Medical institution recommendation method and device, computer equipment and readable storage medium

Abstract

The invention relates to a medical institution recommendation method and device, computer equipment and a readable storage medium. The method comprises the steps of obtaining object information of a target object, and determining medical reference knowledge matched with the object information; determining disease reference information of the target object by integrating the medical reference knowledge; performing disease grading on the target object based on grading reference information associated with the disease reference information, and determining a disease grading result for representing the severity of the disease; and inputting the object information, the disease grading result and the mechanism resource information into a recommendation shunting model, and determining a recommended medical seeing mechanism of the target object based on the output of the recommendation shunting model. By adopting the method, the patient shunting accuracy can be improved, and grading diagnosis and treatment can be promoted.

Description

Technical Field

[0001] This application relates to the field of internet healthcare technology, and in particular to a method, apparatus, computer device, and readable storage medium for recommending medical institutions. Background Technology

[0002] Despite the increasing abundance of medical resources, during peak seasons for certain diseases, such as flu season, large influxes of patients still pose a significant challenge to both tertiary hospitals and primary care clinics. Primary care clinics often lack sufficient testing capabilities and procedures, necessitating the referral of severely ill patients to larger, tertiary hospitals. Meanwhile, tertiary hospitals experience long waiting times during peak periods, and follow-up appointments incur high costs for patients. This process can also lead to cross-infection, causing mild cases to worsen or infecting accompanying persons.

[0003] By triaging patients, the waste of social medical resources and the secondary transmission caused by concentrated patient gatherings can be reduced. Among related technologies, online consultations through internet-based medical platforms are one measure for patient triage. However, internet-based medical platforms still have certain limitations; their reliability and professionalism cannot reach the level of serious medical practice, patients have a lower level of trust in the diagnostic and treatment results provided by internet platforms, resulting in poor triage effectiveness. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, device, computer equipment, and readable storage medium for recommending medical institutions that can improve the accuracy of patient triage, in order to address the aforementioned technical problems.

[0005] Firstly, this application provides a method for recommending medical institutions, including:

[0006] Obtain object information of the target object and determine medical reference knowledge that matches the object information;

[0007] By integrating the aforementioned medical reference knowledge, the disease reference information of the target object is determined;

[0008] Based on the hierarchical reference information associated with the symptom reference information, the target object is classified into symptom levels to determine the symptom classification result used to characterize the severity of the symptom.

[0009] The object information, the disease classification result, and the institution resource information are input into the recommendation and triage model. Based on the output of the recommendation and triage model, the recommended medical institution for the target object is determined.

[0010] Secondly, this application also provides a medical institution recommendation device, comprising:

[0011] The object information acquisition module is used to acquire object information of the target object and determine medical reference knowledge that matches the object information;

[0012] The reference knowledge integration module is used to determine the disease reference information of the target object by integrating the medical reference knowledge.

[0013] The disease grading determination module is used to grade the target object based on the grading reference information associated with the disease reference information, and determine the disease grading result used to characterize the severity of the disease.

[0014] The object recommendation and triage module is used to input the object information, the disease classification result, and the institution resource information into the recommendation and triage model, and determine the recommended medical institution for the target object based on the output of the recommendation and triage model.

[0015] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the steps described above.

[0016] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the steps described above.

[0017] The aforementioned method, apparatus, computer equipment, and readable storage medium for recommending medical institutions acquire object information of the target object. This object information records the target object's basic information and disease-related details. It identifies medical reference knowledge matching the object information, which serves as the basis for diagnosing the target object's disease. By integrating this medical reference knowledge, it determines the target object's disease reference information, which provides a reference for the target object's illness. Based on the hierarchical reference information associated with the disease reference information, it classifies the target object's disease, providing a reference for the severity of the target object's illness. This determines the disease classification result used to characterize the severity of the disease. The object information, disease classification result, and institution resource information are input into a recommendation and triage model. Based on the output of the recommendation and triage model, it determines the recommended medical institution for the target object. Therefore, this application, based on providing disease reference and disease classification for the target object, adapts to the severity of the target object's illness, guiding the target object to a recommended medical institution with sufficient medical resources for diagnosis and treatment, improving the accuracy of triage, and promoting the implementation of hierarchical medical treatment. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a diagram illustrating the application environment of a medical institution recommendation method in one embodiment.

[0020] Figure 2 This is a flowchart illustrating a method for recommending medical institutions in one embodiment;

[0021] Figure 3 This is a flowchart illustrating the process of determining medical reference knowledge that matches object information in one embodiment.

[0022] Figure 4 This is a flowchart illustrating the process of determining medical reference knowledge based on a medical knowledge graph in one embodiment.

[0023] Figure 5 This is a flowchart illustrating a process for determining the disease reference information of a target object by integrating medical reference knowledge in one embodiment.

[0024] Figure 6 This is a flowchart illustrating the process of obtaining hierarchical reference information associated with disease reference information in one embodiment;

[0025] Figure 7 This is a flowchart illustrating the process of determining disease classification results based on a medical classification knowledge base in one embodiment.

[0026] Figure 8 This is a flowchart illustrating a process for classifying a target object based on hierarchical reference information associated with symptom reference information, and determining the symptom classification result used to characterize the severity of the symptom in one embodiment.

[0027] Figure 9 This is a flowchart illustrating the incremental training process of a recommendation traffic splitting model in one embodiment;

[0028] Figure 10 This is a flowchart illustrating the process of inputting object information and disease classification results into a recommendation and triage model in one embodiment, and determining the recommended medical institution for the target object based on the output of the recommendation and triage model.

[0029] Figure 11 This is a schematic diagram of a recommendation and traffic splitting process using a machine learning recommendation and traffic splitting model in one embodiment;

[0030] Figure 12This is a schematic diagram of the recommendation and traffic splitting process using the LLM recommendation and traffic splitting model in one embodiment;

[0031] Figure 13 This is a flowchart illustrating the medical institution recommendation method in another embodiment;

[0032] Figure 14 This is a schematic diagram of the data access process in one embodiment;

[0033] Figure 15 This is a flowchart illustrating the hierarchical triggering of objects in one embodiment;

[0034] Figure 16 This is a schematic diagram illustrating the process of information push and feedback from a medical institution recommendation platform in one embodiment;

[0035] Figure 17 This is a schematic diagram of the process of constructing a transformed dataset in one embodiment;

[0036] Figure 18 This is a schematic diagram illustrating the training process of a machine learning-based medical institution recommendation model in one embodiment.

[0037] Figure 19 This is a structural block diagram of a medical institution recommendation device in one embodiment;

[0038] Figure 20 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0040] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0041] The medical institution recommendation method provided in this application embodiment can be applied to, for example, Figure 1In the application environment shown, user terminal 102 and institution terminal 104 communicate with the medical institution recommendation platform 106 via a network. A data storage system can store the data that the medical institution recommendation platform 106 needs to process. The data storage system can be integrated into the medical institution recommendation platform 106 or placed in the cloud or on other network servers. The medical institution recommendation platform 106 obtains the object information of the target object sent by user terminal 102, determines the medical reference knowledge matching the object information, determines the symptom reference information of the target object by integrating the medical reference knowledge, classifies the symptom of the target object based on the hierarchical reference information associated with the symptom reference information, and determines the symptom classification result used to characterize the severity of the symptom; it inputs the object information, symptom classification result, and institution resource information into the recommendation triage model, and determines the recommended medical institution for the target object based on the output of the recommendation triage model. The institution resource information can be obtained by the medical institution recommendation platform 106 from the institution terminal 104. User terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. Portable wearable devices can be smartwatches, smart bracelets, smart glasses, etc. The medical institution recommendation platform 106 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services.

[0042] It should be noted that the object information, institutional resource information, and data stored in the medical knowledge graph and medical hierarchical knowledge base involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with relevant regulations to ensure the security of user information.

[0043] In one exemplary embodiment, such as Figure 2 As shown, a method for recommending medical institutions is provided, which is then applied to... Figure 1 The following steps, from 202 to 208, will be used as an example to illustrate the process of recommending medical institutions in China.

[0044] Step 202: Obtain the object information of the target object and determine the medical reference knowledge that matches the object information.

[0045] The target object refers to the object that needs to be triaged. Object information describes the current state of the target object and may include basic information, location information, and medical-related information. The collection of object information requires authorization from the user's terminal and does not involve the user's individual privacy information. In one embodiment, basic information includes age, gender, etc., and medical-related information includes medication purchase information, online consultation information, home testing information, etc. Medical reference knowledge is medical knowledge used to assist in disease diagnosis. In one embodiment, medical reference knowledge includes official medical literature, electronic medical records, and medical research results published in authoritative journals, which can provide a basis for determining the type of disease.

[0046] For example, the system obtains the object information of the target object that meets the triggering triage conditions, and queries a medical database for medical reference knowledge that matches the object information. Triggering triage conditions refer to the conditions that the object information must meet to trigger medical triage. These conditions can include multiple judgment rules, such as whether the user has a medication purchase record or an online consultation record within the past three days. The medical database can store medical knowledge from different sources, and the medical knowledge that matches the object information is identified as medical reference knowledge.

[0047] Step 204: By integrating medical reference knowledge, determine the disease reference information of the target subject.

[0048] Disease reference information describes the type of disease a target individual suffers from and the basis for that diagnosis. Integrating medical reference knowledge can be viewed as an information processing operation aimed at extracting relevant information from medical reference knowledge to provide a diagnostic reference for the target individual's disease. In one embodiment, disease reference information includes disease type, diagnostic basis, and the original text cited in the diagnostic basis. The disease type is a highly generalized description, the diagnostic basis can be represented as the source name of the diagnostic reference, and the cited text can be retrieved from a database storing medical reference knowledge.

[0049] For example, medical reference knowledge is sorted by relevance, and medical reference knowledge that does not meet the relevance criteria is filtered out to determine the disease reference information of the target object.

[0050] Step 206: Based on the hierarchical reference information associated with the symptom reference information, classify the target object according to its symptom severity and determine the symptom classification result used to characterize the severity of the symptom.

[0051] Grading reference information is used to determine the severity of a target individual's illness. Disease grading refers to the process of classifying the severity of a disease. The result of disease grading is a general description of the severity of the target individual's illness and may include the disease level, the basis for grading, and the name of the medical literature source from which the grading was referenced.

[0052] For example, based on the hierarchical reference information associated with the symptom reference information, the object information is matched with the hierarchical reference information by similarity to determine the symptom classification result used to characterize the severity of the symptom.

[0053] Step 208: Input the object information, disease classification results and institutional resource information into the recommendation and triage model, and determine the recommended medical institutions for the target object based on the output of the recommendation and triage model.

[0054] Institutional resource information refers to information about the medical resources that a medical institution can provide. In one embodiment, institutional resource information includes the number of available appointments and the queuing status for testing items. The recommendation and triage model is a model used to recommend suitable medical institutions to a target individual. This embodiment does not specifically limit the type of recommendation and triage model; it only needs to be able to output suitable recommended medical institutions based on the individual's information, disease classification results, and institutional resource information. Recommended medical institutions are medical institutions recommended for the target individual to receive treatment. The individual's information may also include the geographical location reported by the user terminal. The recommendation and triage model can output recommended medical institutions that are appropriate in distance and have sufficient medical resources, matching the disease level, based on the disease classification results.

[0055] For example, object information, disease classification results, and institutional resource information are input into the recommendation and triage model to obtain the classification recommendation results output by the model. Based on the classification recommendation results, the recommended medical institutions for the target object are determined. The classification recommendation results can indicate the level of medical institutions that matches the disease classification results, thereby determining the recommended medical institutions from among the medical institutions that meet the corresponding levels.

[0056] In the aforementioned method for recommending medical institutions, the target information is obtained, which records the target's basic information and disease-related details. Medical reference knowledge matching the target information is identified, serving as the basis for diagnosing the target's disease. By integrating this medical reference knowledge, disease reference information is determined, providing a reference for the target's illness. Based on the hierarchical reference information associated with the disease reference information, the target's disease is classified, providing a reference for the severity of the target's illness. This determines the disease classification result used to characterize the severity of the disease. The target information, disease classification result, and institution resource information are input into a recommendation and triage model. Based on the output of the recommendation and triage model, the recommended medical institution for the target is determined. Therefore, this application, by providing disease reference and disease classification for the target, adapts to the severity of the target's illness and guides the target to a recommended medical institution with sufficient medical resources for diagnosis and treatment, improving the accuracy of triage and promoting the implementation of hierarchical medical treatment.

[0057] In one exemplary embodiment, such as Figure 3As shown, determining the medical reference knowledge that matches the object information includes steps 302 to 304.

[0058] Step 302: Obtain the medical knowledge graph that matches the target object. The medical knowledge graph includes multiple nodes.

[0059] A medical knowledge graph is a visual representation of medical knowledge. (See reference...) Figure 4 , Figure 4 This is a flowchart illustrating the process of determining medical reference knowledge based on a medical knowledge graph in one embodiment. The medical knowledge graph in this embodiment includes multiple nodes and edges connecting the nodes. Nodes can correspond to medical entities, and edges can correspond to the relationships between medical entities. Medical entities can be diseases, symptoms, etc.

[0060] In an exemplary embodiment, medical data is acquired, and medical entities and their relationships are extracted. A medical knowledge graph is then constructed based on these entities and relationships. The medical knowledge graph can be a GraphRAG combined with Retrieval-augmented Generation (RAG). Official medical literature, electronic medical records, and cutting-edge medical research findings can be obtained from publicly available medical databases, journal articles, and other sources. The text is then sliced. A Large Language Model (LLM) is used to extract medical entities and their relationships from the text slices based on prompts. This extracted information is then imported into a graph knowledge base such as Neo4j to construct the medical knowledge graph. To ensure the consistency of medical entity names, disambiguation can be performed on the medical entities based on a unified medical language database. For example, all disease entities can be mapped to the official names of ICD-10 (International Classification of Diseases).

[0061] In an exemplary embodiment, the medical knowledge graph can be further segmented into community subgraphs based on various themes using community mining algorithms (such as k-clique, Leidan, etc.). Elements within these subgraphs are highly correlated; for example, diseases and drugs related to fever symptoms will have more medical relationships extracted and thus be clustered. After extracting the community subgraphs, LLM can be used to generate a summary report for each subgraph. The generated summary report contains a summary of information about the nodes in that subgraph. For example, if the elements in the subgraph are all fever-related medical entities and relationships, then the output summary report would be a fever-related medical evidence report. Medical knowledge graphs constructed from a large amount of medical literature cover numerous elements and cannot be efficiently used for retrieval. Segmenting them into community subgraphs can improve retrieval efficiency.

[0062] Step 304 involves semantically matching the keywords extracted from the object information with the node elements of each node to determine the medical reference knowledge of the target object from the medical knowledge graph.

[0063] Node elements refer to the medical entities corresponding to those nodes and the medical relationships that exist within those entities. Keywords extracted from object information can include medical-related fields, such as body temperature, virus test results, and chief complaint of symptoms. Semantic matching of keywords with medical entities and medical relationships can filter out medical entities and relationships with semantic similarity exceeding a threshold, integrating them into medical reference knowledge.

[0064] For example, basic information, chief complaint, medication purchase information, and testing information are extracted from the object information and organized into query text. The query text is then converted into a text vector, and content with similar semantics in GraphRAG is retrieved based on the text vector. The operation of converting query text into text vectors can be performed using an embedding model. Querying in GraphRAG can be done through local or global retrieval, which can be selected according to the actual situation. In local retrieval, medical entities are extracted from the query text, and similar medical entities in GraphRAG are identified. Then, medical relationships, community subgraphs, and original text slices of these medical entities are obtained through reverse retrieval. In global retrieval, similarity matching is performed between the query text and community reports extracted from GraphRAG to find the topic with the highest relevance to the query text, and then relevant elements in the community subgraph are obtained. After retrieval, medical entities and relationships related to the user's query text, community reports containing medical subject information, and the original source can be traced back through the medical knowledge graph. Combined with text similarity ranking, this information is further filtered to obtain medical reference knowledge.

[0065] In this embodiment, by retrieving medical knowledge from the medical knowledge graph, medical reference knowledge can be obtained, which can provide effective reference for determining the disease type of the target object.

[0066] In another exemplary embodiment, determining medical reference knowledge matching the object information includes: acquiring object information of the target object; determining medical knowledge matching the object type of the target object; and filtering medical reference knowledge that matches the object information from the medical knowledge. Medical knowledge can be categorized into different fields based on the medical domain involved, such as human medical knowledge and animal medical knowledge. The object type of the target object refers to the type of medical domain to which the target object belongs. The user terminal can push object information of different target objects to the medical institution recommendation platform. If the object type of the target object is human, the matching medical knowledge is determined to be human medical knowledge. If the object type of the target object is animal, the matching medical knowledge is determined to be animal medical knowledge. From the medical knowledge in the corresponding field, medical reference knowledge that matches the symptom description information in the object information is filtered out.

[0067] In one exemplary embodiment, such as Figure 5 As shown, by integrating medical reference knowledge, the disease reference information of the target object is determined, including steps 502 to 504.

[0068] Step 502: Integrate medical reference knowledge and keywords to obtain fused information.

[0069] The fused information integrates medical reference knowledge and keywords, and can serve as a basis for analyzing the symptoms of a target object. For example, medical reference knowledge can be incorporated into the query text to construct a new query statement, thereby obtaining the fused information.

[0070] Step 504: Analyze the symptom status of the target object based on the fused information to determine the symptom reference information of the target object.

[0071] Symptom status analysis refers to the operation of analyzing the type of symptom suffered by a target individual. For example, prompt words are constructed based on fused information, explicitly specifying the target output as determining the symptom type. These prompt words are then input into an LLM (Label Management Model). The LLM can then perform assisted diagnosis based on medical reference knowledge within the fused information, analyzing the degree of match between the target individual's chief complaint symptoms, test results, and medication records described by the keywords and relevant symptoms in the medical reference knowledge. It can output the target individual's disease type and also provide the basis for the diagnosis and the original text of the application.

[0072] In this embodiment, based on the semantic understanding and logical judgment capabilities of LLM, symptom reference information is obtained, which can be used as a reference for the type of symptom of the target object, and help to provide suitable medical institutions for the target object.

[0073] In one exemplary embodiment, such as Figure 6 As shown, the recommended method by medical institutions also includes steps 602 to 606.

[0074] Step 602: Obtain the medical grading knowledge base; the medical grading knowledge base includes multiple knowledge summaries and the corresponding medical grading knowledge for each knowledge summary.

[0075] Medical classification knowledge refers to medical knowledge related to the grading of the severity of diseases. A medical classification knowledge base is a database storing medical knowledge related to disease classification. (See also...) Figure 7 , Figure 7 This is a flowchart illustrating the process of determining disease classification results based on a medical classification knowledge base in one embodiment. Since a single disease may belong to classification knowledge from multiple sources (e.g., pneumonia caused by influenza can be classified according to both pneumonia guidelines and cold guidelines), and classification knowledge from the same source can classify diseases caused by multiple etiologies (e.g., pneumonia guidelines can be used for influenza pneumonia as well as bacterial pneumonia), the medical classification knowledge base can be configured with a hierarchical storage structure to facilitate text matching. A knowledge summary is a text segment that summarizes the medical classification knowledge of various disease categories. Each knowledge summary can correspond to multiple categories of medical classification knowledge, and multiple knowledge summaries can also correspond to the same category of medical classification knowledge.

[0076] For example, medical classification knowledge can come from officially published medical guidelines or include custom classification rules. Medical classification knowledge can be manually imported into a medical classification knowledge base or extracted from files using an LLM (Local Level Management) library.

[0077] For example, the medical classification knowledge base uses a hierarchical RAG, including search blocks and recall blocks. Each search block can correspond to a knowledge summary, and each recall block can correspond to a category of medical classification knowledge. Recall blocks can be built using an index and content approach, where the index represents the file name of the medical classification knowledge, and the content represents the specific text in the file. There is an N:N correspondence between recall blocks and search blocks. A search block can be simple disease description text, such as "a disease classification guide for the common cold," and it can be associated with multiple recall blocks, which represent classification literature related to the common cold.

[0078] Step 604: By performing text similarity matching between the disease reference information and each knowledge summary, target summaries whose text similarity meets the similarity conditions are determined from each knowledge summary.

[0079] The disease reference information can be input into the embedding model, and text similarity matching can be performed with each knowledge summary in the search block. The knowledge summary whose text similarity meets the similarity criteria is used as the target summary. The similarity criteria can be set according to the actual situation. For example, the similarity criteria can be set to the knowledge summary with the highest similarity, or the knowledge summary with a similarity exceeding a threshold can be set as the target summary. Since the disease types in the disease reference information and the search block use relatively simple disease descriptions, efficient and accurate retrieval can be achieved.

[0080] Step 606: The medical classification knowledge associated with the target summary is identified as the classification reference information associated with the disease reference information.

[0081] Based on the search blocks corresponding to the target abstract, related recall blocks can be found, and the names and specific text content of the corresponding guideline documents can be determined as hierarchical reference information. For example, the medical hierarchical knowledge associated with the target abstract includes at least one hierarchical guideline, and different hierarchical guidelines use different methods and bases for disease hierarchical classification.

[0082] In this embodiment, hierarchical retrieval is performed based on a medical grading knowledge base to obtain grading reference information, thereby improving retrieval efficiency.

[0083] In one exemplary embodiment, such as Figure 8 As shown, based on the hierarchical reference information associated with the symptom reference information, the target object is classified into symptom grades to determine the symptom grade result used to characterize the severity of the symptom, including steps 802 to 804.

[0084] Step 802: Generate graded prompt words based on the object information and the graded reference information associated with the symptom reference information.

[0085] The grading prompts are inputs used to guide the LLM output of disease grading results. The query text and grading reference information from the above embodiments can be concatenated into grading prompts, and the format of the disease grading results output by the LLM can be specified in the grading prompts.

[0086] Step 804: Input the graded prompt words into the large language model so that the large language model can classify the target object into disease levels under the guidance of the graded prompt words and output the disease classification results used to characterize the severity of the disease.

[0087] The LLM (Local Management Module) inputs grading prompts and analyzes the target object's grading results based on various official grading guidelines and custom grading rules from the grading reference information, outputting the final disease grading result for the target object. For example, the disease grading result may include the disease level and the grading criteria. For instance, the disease level classification may include high risk, intermediate risk, low risk, and no medical attention required.

[0088] In this embodiment, based on the LLM output of the disease classification results, subsequent medical advice for the target individuals can be given based on the disease classification results, which helps to improve the accuracy of the recommendation and triage.

[0089] In an exemplary embodiment, after determining the recommended medical institutions for the target object, a review process can be set up to manually review the <object, hospital> recommended combinations output by the recommendation triage model to determine whether these recommended combinations are reasonable.

[0090] In one exemplary embodiment, such as Figure 9 As shown, the recommended method by medical institutions also includes steps 902 to 904.

[0091] Step 902: When the target object seeks medical treatment at the recommended medical institution, construct a training sample containing object information and institution information of the recommended medical institution.

[0092] For example, a platform app can be installed on the user terminal to access the medical institution recommendation platform, and the institution terminal can access the medical institution recommendation platform through a mini-program or data interface. After identifying the recommended medical institution for the target, the medical institution recommendation platform can send push notifications about the recommended medical institution to the user terminal. The push notifications can take the form of, but are not limited to, in-app notifications, phone calls, and SMS messages, to notify the target and send relevant information about the target to the recommended medical institution.

[0093] The medical institution recommendation platform can also track conversion information for recommended combinations, where conversion refers to the effectiveness of the recommendation combination. Conversion information can be obtained by having the target user rate the recommended medical institutions on the platform's app, and the institution's terminal sending feedback via application or API to confirm whether the target user visited the recommended institution within a specified number of days. Therefore, the medical institution recommendation platform can use this conversion information to determine whether the target user should seek medical treatment at the recommended institution.

[0094] For example, the <object, hospital> recommendation combination is used as a training sample, and conversion information is used as a label to annotate the training sample, constructing a complete labeled training sample. The medical institution recommendation platform can obtain the original information of the target object and the recommended medical institution from the database, extract features, object features include age, gender, blood pressure, and the reported geographical location that triggered the push, and institution features include hospital level, number of medical staff, and the number of real-time registrations when the push is triggered, etc., and continuously collect feedback data from user terminals and institution terminals to form <object, hospital, conversion> combinations to construct new training samples. The combined features of <object, hospital> are calculated, such as the straight-line distance between the user terminal's reported address and the hospital address, the target object's historical medical records at the hospital, etc., and the collected conversion information is added as a label.

[0095] Step 904: Use the training set containing training samples to incrementally train the recommendation traffic splitting model to obtain the updated recommendation traffic splitting model.

[0096] For example, the dataset of training samples can be divided into a training set, a test set, and a validation set. The division can be based on a time range or random division. For example, 80% of the labeled data can be randomly selected as the training set, 10% as the test set, and 10% as the validation set.

[0097] For example, the referral model can be an LLM or a machine learning model. Taking incremental training of a machine learning model as an example, supervised learning can be used. The essence of recommending medical institutions is regarded as a binary classification task. The input <object features, hospital features> is fed into the model to predict whether a conversion will occur between the target object and the medical institution. The prediction output can be the probability of conversion. The conversion is determined by setting a probability threshold. The purpose of training is to minimize the error between the predicted conversion and the conversion label.

[0098] For example, there are no specific restrictions on the type of machine learning model. For instance, it could be a tree-based ensemble learning model like LightGBM or Xgboost, which are not sensitive to differences in feature scale (different features, such as age and blood pressure, have completely different value ranges), or a deep learning model like DeepFFM (Field-aware Factorization Machine).

[0099] In this embodiment, by acquiring the conversion information between the target object and the recommended medical institution, the recommendation and triage model is incrementally trained. This allows us to understand the target object's preference for different medical institutions, provide the target object with more suitable recommended medical institutions, and improve the accuracy of triage.

[0100] In one exemplary embodiment, such as Figure 10 As shown, the object information and disease classification results are input into the recommendation triage model. Based on the output of the recommendation triage model, the recommended medical institutions for the target object are determined, including steps 1002 to 1006.

[0101] Step 1002: Input the object information, disease classification results and institutional resource information into the recommendation and triage model to obtain multiple hierarchical medical institutions recommended by the recommendation and triage model.

[0102] Tiered medical institutions refer to medical institutions evaluated and classified according to dimensions such as total resources, testing capabilities, and professionalism. It is understandable that the tiered classification of medical institutions can be achieved in various ways. For example, it can be directly classified according to the current "Hospital Tiered Management Measures," or it can be classified according to the size of the institution and the number of specialists. Lower-level community hospitals are suitable for patients with mild cases, while higher-level tertiary hospitals are suitable for patients with acute and severe cases.

[0103] For example, primary care clinics and graded hospitals can each have their own referral and triage models, using the same model architecture. The referral and triage model for primary care clinics is responsible for recommending patients to primary care clinics and receiving patients whose disease classification results are low-risk. The referral and triage model for graded hospitals is responsible for recommending patients to graded hospitals and receiving patients whose disease classification results are medium- or high-risk, thereby improving the efficiency and accuracy of triage.

[0104] For example, the object information also includes location information. The recommendation triage model recommends multiple graded medical institutions with sufficient medical resources that are appropriately located and match the disease level in the symptom grading results for the target object.

[0105] Step 1004: For each tiered medical institution, determine the conversion rate of the tiered medical institution corresponding to the target object based on the target object's historical medical records.

[0106] Historical medical records document a target individual's visits to medical institutions. Medical institution recommendation platforms can determine whether a conversion has occurred between the target individual and the recommended institution by obtaining feedback data from both user and institution terminals, thus creating a historical medical record. Conversion rate refers to the probability that a target individual chooses a recommended medical institution for treatment. (Refer to...) Figure 11 , Figure 11 This is a schematic diagram of the recommendation and triage process using a machine learning recommendation and triage model in one embodiment. After determining whether the disease classification result is low-risk, the alternative options for accessing primary hospitals or higher-level hospitals are determined. After filtering out medical institutions that are too far away using distance, the combined features of <object, institution> are calculated using object features and institution features constructed using object information, and the conversion rate under this combined feature is determined.

[0107] Step 1006: From the various tiered medical institutions, identify the recommended medical institutions whose conversion rates meet the conversion criteria.

[0108] Conversion criteria refer to the conversion rate conditions that the recommended medical institutions should meet for the corresponding target groups. This can be set to identify tiered medical institutions that exceed a conversion rate threshold as recommended medical institutions. Recommended medical institutions can be grouped according to the target groups, and the N recommended medical institutions with the highest conversion rates for each target group can be obtained and output.

[0109] For example, the medical institution recommendation platform pushes information about recommended medical institutions to user terminals for target users to refer to when seeking medical treatment, and pushes the de-identified target information to institution terminals, monitors the feedback data pushed by user terminals and institution terminals, and determines whether a conversion has occurred.

[0110] In this embodiment, the conversion rate between tiered medical institutions and target individuals is used as one of the bases for triage. This approach can more fully consider the target individuals' preferences for different medical institutions and improve the prediction accuracy of the recommendation triage model.

[0111] In one exemplary embodiment, refer to Figure 12 , Figure 12 This is a schematic diagram of the recommendation and triage process using an LLM recommendation and triage model in one embodiment. First, based on whether the symptom classification result is low-risk, alternative options for accessing primary care hospitals or higher-level hospitals are determined. Then, based on the address in the target information and the real-time location information obtained from the platform APP, distance is used to filter out medical institutions that are too far away. The target information, symptom classification result, and institutional resource information are integrated into recommendation prompts. The LLM selects the N most suitable hospitals and outputs a suitability score between the target target and the medical institutions.

[0112] For example, when there are too many alternative hospitals, a batch recommendation approach can be adopted, dividing the alternative hospitals into multiple groups for LLM to select in parallel, and finally integrating the recommended hospitals from each group and filtering them a second time to select the final recommended medical institution, so as to ensure accuracy and reasoning speed.

[0113] In this embodiment, the LLM-based recommendation does not require prior knowledge. It uses AI's reasoning ability to provide the most suitable hospital for the patient, making it particularly suitable for the early initiation stage where there is a lack of prior knowledge of transformational information.

[0114] In one exemplary embodiment, the medical institution recommendation platform detects object information pushed by the user terminal, determines whether the object information meets preset triggering rules, and identifies the object corresponding to the object information as the target object if the object information meets the triggering rules. The medical institution recommendation platform can periodically monitor the accessed object information and filter the object information of the target objects to be triaged through triggering rules. For example, if a user has purchased medication, conducted online consultations, performed home testing (nucleic acid, blood routine), etc., and obtained test results within 3 days, the triggering rule can be set according to needs to flexibly control the user group triggered for triage.

[0115] In one exemplary embodiment, refer to Figure 13 , Figure 13 This is a flowchart illustrating a medical institution recommendation method in another embodiment, which includes steps S1 to S8.

[0116] S1, Data Access. (Refer to...) Figure 14 , Figure 14 This is a schematic diagram of the data access process in one embodiment. The accessed data includes data from both the user side and the institution side. On the user side, the user terminal accesses object information, such as patient disease information, records and results of medication purchases and home testing. On the institution side, data from primary hospitals and tertiary hospitals is accessed, such as basic institution information, scale, and real-time resources (bed availability, current number of registered appointments, etc.). This data will be collected and processed uniformly on the medical institution recommendation platform.

[0117] S2. Object Classification. First, the target objects are classified into levels (e.g., high, medium, low). This classification process can be completed using an LLM agent. Using pre-collected medical data, such as official clinical literature and custom classification criteria (e.g., cutting-edge patient classification studies not found in the literature), a two-layer object classification agent is constructed. The first layer uses LLM to build a knowledge graph and GraphRAG to perform disease-assisted diagnosis based on object information. The second layer accurately matches classification information (e.g., medical guidelines) based on the auxiliary diagnostic results to complete the object classification.

[0118] S3, object-level triggering. (See reference...) Figure 15 , Figure 15 This is a flowchart illustrating the object-level triggering process in one embodiment. The medical institution recommendation platform monitors users' drug purchase and home testing behaviors to determine whether the triage conditions are triggered. If the triage conditions are triggered, the object information is used for object triage. If the triage conditions are not triggered, it is determined that the target object does not need to seek medical treatment offline.

[0119] S4, LLM Triage. During the object triage initiation phase, due to the lack of prior knowledge for recommendations, triage can be divided into two stages: LLM triage and machine learning triage. In the LLM triage stage, the agent accesses the tagged user information and combines it with dynamic medical institution information. Utilizing the general reasoning capabilities of LLM, it recommends the most suitable medical institution to patients at each level (e.g., recommending mild cases to the nearest primary care clinic, and suspected severe cases to hospitals with available resources). This process is directly completed by LLM.

[0120] S5, Information Push and Feedback. (Refer to...) Figure 16 , Figure 16 This is a flowchart illustrating the information push and feedback process of a medical institution recommendation platform in one embodiment. The platform pushes the <object, institution> recommendation combination output by the recommendation triage model to the user's internet app and the institution's system. A manual review process can be added to the push process. If a patient selects a recommended institution, it is considered a conversion, and this feedback data can be collected. The platform also collects medical information about the target patient from the institution's side.

[0121] S6. Transform and construct the dataset. (Refer to...) Figure 17 , Figure 17 This is a schematic diagram illustrating the process of constructing a conversion dataset in one embodiment. The medical institution recommendation platform can obtain the original information of the target object and the recommended medical institution from the database, extract features, such as object features including age, gender, blood pressure, and the reported geographical location that triggered the push, and institution features including hospital level, number of medical staff, and the number of real-time registrations when the push was triggered. It continuously collects feedback data from user terminals and institution terminals to form <object, hospital, conversion> combinations, and constructs new training samples. The combined features of <object, hospital> are calculated, such as the straight-line distance between the user terminal's reported address and the hospital address, and the target object's historical medical records at that hospital, and the collected conversion information is added as labels.

[0122] S7, Machine Learning Triage Model Training. (Refer to...) Figure 18 , Figure 18 This is a flowchart illustrating the training process of a machine learning-based medical institution recommendation model in one embodiment. The statistical triage template is responsible for building the machine learning recommendation model. Based on real-world conversion rates, it iteratively recommends patients to the institutions with the highest conversion probability (most likely to be visited). Since machine learning also involves adjusting hyperparameters (e.g., learning rate, maximum tree depth), AutoML (Auto Machine Learning) frameworks such as Optuna can be used to search for the optimal hyperparameter combination on the test set to maximize model prediction performance. To further improve prediction accuracy, a model fusion step can be added. First, multiple base models are trained and their hyperparameters tuned using a dataset. Then, the prediction results of each base model are fused using weighted averaging, model stacking, or other methods. After model training and fusion are complete, a validation set is input into the model to calculate the validation error. The error is evaluated to determine if it is within an acceptable range, resulting in the trained recommendation triage model.

[0123] S8. Machine Learning Triage. After the triage initiation phase is completed, as the accuracy of the recommendation system increases, when patient triage is subsequently triggered, patient information will be simultaneously input into both the LLM triage and machine learning triage modules, collaboratively pushing patients to the most suitable medical institutions.

[0124] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0125] Based on the same inventive concept, this application also provides a medical institution recommendation device for implementing the aforementioned medical institution recommendation method. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations in one or more embodiments of the medical institution recommendation device provided below can be found in the limitations of the medical institution recommendation method described above, and will not be repeated here.

[0126] In one exemplary embodiment, such as Figure 19 As shown, a medical institution recommendation device 1900 is provided, including: a subject information acquisition module 1901, a reference knowledge integration module 1902, a disease classification determination module 1903, and a subject recommendation and triage module 1904, wherein:

[0127] The object information acquisition module 1901 is used to acquire object information of the target object and determine medical reference knowledge that matches the object information.

[0128] The reference knowledge integration module 1902 is used to determine the disease reference information of the target object by integrating the medical reference knowledge.

[0129] The symptom grading determination module 1903 is used to grade the target object based on the grading reference information associated with the symptom reference information, and determine the symptom grading result used to characterize the severity of the symptom.

[0130] The object recommendation and triage module 1904 is used to input the object information, the disease classification result and the institution resource information into the recommendation and triage model, and determine the recommended medical institution for the target object based on the output of the recommendation and triage model.

[0131] In an exemplary embodiment, the object information acquisition module 1901 is further configured to: acquire a medical knowledge graph matching the target object, the medical knowledge graph including multiple nodes; perform semantic matching between keywords extracted from the object information and the node elements of each node, and determine the medical reference knowledge of the target object from the medical knowledge graph.

[0132] In an exemplary embodiment, the reference knowledge integration module 1902 is further configured to: integrate medical reference knowledge and keywords to obtain integrated information; and analyze the symptom status of the target object based on the integrated information to determine the symptom reference information of the target object.

[0133] In an exemplary embodiment, the medical institution recommendation device 1900 further includes a hierarchical reference information acquisition module, used to: acquire a medical hierarchical knowledge base; the medical knowledge hierarchical base includes multiple knowledge summaries and medical hierarchical knowledge corresponding to each knowledge summary; determine target summaries whose text similarity meets the similarity condition from each knowledge summary by performing text similarity matching between the disease reference information and each knowledge summary; and determine the medical hierarchical knowledge associated with the target summary as hierarchical reference information associated with the disease reference information.

[0134] In an exemplary embodiment, the symptom grading determination module 1903 is further configured to: generate grading prompt words based on object information and grading reference information associated with symptom reference information; input the grading prompt words into a large language model, so that the large language model, guided by the grading prompt words, performs symptom grading on the target object and outputs a symptom grading result to characterize the severity of the symptom.

[0135] In an exemplary embodiment, the medical institution recommendation device 1900 further includes a model incremental training module, used to: construct training samples containing object information and institution information of the recommended medical institution when the target object seeks medical treatment according to the recommended medical institution; and use the training set containing the training samples to incrementally train the recommendation triage model to obtain an updated recommendation triage model.

[0136] In an exemplary embodiment, the object recommendation and triage module 1904 is further configured to: input object information, disease classification results, and institutional resource information into the recommendation and triage model to obtain multiple tiered medical institutions recommended by the recommendation and triage model; for each tiered medical institution, determine the conversion rate of the tiered medical institution corresponding to the target object based on the target object's historical medical records; and determine the recommended medical institutions whose conversion rates meet the conversion conditions from among the tiered medical institutions.

[0137] The modules in the aforementioned medical institution recommendation device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0138] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 20 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores object information sent by user terminals and institutional resource information sent by institutional terminals. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a medical institution recommendation method.

[0139] Those skilled in the art will understand that Figure 20 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0140] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method embodiments.

[0141] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method embodiments.

[0142] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the above-described method embodiments.

[0143] It should be noted that the information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with relevant regulations.

[0144] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0145] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0146] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for recommending medical institutions, characterized in that, The method includes: Obtain object information of the target object and determine medical reference knowledge that matches the object information; By integrating the aforementioned medical reference knowledge, the disease reference information of the target object is determined; Based on the hierarchical reference information associated with the symptom reference information, the target object is classified into symptom levels to determine the symptom classification result used to characterize the severity of the symptom. The object information, the disease classification result, and the institution resource information are input into the recommendation and triage model. Based on the output of the recommendation and triage model, the recommended medical institution for the target object is determined.

2. The method according to claim 1, characterized in that, The determination of medical reference knowledge matching the object information includes: Obtain the medical knowledge graph that matches the target object, the medical knowledge graph including multiple nodes; Keywords extracted from the object information are semantically matched with the node elements of each node to determine the medical reference knowledge of the target object from the medical knowledge graph.

3. The method according to claim 2, characterized in that, The process of integrating the medical reference knowledge to determine the disease reference information of the target object includes: By integrating the aforementioned medical reference knowledge and the aforementioned keywords, fused information is obtained; Based on the analysis of the symptom status of the target object using the fused information, reference information about the target object's symptoms is determined.

4. The method according to claim 1, characterized in that, The method further includes: Obtain a medical grading knowledge base; the medical grading knowledge base includes multiple knowledge summaries and the medical grading knowledge corresponding to each knowledge summary; By performing text similarity matching between the disease reference information and each of the knowledge summaries, target summaries whose text similarity meets the similarity conditions are determined from each of the knowledge summaries; The medical classification knowledge associated with the target summary is identified as the classification reference information associated with the disease reference information.

5. The method according to claim 1, characterized in that, The step of classifying the target object based on the hierarchical reference information associated with the symptom reference information, and determining the symptom classification result used to characterize the severity of the symptom, includes: Based on the object information and the hierarchical reference information associated with the symptom reference information, hierarchical prompt words are generated; The grading prompts are input into a large language model, which, guided by the grading prompts, performs symptom grading on the target object and outputs symptom grading results that characterize the severity of the symptom.

6. The method according to claim 1, characterized in that, The method further includes: When the target object seeks medical treatment at the recommended medical institution, a training sample is constructed that includes the object's information and the institution's information. The recommendation traffic splitting model is incrementally trained using a training set containing the training samples to obtain an updated recommendation traffic splitting model.

7. The method according to claim 1, characterized in that, The step of inputting the object information and the disease classification result into the recommendation and triage model, and determining the recommended medical institution for the target object based on the output of the recommendation and triage model, includes: Input the object information, the disease classification result, and the institution resource information into the recommendation and triage model to obtain multiple tiered medical institutions recommended by the recommendation and triage model; For each of the aforementioned tiered medical institutions, the conversion rate of the tiered medical institution corresponding to the target object is determined based on the target object's historical medical records; From the various tiered medical institutions, identify the recommended medical institutions whose conversion rates meet the conversion criteria.

8. A medical institution recommendation device, characterized in that, The device includes: The object information acquisition module is used to acquire object information of the target object and determine medical reference knowledge that matches the object information; The reference knowledge integration module is used to determine the disease reference information of the target object by integrating the medical reference knowledge. The disease grading determination module is used to grade the target object based on the grading reference information associated with the disease reference information, and determine the disease grading result used to characterize the severity of the disease. The object recommendation and triage module is used to input the object information, the disease classification result, and the institution resource information into the recommendation and triage model, and determine the recommended medical institution for the target object based on the output of the recommendation and triage model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images