knowledge unit network, and connect the multiple symptom entities to the department entity through the common disease entity to form a multi-level association path containing multiple starting nodes.

[0087] As an example, the symptoms "headache" and "dizziness" in the symptom entity combination have a causal relationship. The common disease entity that is causally related to both is hypertension. Therefore, a multi-level association path is formed: headache, dizziness -> hypertension -> internal medicine. This path includes the causal relationship between headache and hypertension (Q1'), the causal relationship between dizziness and hypertension (Q6), and the attribution relationship between hypertension and internal medicine (Q2').

[0088] Step S135: When there are multiple associated paths that connect to the same department entity through different intermediate nodes, the associated paths are merged, and the intermediate nodes and relationships in each path are retained to form a composite associated path containing multiple intermediate node branches.

[0089] As an example, the paths connecting to internal medicine include path 1, path 4, and the aforementioned multi-level associated paths. These are merged to form composite associated paths: headache -> hypertension -> internal medicine; dizziness -> hypertension -> internal medicine (the merged path includes these two branches and their corresponding relationships and strengths).

[0090] Step S136: Assign a path identifier to each associated path, record the entity sequence, association relationship type sequence and association strength feature sequence in the path, and generate a multi-level associated path sequence from symptom entity combination to department entity.

[0091] As an example, identifiers are assigned to each path, such as P1 corresponding to path 1, P2 to path 2, P3 to path 3, P4 to path 4, P5 to path 5, P6 to a multi-level association path, and P7 to a composite association path. The entity sequence, association relationship type sequence, and association strength feature sequence for each path are recorded. For example, the entity sequence of P1 is [headache, hypertension, internal medicine], the association relationship type sequence is [causal relationship, attribution relationship], and the association strength feature sequence is [Q1', Q2']; the entity sequence of P6 is [headache, dizziness, hypertension, internal medicine], the association relationship type sequence is [causal relationship, causal relationship, attribution relationship], and the association strength feature sequence is [Q1', Q6, Q2'], etc., thus generating a multi-level association path sequence.

[0092] Step S140: Calculate the recommendation priority score of the department entity based on the association strength characteristics of each association path in the multi-level association path sequence, and generate a triage recommendation scheme containing the recommended department sequence based on the recommendation priority score.

[0093] In this embodiment, S140 may include the following sub-steps S141-S142, which will be described in detail below.

[0094] Step S141: Calculate the recommendation priority score of the department entity based on the association strength characteristics of each association path in the multi-level association path sequence.

[0095] In this embodiment, S141 may include the following sub-steps S1411-S1416, which will be described in detail below.

[0096] Step S1411: For each associated path in the multi-level associated path sequence, extract the association strength features of each association relationship in the path, and form an association strength feature sequence according to the position order of the association relationship in the path.

[0097] For example, the association strength feature sequence of path P1 is [Q1', Q2']; that of path P2 is [Q4, Q8]; that of path P3 is [Q5, Q9]; and that of path P6 is [Q1', Q6, Q2'], etc.

[0098] Step S1412: Perform a weighted summation on the association strength feature sequence, wherein the association strength feature weight at the starting position of the path is higher than the association strength feature weight at subsequent positions, to obtain the path strength value of a single association path.

[0099] As an example, let the weight of the starting position association strength be c1, and that of the subsequent position be c2, and c1 > c2. For example, c1 = 0.6 and c2 = 0.4. For path P1, the path strength value S1 = c1 × Q1' + c2 × Q2'; the path strength value of path P2, S2 = c1 × Q4 + c2 × Q8; S3 of path P3 = c1 × Q5 + c2 × Q9; for path P6, since there are three positions, let the weight of the third position be c3 (c3 < c2), such as c3 = 0.3, and the path strength value S6 = c1 × Q1' + c2 × Q6 + c3 × Q2'.

[0100] Step S1413: For multiple association paths pointing to the same department entity, add up the path strength values of each association path to obtain the total path strength value of this department entity.

[0101] As an example, the paths pointing to the internal medicine department are P1, P4, P6, etc. Assuming the path strength value of P4 is S4, then the total path strength value of the internal medicine department Z_internal_medicine = S1 + S4 + S6; the path pointing to the neurology department is P2, and the total path strength value Z_neurology = S2; the path pointing to the orthopedics department is P3, and the total path strength value Z_orthopedics = S3; the path pointing to the hematology department is P5. Assuming its path strength value is S5, the total path strength value Z_hematology = S5.

[0102] Step S1414: Extract the attribute features of each symptom entity in the symptom entity combination, and calculate the symptom severity score according to the symptom location feature, symptom nature feature, symptom onset frequency feature, and symptom duration feature of the symptom entity. The symptom severity score is positively correlated with the symptom onset frequency feature and positively correlated with the symptom duration feature.

[0103] As an example, for "headache", the symptom onset frequency level is F3, and the corresponding frequency score f3 = 0.6; the duration level is T2, and the corresponding time score t2 = 0.5. The symptom severity score Y_headache = f3 + t2 (here it is weighted addition, and the weights can be set to 0.5 and 0.5, that is, Y_headache = 0.5 × 0.6 + 0.5 × 0.5 = 0.55). For "dizziness", due to incomplete attribute features, it is temporarily calculated according to a lower score, such as Y_dizziness = 0.3.

[0104] Step S1415: Multiply the symptom severity score by the total path strength value of the corresponding department entity to obtain the preliminary recommended priority score of the department entity.

[0105] As an example, the severity score for symptoms in internal medicine is a combined score of headache and dizziness. Assuming the combined score Y_internal = Y_headache + Y_dizziness = 0.55 + 0.3 = 0.85, then the initial recommendation priority score for internal medicine is C_internal_initial = Z_internal medicine × Y_internal medicine; the symptom corresponding to neurology is headache, so C_neurology_initial = Z_neurology × Y_headache; the symptom corresponding to orthopedics is headache, so C_orthopedics_initial = Z_orthopedics × Y_headache; and the symptom corresponding to hematology is dizziness, so C_hematology_initial = Z_hematology × Y_dizziness.

[0106] Step S1416: Query the current patient volume information and doctor resource information of each department entity in the outpatient knowledge unit network, and calculate the department load coefficient based on the current patient volume information and doctor resource information. The department load coefficient is positively correlated with the current patient volume information and negatively correlated with the doctor resource information.

[0107] As an example, the current patient volume for Internal Medicine is A1, and the number of doctor resources is D1; ​​the current patient volume for Neurology is A2, and the number of doctor resources is D2; the current patient volume for Orthopedics is A3, and the number of doctor resources is D3; and the current patient volume for Hematology is A4, and the number of doctor resources is D4. The department load coefficient is calculated as: Load coefficient = Patient volume / Number of doctor resources (normalized to the [0,1] interval). Therefore, the load coefficient for Internal Medicine is F_Internal Medicine = (A1 / D1) / max(A1 / D1,A2 / D2,A3 / D3,A4 / D4); similarly, F_Neurology, F_Orthopedics, and F_Hematology can be obtained.

[0108] Step S1417: Based on the preliminary recommendation priority score of the department entity and the department load coefficient, obtain the final recommendation priority score of the department entity.

[0109] As an example, the initial recommendation priority score for a department entity can be divided by the department load coefficient to obtain the final recommendation priority score for that department entity. For example, the final recommendation priority score for internal medicine is C_internal_final = C_internal_initial / F_internal_; for neurology, C_neurology_final = C_neurology_initial / F_neurology; for orthopedics, C_orthopedics_final = C_orthopedics_initial / F_orthopedics; and for hematology, C_hematology_final = C_hematology_initial / F_hematology.

[0110] Step S142: Generate a triage recommendation scheme containing a sequence of recommended departments based on the recommendation priority score.

[0111] In this embodiment, S142 may include the following sub-steps S1421-S1425, which will be described in detail below.

[0112] Step S1421: Sort the final recommendation priority scores of each department entity in descending order to obtain the department entity ranking sequence.

[0113] As an example, suppose that after calculation and sorting, the department entity sorting sequence is Internal Medicine, Neurology, Orthopedics, Hematology.

[0114] Step S1422: Select a preset number of departmental entities that rank highly from the departmental entity sorting sequence as recommended departments.

[0115] If the preset number is 3, then internal medicine, neurology, and orthopedics are selected as recommended departments.

[0116] Step S1423: For each recommended department, extract all associated paths pointing to the recommended department from the multi-level associated path sequence, analyze the intermediate disease entities in the associated paths, and determine the main disease entities associated with the recommended department.

[0117] As an example, for internal medicine, the intermediate disease entity in the association path is hypertension, so the primary disease entity is hypertension; for neurology, the intermediate disease entity is migraine, so the primary disease entity is migraine; for orthopedics, the intermediate disease entity is cervical spondylosis, so the primary disease entity is cervical spondylosis.

[0118] Step S1424: Query the doctor entities and their professional direction characteristics of the recommended departments in the outpatient knowledge unit network. Based on the matching degree between the main disease entities and the doctor professional direction characteristics, select doctor entities that are good at handling the main disease entities for the recommended departments as recommended doctors.

[0119] As an example, the search for internal medicine doctors includes Dr. Zhang and Dr. Li. Dr. Zhang specializes in hypertension, making him a high-match doctor; Dr. Li does not specialize in hypertension, making him a lower-match doctor. Therefore, Dr. Zhang is recommended for internal medicine. Similarly, Dr. Wang specializing in migraines is selected for neurology, and Dr. Zhao specializing in cervical spondylosis is selected for orthopedics.

[0120] Step S1425: Extract the departmental treatment scope characteristics of the recommended departments and the doctor's professional direction characteristics of the recommended doctors, and generate detailed information about the recommended departments and doctors. The detailed information includes the disease types treated by the departments, the treatment technology characteristics, the disease types the recommended doctors are good at, and the clinical experience characteristics.

[0121] As an example, the detailed information for internal medicine includes: disease types treated include hypertension and coronary heart disease; diagnostic and treatment techniques include electrocardiogram (ECG) examinations. Dr. Zhang's areas of expertise are hypertension and coronary heart disease, and his clinical experience is E3 (10-20 years). Similarly, detailed information for neurology and orthopedics, along with recommended doctors, is generated.

[0122] Step S1426: Combine the recommended department sequence, the main disease entities corresponding to each recommended department, the recommended doctor and detailed information to generate a triage recommendation scheme. The triage recommendation scheme also includes the location information of the recommended departments and the current waiting time information.

[0123] As an example, the triage recommendation scheme includes a recommended department sequence [Internal Medicine, Neurology, Orthopedics]; the main disease entities corresponding to each department [Hypertension, Migraine, Cervical Spondylosis]; recommended doctors [Dr. Zhang, Dr. Wang, Dr. Zhao]; detailed information; and the location information of each department (e.g., Internal Medicine is on the east side of the 3rd floor of the outpatient building, Neurology is on the west side of the 2nd floor of the outpatient building) and the current waiting time information (e.g., the current waiting time for Internal Medicine is 30 minutes, the current waiting time for Neurology is 45 minutes).

[0124] Step S1427: Based on the clinical characteristics of the main disease entities corresponding to the recommended departments, extract the typical symptom features and key points for differential diagnosis of the diseases.

[0125] For example, typical symptoms of hypertension include headache and dizziness; key points for differential diagnosis include distinguishing it from secondary hypertension. Typical symptoms of migraine include unilateral throbbing headache; key points for differential diagnosis include distinguishing it from tension headache. Typical symptoms of cervical spondylosis include neck and shoulder pain and numbness in the upper limbs; key points for differential diagnosis include distinguishing it from frozen shoulder.

[0126] Step S1428: Compare the symptom description information entered by the patient with the typical symptom features of the main disease entities to generate a symptom matching degree analysis report, which includes matching symptom features and non-matching symptom features.

[0127] As examples, when comparing the patient's symptoms with typical symptoms of hypertension, the matching symptom features were headache and dizziness; no unmatching symptom features were found at the moment. When comparing with typical symptoms of migraine, the matching symptom feature was headache; the unmatching symptom feature was unilateral throbbing headache (described by the patient as frontal throbbing pain). When comparing with typical symptoms of cervical spondylosis, the matching symptom feature was headache; the unmatching symptom features included neck and shoulder pain, etc.

[0128] Step S1429: Based on the differential diagnosis points and symptom matching analysis report, generate preliminary medical consultation precautions, which include recommended examination materials and symptom details that need to be described to the doctor.

[0129] As an example, for internal medicine (hypertension), it is recommended to bring previous blood pressure records; you need to describe in detail to the doctor the specific time the headache occurred, and whether it is accompanied by other symptoms such as blurred vision. Similarly, generate the consultation precautions for other departments.

[0130] Step S1430: Combine the current waiting time information and location information of the recommended departments to generate suggestions for the route to the clinic and the best time to visit the clinic.

[0131] As an example, the current wait time for the Internal Medicine department is 30 minutes. The department is located on the east side of the 3rd floor of the outpatient building. It is recommended that the patient take the elevator from their current location (assuming they are in the lobby on the 1st floor of the outpatient building) to the 3rd floor and then walk east. The optimal time to visit is suggested to be during the current time slot, as the wait time is relatively short. Similar suggestions can be generated for other departments.

[0132] Step S1431: Add the symptom matching analysis report, precautions for seeking medical treatment, suggestions for the route to the medical treatment site, and suggestions for the best time to seek medical treatment as medical treatment suggestion information to the triage recommendation plan.

[0133] As an example, the final triage recommendation scheme includes a recommended department sequence, main disease entities, recommended doctors, detailed information, location information, current waiting time information, and consultation suggestions.

[0134] Step S150: Feed back the triage recommendation scheme to the outpatient triage system, receive the triage execution result returned by the outpatient triage system, and update the association strength feature of the corresponding association path in the outpatient knowledge unit network according to the triage execution result.

[0135] In this embodiment, S150 may include the following sub-steps S151-S157, which will be described in detail below.

[0136] Step S151: Send the triage recommendation plan to the outpatient triage system in a structured data format. The structured data format includes a sequence of recommended departments, a priority score for each recommended department, main disease entities, and information on recommended doctors.

[0137] As an example, it is sent in JSON format, for example: { Recommended Department Sequence: ["Internal Medicine", "Neurology", "Orthopedics"] "Recommendation Priority Score":{"Internal Medicine": C_Internal Medicine Final, "Neurology": C_Neurology Final, "Orthopedics": C_Orthopedics Final}, "Main Disease Entities":{"Internal Medicine":"Hypertension","Neurology":"Migraine","Orthopedics":"Cervical Spondylosis"}, Recommended Doctors: {"Internal Medicine": "Dr. Zhang", "Neurology": "Dr. Wang", "Orthopedics": "Dr. Zhao"} } Step S152: Receive the triage execution result returned by the outpatient triage system. The triage execution result includes the actual department selected by the patient, the diagnosis result in that department, and the feedback on the treatment effect.

[0138] As an example, suppose the patient actually chooses the internal medicine department, is diagnosed with hypertension, and the treatment feedback is that it is effective.

[0139] Step S153: Extract the actual department entity and corresponding diagnosis result selected by the patient from the triage execution result, and determine the actual associated disease entity.

[0140] As an example, the actual selected department entity is extracted as internal medicine, and the corresponding diagnosis result is hypertension. Therefore, the actual associated disease entity is hypertension.

[0141] Step S154: In the outpatient knowledge unit network, find the association path from the symptom entity combination to the actually selected department entity and through the actually associated disease entity, and use it as the target association path.

[0142] As an example, the target associated paths found are path 1, path 4, multi-level associated paths, and composite associated paths.

[0143] Step S155: Adjust the association strength features of each relationship in the target association path according to the treatment effect feedback. If the treatment effect feedback is effective, increase the value of the association strength feature; if the treatment effect feedback is ineffective, decrease the value of the association strength feature.

[0144] As an example, if the treatment effect feedback is effective, the correlation strength feature value of each correlation in the target correlation path will be increased by a certain percentage, such as 5%. For example, the correlation strength Q1' between headache and hypertension will be adjusted to Q1'' = Q1' × 1.05; the correlation strength Q6 between dizziness and hypertension will be adjusted to Q6' = Q6 × 1.05; and the correlation strength Q2' between hypertension and internal medicine will be adjusted to Q2'' = Q2' × 1.05.

[0145] Step S156: When a patient selects a department entity that was not recommended in the triage execution results, search for the association path from the symptom entity combination to the department entity in the outpatient knowledge unit network, analyze the reason for not being recommended, and if it is not recommended because the association strength feature is too low, then appropriately increase the association strength feature of the association path.

[0146] As an example, in this case, the patient selected internal medicine from the recommended departments. There were no unrecommended departments selected, so this step will not be performed at this time.

[0147] Step S157: Update the adjusted association strength features to the topology of the outpatient knowledge unit network to complete the update of the knowledge unit network.

[0148] As an example, the adjusted association strength features such as Q1'', Q6', and Q2'' are updated to the corresponding association paths in the outpatient knowledge unit network to complete this knowledge unit network update.

[0149] The aforementioned prediction of the associated path strength can be achieved using a machine learning model. Based on this, the method further includes a model training step, described in steps S210-S240 as follows: Step S210: Collect historical triage data, which includes the patient's symptom description information, the actual department selected, the diagnosis result, and the treatment effect feedback.

[0150] As an example, collect historical triage data over a period of time, such as 1,000 data entries, each containing the above information.

[0151] Step S211: Extract entities and attribute features from the symptom description information in the historical triage data to obtain the historical symptom entity combination and attribute features.

[0152] As an example, the same method as in step S120 is used to process the symptom description information of each historical data, and extract the corresponding symptom entity combination and attribute features.

[0153] Step S212: Construct historical association paths based on the actual selected department entities and diagnosis results in historical data.

[0154] As an example, a historical association path from symptom entity combinations to department entities is constructed based on the symptom entity combinations, diagnosed disease entities, and actually selected department entities in historical data.

[0155] Step S213: Input the historical symptom entity combinations and attribute features, and historical association paths as training data into a preset machine learning model. The machine learning model is used to learn the mapping relationship between symptom entity combinations and attribute features and the strength of association paths.

[0156] As an example, the preset machine learning model can be a random forest model, with the input being the attribute feature vector of historical symptom entity combinations (such as the encoded values ​​of features such as symptom location, nature, frequency of onset, duration, etc.) and the features of historical association paths (such as the features of intermediate disease entities, etc.), and the output being the predicted value of the strength of the association path.

[0157] Step S220: Set the training parameters of the machine learning model, including the learning rate, number of iterations, and regularization coefficient.

[0158] As an example, the learning rate is set to 0.01, the number of iterations is 1000, and the regularization coefficient is 0.001.

[0159] Step S221: Train the machine learning model using the gradient descent algorithm and calculate the loss value between the model's predicted value and the actual associated path strength.

[0160] The loss value is calculated using the mean squared error, and the loss value L = (1 / n) × sum((predicted value - actual value)^2), where n is the number of training samples.

[0161] Step S222: Adjust the model parameters according to the loss value until the loss value is less than the preset threshold or the maximum number of iterations is reached.

[0162] Training stops when the loss value is less than 0.01 or the number of iterations reaches 1000.

[0163] Step S230: Validate the trained machine learning model using a validation dataset, which includes historical triage data that was not used in the training.

[0164] 20% of the historical triage data was selected as the validation dataset and input into the trained model to obtain the validation prediction results.

[0165] Step S231: Calculate the accuracy of the verification prediction results and the actual associated path strength in the verification dataset. If the accuracy is higher than the preset accuracy threshold, the model training is complete.

[0166] Accuracy is calculated by dividing the number of accurately predicted samples by the total number of validation samples. If the accuracy is higher than 80%, the model training is complete.

[0167] Step S232: If the accuracy is lower than the preset accuracy threshold, adjust the training parameters of the model and return to step S220 to retrain.

[0168] For example, adjust the learning rate to 0.005, increase the number of iterations to 1500, and retrain.

[0169] Step S240: Apply the trained machine learning model to predict the strength of the associated path. When new triage data is input, use the trained machine learning model to predict the strength of the associated path and assist in calculating the department recommendation priority score.

[0170] Thus, when calculating the path strength value in step S1412, the calculation result is corrected by combining the associated path strength predicted by the model, thereby improving the accuracy of the department recommendation priority score.

[0171] The method may also include patient privacy protection steps, described below by steps S310-S330: Step S310: When collecting patients' symptom descriptions and other personal information, anonymize the personal identification information and remove sensitive information such as name, ID number, and contact information.

[0172] For example, the patient's name can be replaced with an anonymous identifier, such as P001 or P002, and information such as ID number and contact information can be deleted.

[0173] Step S311: The anonymized data is encrypted and stored using a symmetric encryption algorithm. The key of the symmetric encryption algorithm is managed by a secure key management system.

[0174] As an example, the data is encrypted using the AES encryption algorithm with a key length of 256 bits. The key is stored in a secure key management system and is changed periodically.

[0175] Step S320: During data transmission, use the SSL / TLS protocol for encrypted transmission to ensure that the data is not leaked during transmission.

[0176] As an example, when the outpatient triage system transmits data between itself and other systems, the SSL / TLS protocol is enabled to encrypt the transmitted data.

[0177] Step S321: Set data access permissions so that only authorized personnel can access the patient's relevant data. The authorized personnel must pass identity authentication and permission verification.

[0178] As an example, different access permissions can be set for different roles. For instance, doctors can only access data of the patients they have treated, while administrators have higher access permissions but need to go through multiple authentication methods (such as password + dynamic verification code).

[0179] Step S330: Regularly back up and security audit the data to check for any data leaks or abnormal access.

[0180] As an example, data is backed up weekly, and the backup data is stored on a secure offline storage device. A security audit is performed monthly, reviewing data access logs and checking for any unusual access records.

[0181] Based on the above, such as Figure 3The diagram shown is a schematic of a knowledge graph-based intelligent triage and recommendation system for smart clinics provided in this application. The system includes components such as a processor, a machine-readable storage medium, and input / output devices. The machine-readable storage medium is connected to the processor and is used to store programs, instructions, or code. The processor executes the programs, instructions, or code in the machine-readable storage medium to implement the aforementioned knowledge graph-based intelligent triage and recommendation method for smart clinics. The knowledge graph-based intelligent triage and recommendation system for smart clinics can be understood as the system described in this application. Figure 1 This could be a part of the intelligent triage platform in the application scenario shown, or it could be the intelligent triage platform itself.

[0182] The machine-readable storage medium may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable read-only memory (EPROM), electrically erasable read-only memory (EEPROM), etc. The machine-readable storage medium is used to store a program, which the processor executes upon receiving an execution instruction.

[0183] The processor may be an integrated circuit chip with signal processing capabilities. The processor mentioned above can be, but is not limited to, a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), etc.

[0184] In summary, the knowledge graph-based intelligent outpatient triage recommendation method and system provided in this application provides comprehensive and structured knowledge support for intelligent triage by constructing an outpatient knowledge unit network that includes medical entities, their attribute features, and relationships. This network covers multiple types of entities, such as symptoms, diseases, departments, and doctors, as well as various relationships between entities, including causal and attribution relationships, and assigns strength features to these relationships. This allows for clear characterization of complex connections between different entities, laying a solid foundation for subsequent path analysis. When receiving patient symptom descriptions and performing entity extraction, this method can accurately identify symptom entities and their attribute features. By combining synonym expansion, exact matching, and fuzzy matching, it improves the accuracy and completeness of symptom entity identification, avoiding information loss or misjudgment due to differences in symptom descriptions. Furthermore, contextual semantic analysis is used to verify candidate symptom entities, further ensuring the consistency between the extraction results and the patient's actual symptoms.

[0185] Furthermore, the association path analysis based on the outpatient knowledge unit network can start from symptom entity combinations and mine multi-level association paths of department entities through breadth-first search and other methods. It fully considers the multi-level associations between symptoms and diseases, diseases and departments, as well as the accompanying relationships between symptoms and the concurrent relationships between diseases. The generated multi-level association path sequence comprehensively reflects the possible paths from symptoms to departments, providing rich evidence for department recommendations. When calculating the department recommendation priority score, this method comprehensively considers the strength of the association path, the severity of symptoms, and the department's workload. Through reasonable weighted calculation and normalization, the scientific and reasonable nature of the score is ensured. Specifically, the calculation of the association path strength considers the weight differences of association relationships at different locations; the assessment of symptom severity combines multiple attribute characteristics of symptoms; and the consideration of department workload makes the recommendation results more consistent with the actual operating conditions of medical institutions, improving the practicality of triage recommendations.

[0186] Furthermore, this application achieves dynamic optimization of the knowledge network by receiving triage execution results and updating the association strength characteristics of corresponding related paths in the outpatient knowledge unit network. With the continuous accumulation of actual clinical data, the knowledge network can continuously learn and evolve, making the strength characteristics of associations more closely match actual clinical situations, thereby continuously improving the accuracy and adaptability of triage recommendations, forming a closed-loop intelligent optimization system. Overall, this application, through the combination of knowledge graph technology and intelligent algorithms, achieves intelligent processing of the entire process from symptom input to department recommendation, effectively improving the accuracy, efficiency, and intelligence level of outpatient triage, reducing the waste of medical resources and patient treatment delays caused by triage errors, and enhancing the patient's medical experience and the service quality of medical institutions.

[0187] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not described in detail in a certain embodiment can be referred to in the relevant descriptions of other embodiments. The embodiments, implementation methods, and related technical features of this application can be combined and substituted with each other without conflict. The above are merely preferred embodiments of this application and are not intended to limit this application in any way. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of this application without departing from the scope of the technical solution of this application shall still fall within the scope of the technical solution of this application.

Claims

1. A knowledge graph-based intelligent outpatient service intelligent triage recommendation method, characterized in that, The method comprises: constructing an outpatient knowledge unit network comprising medical entities, attribute features of medical entities, and association relationships, wherein the medical entities comprise symptom entities, department entities, disease entities, and doctor entities, the attribute features comprise symptom manifestation features, department diagnosis and treatment range features, disease clinical features, and doctor professional direction features, and the association relationships comprise cause-effect relationships between symptoms and diseases, attribution relationships between diseases and departments, and affiliation relationships between doctors and departments; receiving symptom description information input by a patient and performing entity extraction processing to obtain a symptom entity combination corresponding to the symptom description information and attribute features of each symptom entity; performing association path analysis on the symptom entity combination and the attribute features of each symptom entity based on the outpatient knowledge unit network to generate a multi-level association path sequence from the symptom entity combination to a department entity, wherein each association path in the multi-level association path sequence comprises intermediate entities on the path, association relationship types between the entities, and association strength features; calculating a recommendation priority score of the department entity according to the association strength features of each association path in the multi-level association path sequence, and generating a triage recommendation scheme comprising a recommended department sequence based on the recommendation priority score; feeding back the triage recommendation scheme to an outpatient triage system and receiving a triage execution result returned by the outpatient triage system, and updating the association strength features of the corresponding association paths in the outpatient knowledge unit network according to the triage execution result.

2. The knowledge graph-based intelligent outpatient service intelligent triage recommendation method according to claim 1, characterized in that, The method comprises: extracting symptom entities, department entities, disease entities, and doctor entities from a medical knowledge base, assigning a unique identifier to each entity, and establishing an entity basic information table; extracting symptom manifestation features of each symptom entity; extracting department diagnosis and treatment range features of each department entity; extracting disease clinical features of each disease entity; extracting doctor professional direction features of each doctor entity; defining association relationship types between entities, wherein the association relationship types comprise at least cause-effect relationships between symptoms and diseases, attribution relationships between diseases and departments, affiliation relationships between doctors and departments, concurrent relationships between diseases, and accompanying relationships between symptoms; setting initial values of association strength features for each association relationship type, wherein the initial values of the association strength features are determined based on evidence support degrees in the medical knowledge base; constructing a topological structure of the outpatient knowledge unit network based on the entity basic information table, the attribute features of each entity, and the association relationship types, wherein each node in the topological structure corresponds to a medical entity, and each edge corresponds to an association relationship type and is labeled with an association strength feature. 3.The knowledge graph-based intelligent outpatient service intelligent triage recommendation method according to claim 2, characterized in that, The method comprises: collecting the number of clinical literatures and the number of clinical cases supporting each association relationship type in the medical knowledge base; standardizing the number of clinical literatures and the number of clinical cases to obtain literature support degrees and case support degrees; calculating evidence support degrees of the association relationships according to the literature support degrees and the case support degrees. Map the evidence support degree to a preset correlation strength characteristic value range to obtain an initial value of the correlation strength characteristic of each correlation type; For the causal relationship between a symptom and a disease, if the symptom is a specific symptom of the disease, the initial value of the correlation strength characteristic of the causal relationship is increased; For the attribution relationship between a disease and a department, if the disease is a specified diagnosis and treatment disease of the department, the initial value of the attribution relationship is increased; For the affiliation relationship between a doctor and a department, if the doctor is a specified doctor of the department, the initial value of the affiliation relationship is increased.

4. The knowledge graph-based intelligent outpatient service intelligent triage recommendation method according to claim 1, characterized in that, The symptom description information input by the patient is received and entity extraction processing is performed to obtain a symptom entity combination corresponding to the symptom description information and attribute characteristics of each symptom entity, including: Receiving text-form symptom description information input by the patient through an outpatient triage system, performing word segmentation processing on the symptom description information to obtain a plurality of word units; Matching the plurality of word units with symptom entity names in an outpatient knowledge unit network, and identifying word units matching the symptom entity names as candidate symptom entities; Performing context semantic analysis on the candidate symptom entities to determine the semantic roles of each candidate symptom entity in the symptom description information, the semantic roles including main symptoms, accompanying symptoms, and inducing factors; Screening the candidate symptom entities based on the semantic roles, removing candidate symptom entities with the semantic role of inducing factors, and obtaining a symptom entity combination; Extracting attribute description information related to each symptom entity in the symptom entity combination from the symptom description information, the attribute description information including text segments describing symptom locations, text segments describing symptom properties, text segments describing symptom onset frequencies, and text segments describing symptom durations; Matching the attribute description information with symptom manifestation characteristics of the corresponding symptom entities to determine specific values of symptom location characteristics, symptom property characteristics, symptom onset frequency characteristics, and symptom duration characteristics for each symptom entity; Determining the specific values of the symptom location characteristics, symptom property characteristics, symptom onset frequency characteristics, and symptom duration characteristics as the attribute characteristics of each symptom entity to obtain a symptom entity combination containing symptom entities and their attribute characteristics.

5. The knowledge graph-based intelligent outpatient service intelligent triage recommendation method according to claim 4, characterized in that, The matching of the plurality of word units with the symptom entity names in the outpatient knowledge unit network to identify word units matching the symptom entity names as candidate symptom entities includes: Performing synonym expansion processing on the symptom entity names in the outpatient knowledge unit network to establish a synonym table of symptom entity names; Matching the plurality of word units with the symptom entity names and their synonym tables respectively to identify word units that completely match as first candidate symptom entities; For unmatched word units, performing fuzzy matching processing to calculate the text similarity between the word units and the symptom entity names, and identifying word units corresponding to symptom entity names with a text similarity higher than a preset threshold as second candidate symptom entities; Performing deduplication processing on the first candidate symptom entities and the second candidate symptom entities to retain unique symptom entities; The context of the symptom description information is combined, and semantic verification is performed on the de-duplicated candidate symptom entity to remove candidate symptom entities that do not match the context; The candidate symptom entity that has passed the semantic verification is used as the final candidate symptom entity.

6. The knowledge graph-based intelligent outpatient service intelligent triage recommendation method according to claim 1, characterized in that, The attribute characteristics of the symptom entity combination and each symptom entity are analyzed based on the outpatient knowledge unit network to generate a multi-level association path sequence from the symptom entity combination to the department entity, including: Each symptom entity in the symptom entity combination is used as a starting node, and a breadth-first search is performed in the outpatient knowledge unit network to find disease entities directly associated with the starting node, and the causal relationship between the symptom entity and the disease entity and the corresponding association strength characteristics are recorded; The disease entity found is used as an intermediate node, and a breadth-first search is continued in the outpatient knowledge unit network to find department entities directly associated with the intermediate node, and the attribution relationship between the disease entity and the department entity and the corresponding association strength characteristics are recorded; The starting node, intermediate node, and terminal node are connected in the order of the search path to form a two-level association path from the symptom entity to the department entity, and the two-level association path includes the symptom entity, disease entity, department entity, causal relationship and association strength characteristics between the symptom entity and the disease entity, attribution relationship and association strength characteristics between the disease entity and the department entity; For multiple symptom entities with accompanying relationships in the symptom entity combination, a common disease entity that has a causal relationship with the symptom entities is found in the outpatient knowledge unit network, and the multiple symptom entities are connected to the department entity through the common disease entity to form a multi-level association path including multiple starting nodes; When there are multiple association paths connected to the same department entity through different intermediate nodes, the association paths are merged to retain the intermediate nodes and association relationships in each path to form a composite association path including multiple intermediate node branches; Each association path is assigned a path identifier, and the entity sequence, association relationship type sequence, and association strength characteristic sequence in the path are recorded to generate a multi-level association path sequence from the symptom entity combination to the department entity.

7. The knowledge graph-based intelligent outpatient service intelligent triage recommendation method according to claim 6, characterized in that, The recommendation priority score of the department entity is calculated according to the association strength characteristics of each association path in the multi-level association path sequence, including: For each association path in the multi-level association path sequence, the association strength characteristics of each association relationship in the association path are extracted, and the association strength characteristics are ordered according to the position of the association relationship in the association path to form an association strength characteristic sequence; The association strength characteristic sequence is processed by weighted summation to obtain the path strength value of a single association path; For multiple association paths pointing to the same department entity, the path strength values of each association path are added to obtain the total path strength value of the department entity; The attribute characteristics of each symptom entity in the symptom entity combination are extracted, and the symptom severity score is calculated according to the symptom site characteristics, symptom nature characteristics, symptom frequency characteristics, and symptom duration characteristics of the symptom entity; The symptom severity score is multiplied by the total path strength value of the corresponding department entity to obtain the preliminary recommendation priority score of the department entity; query the current reception capacity information and the doctor resource information of each department entity in the outpatient knowledge unit network, and calculate the department load coefficient according to the current reception capacity information and the doctor resource information; obtain the final recommendation priority score of the department entity according to the preliminary recommendation priority score of the department entity and the department load coefficient. 8.The knowledge graph-based intelligent outpatient service intelligent triage recommendation method according to claim 7, characterized in that, generating a triage recommendation scheme containing a recommended department sequence based on the recommendation priority score, comprising: sorting the final recommendation priority scores of the department entities in descending order to obtain a department entity sorting sequence; selecting the department entities with a ranking in the top of a preset number from the department entity sorting sequence as recommended departments; for each recommended department, extracting all association paths pointing to the recommended department from the multi-level association path sequence, analyzing the intermediate disease entities in the association paths, and determining the main disease entity associated with the recommended department; querying the doctor entity and the doctor professional direction characteristics of the recommended department in the outpatient knowledge unit network, and selecting the doctor entity skilled in handling the main disease entity as the recommended doctor according to the matching degree between the main disease entity and the doctor professional direction characteristics; extracting the department diagnosis and treatment range characteristics of the recommended department and the doctor professional direction characteristics of the recommended doctor, and generating detailed introduction information of the recommended department and the recommended doctor; combining the recommended department sequence, the main disease entity corresponding to each recommended department, the recommended doctor and the detailed introduction information to generate a triage recommendation scheme, wherein the triage recommendation scheme further contains location information and current waiting time information of the recommended department. 9.The knowledge graph-based intelligent outpatient service intelligent triage recommendation method according to claim 1, characterized in that, feedback the triage recommendation scheme to the outpatient triage system, and receive the triage execution result returned by the outpatient triage system, and update the association strength characteristics of the corresponding association path in the outpatient knowledge unit network according to the triage execution result, comprising: sending the triage recommendation scheme to the outpatient triage system in a structured data format; receiving the triage execution result returned by the outpatient triage system; extracting the department entity actually selected by the patient and the corresponding diagnosis result from the triage execution result, and determining the actually associated disease entity; finding the association path from the symptom entity combination to the actually selected department entity and passing through the actually associated disease entity in the outpatient knowledge unit network as the target association path; adjusting the association strength characteristics of each association relationship in the target association path according to the treatment effect feedback; when there is a situation that an un-recommended department entity is selected by the patient in the triage execution result, finding the association path from the symptom entity combination to the department entity in the outpatient knowledge unit network, analyzing the reason for not being recommended, and if the reason is that the association strength characteristics are too low, increasing the association strength characteristics of the association path; updating the adjusted association strength characteristics to the topology structure of the outpatient knowledge unit network, and completing the update of the knowledge unit network.

10. A knowledge graph-based intelligent outpatient service intelligent triage recommendation system, characterized in that, a processor, a machine readable storage medium, the machine readable storage medium being connected with the processor, the machine readable storage medium being used to store programs, instructions or codes, and the processor being used to execute the programs, instructions or codes in the machine readable storage medium to realize the method in any one of claims 1-9.

Images