Physical examination item recommendation method and device based on large language model, equipment and medium

Abstract

According to the physical examination item recommendation method, device and equipment based on the large language model and the computer readable storage medium, interactive physical examination questions and answers are performed with the target user according to the initial risk factors and the related question list corresponding to the candidate disease information list, the collection efficiency of the dynamic health information of the target user is improved, and the user experience is improved. And mining dynamic health information. In combination with dynamic health information, symptom information and health risk factors are further mined, and a disease information list of the target user is enriched, so that personalized physical examination item recommendation of the target user is realized. And the basic health information of the target user is combined with the dynamic health information, and the potential disease information of the target user is determined comprehensively, so that the physical examination item information more adaptive to the target user is obtained according to the comprehensive target disease information list, and the recommendation efficiency and recommendation accuracy of the physical examination item information are improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method, device, computer equipment and computer-readable storage medium for recommending physical examination items based on a large language model. Background Art

[0002] With the improvement of people's living standards and the enhancement of national health awareness, physical examinations have become a rigid need for people to understand their own health status, discover early clues to diseases, and identify health risks. At present, the selection of physical examination packages is usually based on the gender, age and economic conditions of the examinee, and is selected from multiple standardized physical examination packages preset by medical institutions. However, the physical examination items of this physical examination model are usually fixed and fail to be fully determined in combination with the examinee's personal health status. Therefore, for the examinee, the fixed physical examination items are less compatible with themselves.

[0003] The current personalized physical examination item recommendation method on the market needs to recommend corresponding physical examination items based on the basic information, questionnaires and historical medical data of the examinee. However, the above-mentioned physical examination item recommendation method is usually mainly based on a preset questionnaire for recommendation. The information acquisition method of the questionnaire not only has low information collection efficiency, but also the questionnaire cannot cover all health risk information of different examinees, resulting in low accuracy of physical examination recommendation items.

[0004] Therefore, how to solve the low accuracy and efficiency of current physical examination item recommendations has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] The main purpose of the present invention is to provide a method, device, equipment and computer-readable storage medium for recommending physical examination items based on a large language model, aiming to solve the current technical problems of low accuracy and low efficiency in recommending physical examination items.

[0006] To achieve the above object, the present invention provides a method for recommending physical examination items based on a large language model, and the method for recommending physical examination items comprises the following steps:

[0007] Based on the basic health information of the target user, generate the initial risk factors of the target user, and based on the preset disease factor association table, generate a list of candidate disease information corresponding to the initial risk factors;

[0008] Based on the target large language model, a list of related questions corresponding to the initial risk factors and the candidate disease information list is generated, and based on the list of related questions, an interactive physical examination question inquiry is performed to the target user to obtain dynamic health information corresponding to the target user;

[0009] generating target symptom information and target risk factors based on the dynamic health information, and updating the candidate disease information list based on the target symptom information and target risk factors to obtain a target disease information list;

[0010] Based on the physical examination item recommendation model and the target disease information list, target physical examination item information corresponding to the target user is generated.

[0011] In addition, to achieve the above-mentioned purpose, the present invention also provides a physical examination item recommendation device based on a large language model, and the physical examination item recommendation device based on a large language model includes:

[0012] An information list generating module, used to generate initial risk factors of the target user based on the basic health information of the target user, and to generate a candidate disease information list corresponding to the initial risk factors based on a preset disease factor association table;

[0013] A dynamic information generation module, for generating a list of related questions corresponding to the initial risk factors and the candidate disease information list based on a target large language model, and performing interactive physical examination question inquiries to the target user based on the list of related questions to obtain dynamic health information corresponding to the target user;

[0014] An information list updating module, used to generate target symptom information and target risk factors based on the dynamic health information, and update the candidate disease information list based on the target symptom information and target risk factors to obtain a target disease information list;

[0015] The physical examination item recommendation module is used to generate target physical examination item information corresponding to the target user based on the physical examination item recommendation model and the target disease information list.

[0016] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer device, which includes a processor, a memory, and a physical examination item recommendation program based on a large language model stored in the memory and executable by the processor, wherein when the physical examination item recommendation program based on a large language model is executed by the processor, the steps of the physical examination item recommendation method based on a large language model as described above are implemented.

[0017] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, on which is stored a physical examination item recommendation program based on a large language model, wherein when the physical examination item recommendation program based on a large language model is executed by a processor, the steps of the physical examination item recommendation method based on a large language model as described above are implemented.

[0018] The present application provides a method for recommending physical examination items based on a large language model. The method generates the initial risk factors of the target user based on the basic health information of the target user, and generates a list of candidate disease information corresponding to the initial risk factors based on a preset disease factor association table; generates a list of related questions corresponding to the initial risk factors and the candidate disease information list based on the target large language model, and performs interactive physical examination question inquiries to the target user based on the list of related questions to obtain the dynamic health information corresponding to the target user; generates target symptom information and target risk factors based on the dynamic health information, and updates the candidate disease information list based on the target symptom information and the target risk factors to obtain the target disease information list; generates the target physical examination item information corresponding to the target user based on the physical examination item recommendation model and the target disease information list. In the above manner, the present application first screens out the initial risk factors through the basic health information of the target user, and determines the candidate disease information list of the target user according to the initial risk factors. Generate a list of related questions corresponding to the initial risk factors and the candidate disease information list through the target large language model, conduct interactive physical examination Q&A with the target user according to the list of related questions, and mine the dynamic health information of the target user according to the Q&A results. Combined with dynamic health information, the target user's symptom information and health risk factors are further mined to further enrich the target user's disease information list. Finally, based on the target disease information list, personalized physical examination item recommendations are implemented for the target user. Through interactive physical examination questions, the efficiency of collecting dynamic health information of the target user is improved, thereby improving the efficiency of physical examination item recommendations. The target user's basic health information is combined with dynamic health information to comprehensively determine the target user's potential disease information. Based on the comprehensive target disease information list, physical examination item information that is more suitable for the target user is obtained, which improves the accuracy of physical examination item information recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flowchart of the first embodiment of the method for recommending physical examination items based on a large language model of the present invention.

[0020] Figure 2 A schematic diagram of a traditional physical examination item recommendation process provided by an embodiment of the present invention.

[0021] Figure 3 A schematic diagram of an interactive physical examination question inquiry process provided by an embodiment of the present invention.

[0022] Figure 4 It is a schematic diagram of a physical examination item recommendation device based on a large language model of the present invention.

[0023] Figure 5The figure is a schematic diagram of the hardware structure of a computer device involved in an embodiment of the present invention.

[0024] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0026] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0027] It should be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0028] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0029] The method for recommending physical examination items based on a large language model involved in the embodiment of the present invention is mainly applied to a computer device, which may be a device with display and processing functions such as a PC, a portable computer, or a mobile terminal.

[0030] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0031] Reference Figure 1 , Figure 1 It is a flowchart of the first embodiment of the method for recommending physical examination items based on a large language model of the present invention.

[0032] like Figure 1 As shown, an embodiment of the present invention provides a method for recommending physical examination items based on a large language model. The method for recommending physical examination items based on a large language model includes steps S10 to S40.

[0033] Step S10: generating initial risk factors of the target user based on the basic health information of the target user, and generating a candidate disease information list corresponding to the initial risk factors based on a preset disease factor association table;

[0034] The current physical examination recommendation program mainly identifies the potential health problems of the examinee based on the examinee's basic information, questionnaires, and historical medical records, and recommends corresponding physical examination items based on the potential health problems. The existing physical examination recommendation program has the following problems:

[0035] 1. Limitations of information collection. Existing solutions collect relevant data based on initial personal information and preset questionnaires to recommend physical examination items. However, the questions in the questionnaire are limited and cannot cover all potential health risks of the examinee. Due to the fixed nature of the questionnaire and the shallow level of inquiry, the health information collected from the examinee is not comprehensive enough, making it difficult to gain an in-depth understanding of the examinee's health status.

[0036] 2. Limitations of recommendation algorithms that rely on pre-designed knowledge bases. Existing methods rely on pre-set knowledge bases and rules, which are difficult to effectively handle unexpected diseases or different health information of examinees, resulting in the lack of comprehensiveness and pertinence of recommended physical examination items.

[0037] 3. Insufficient personalization. Due to the lack of dynamic interaction with the examinee, the existing physical examination item recommendation scheme relies on the user's static health data, making it difficult to obtain and update the examinee's latest health status in real time. In addition, the examinee's sense of participation is insufficient, and the recommended physical examination items are difficult to adjust according to the examinee's actual needs, resulting in limited personalization.

[0038] In order to solve the above problems, this application adopts a heuristic interaction strategy based on the examinee's own health status, and combines the medical knowledge base and large language model to actively ask targeted questions to the examinee, so as to further explore the possible health risks of the examinee. In this way, the corresponding physical examination items are recommended based on the examinee's actual health risks, and personalized recommendations for physical examination items are achieved, which improves the efficiency and accuracy of recommending physical examination items.

[0039] Specifically, Figure 2 As shown in the figure, the questionnaire information of the target user can be obtained through the resident health questionnaire, so as to obtain the basic health information of the examinee (i.e. any user) from the questionnaire information. Basic health information includes but is not limited to basic information (such as gender, age, occupation), past medical history, family medical history and living habits. Based on the basic health information provided by the examinee, a rule-based comparison method is used to analyze the correspondence between the questionnaire results and known diseases and their associated risk factors.

[0040] By analyzing the basic health information of the examinee, the symptoms and potential disease list are preliminarily evaluated. The preliminarily evaluated symptoms and potential disease list are analyzed through the physical examination big model to obtain a list of relevant questions. Then, based on the list of relevant questions, the model query is initiated to the target user, that is, multiple rounds of dialogue are conducted with the target user to obtain the user's answer, so as to mine more health-related information from the user's answer.

[0041] It is understandable that in addition to obtaining the basic health information of the examinees through questionnaires, in order to improve the efficiency of collecting basic health information, the basic health information corresponding to each examinee can be collected through relevant networks (such as hospitals and physical examination centers at all levels).

[0042] From the basic health information, identify the potential risk factors in the target user's personal health status. Furthermore, the above risk factors can be recorded in the initial risk factor list (user_risk list) for subsequent risk calculation. And according to the preset disease factor association table and the identified risk factors (initial risk factors), calculate the risk of the target user suffering from the corresponding disease.

[0043] In a specific embodiment, based on the corresponding relationship diseasefactors between diseases and risk factors, the risk score DR(i) of the examinee for each disease i is calculated, and the calculation formula is as follows:

[0044]

[0045] Among them, diseasefactors(i) is used to represent all risk factors corresponding to disease i, and the denominator (|diseasefactors(i)|) is used to represent the total number of risk factors corresponding to disease i. user_risk is used to represent risk factors, and the numerator is used to represent the number of risk factors corresponding to disease i that the examinee meets. DR(i) is used to represent the risk of the examinee suffering from a certain disease. The risk value is between 0 and 1. The closer it is to 1, the greater the risk.

[0046] Based on the calculated risk score of each disease, the target user's potential risk for each disease is evaluated, and a candidate disease information list is generated based on each potential risk disease.

[0047] The higher the risk score, the greater the possibility that the examinee will have the disease. In this way, high-risk diseases with risk scores exceeding the preset score threshold are screened out, and physical examination items corresponding to the high-risk diseases are recommended to the examinee.

[0048] Step S20: generating a list of related questions corresponding to the initial risk factors and the candidate disease information list based on the target large language model, and performing interactive physical examination question inquiries to the target user based on the list of related questions to obtain dynamic health information corresponding to the target user;

[0049] In this embodiment, Figure 3 As shown, an interactive physical examination inquiry is performed on the target user through the target large language model (i.e., the physical examination large model), and the information is dynamically updated in combination with the health status of the examinee. During the conversation, the intelligent agent actively asks questions to guide the examinee to describe various health issues, thereby fully understanding the examinee's health risks and potential diseases.

[0050] Specifically, the examinee's health information can be collected through various channels to form a preliminary health record. Health information includes but is not limited to gender, age, occupation, past medical history, family medical history, disease risk factors, living habits, etc. In the subsequent interactive question-and-answer process, the examinee's health record will be dynamically updated based on the information provided by the examinee. For example, more feedback information about symptoms can be obtained and the health record can be updated in real time.

[0051] Among them, health information variables include:

[0052] H_base: basic health information set (such as age, living habits, family medical history, etc.)

[0053] H_dyn: A dynamic set of health information (symptoms or risk factors, etc.), and the dynamic health information will be continuously updated as the conversation progresses.

[0054] Exemplarily, the generating, based on the target large language model, the list of related questions corresponding to the initial risk factors and the candidate disease information list specifically includes:

[0055] Based on the target knowledge graph, obtaining domain context information corresponding to each disease information in the candidate disease information list and domain context information corresponding to each risk factor in the initial risk factors;

[0056] Inputting the candidate disease information and its corresponding domain context information, the initial risk factor and its corresponding domain context information into the target large language model to obtain a candidate symptom list and a candidate risk factor list;

[0057] Based on the target large language model, calculating a first confidence score corresponding to each symptom in the candidate symptom list and a second confidence score corresponding to each risk factor in the candidate risk factor list;

[0058] Based on a preset score threshold, the first confidence score, and the second confidence score, determining at least one relevant symptom in the candidate symptom list and determining at least one relevant risk factor in the candidate risk factor list;

[0059] Based on the respective relevant symptoms and the respective relevant risk factors, the list of relevant questions is generated.

[0060] In this embodiment, the doctor agent asks questions to the examinee based on the health record and physical examination model to discover the examinee's potential health risks. Specifically, the disease risk score DR can be obtained, and the risk of each disease i in the disease list is ranked from high to low in order of disease risk score, and the disease i with the highest score is selected for questioning.

[0061] Specifically, based on the examinee's health status and disease knowledge graph, high-risk symptoms and disease-related information are obtained; for disease i, the GraphRAG retrieval algorithm is executed to obtain relevant domain knowledge such as community reports, symptoms and risk factors. The doctor agent (i.e., the target large language model) combines the examinee's health information (H_base and H_dyn) and the extracted domain context knowledge, analyzes the examinee's potential symptoms and risk factors with the help of a predefined Prompt template, and generates a list of candidate high-risk symptoms and risk factors. Specifically, it includes:

[0062] 1. Information integration and filling: The basic health information H_base, dynamic health information H_dyn and domain context knowledge extracted from the GraphRAG method are filled into the Prompt template to provide comprehensive data support for analysis. That is, based on the big model, the Chain-of-Thought (COT) reasoning and In-Context Learning (ICL) methods are used to guide the model to gradually reason and analyze health information and generate a list of potential symptoms and risk factors. In the analysis process, the interaction between health information, domain knowledge and various factors is comprehensively considered.

[0063] 2. Candidate list generation: Output the structured candidate high-risk symptom list Cq_Sym and risk factor list Cq_Risk through the large model.

[0064] 3. Confidence score: The target large language model (i.e., the physical examination large model) generates a confidence score S_doctor(X_c) for each candidate, where X_c represents a symptom or risk factor. The confidence score is based on the consistency of health information and domain knowledge, the rigor of reasoning logic, and the analysis results of historical samples.

[0065] The doctor agent determines relevant questions based on conditional probability and confidence scores, thereby effectively reducing the information uncertainty of relevant questions and improving the effectiveness of relevant questions.

[0066] For each candidate symptom or risk item X_c∈Cq_Risk∪Cq_Sym, based on the known health information H_base and H_dyn, the conditional probability of getting the disease is calculated: P(i|H_base,H_dyn,X_c).

[0067] The above conditional probabilities represent the impact of the new information X_c on the likelihood of disease, which is evaluated in combination with basic health information and dynamic information.

[0068] By comparing the changes in conditional probabilities, the relative importance of candidate X_c is evaluated, and the calculation formula is as follows:

[0069] ΔP(X_c)=P(i|H_base,H_dyn,X_c)-P(i|H_base,H_dyn)

[0070] Among them, P(i|H_base,H_dyn) is the possibility that the examinee is ill without the new information X_c. ΔP(X_c) means that candidate X_c is more important for disease judgment.

[0071] Combine the doctor agent's score S_doctor(X_c) with the conditional probability to calculate the comprehensive score of candidate X_c:

[0072] S_final(X_c)=w1*ΔP(X_c)+w2*S_doctor(X_c)

[0073] Among them, w1 and w2 are weight parameters.

[0074] The doctor agent score S_doctor(X_c) is an evaluation result of the possible symptoms of the examinee based on the health information H_base and H_dyn.

[0075] ΔP(X_c), analyzes the relative importance of the newly added information X_c through heuristic deduction.

[0076] Select the highest scoring candidate:

[0077] c*=argmax(S_final(X_c)),X_c∈Cq_Risk∪Cq_Sym

[0078] Among them, c* is the high-priority candidate finally selected. The doctor agent generates specific related questions based on the final selected candidate c*. For example:

[0079] If c* is "shortness of breath", the related question generated is: "Have you had significant shortness of breath recently?"

[0080] If c* is “high-risk environment,” the related question generated is: “Do you have long-term work experience in high-risk environments?”

[0081] Furthermore, the performing interactive physical examination question inquiry to the target user based on the relevant question list to obtain the dynamic health information corresponding to the target user specifically includes:

[0082] Calculating the information completeness of the target user based on the basic health information and the dynamic health information of the target user;

[0083] When the number of dialogue turns corresponding to the target user exceeds a preset number threshold or the information completeness corresponding to the target user exceeds a preset ratio threshold, an inquiry stop instruction is generated to terminate the physical examination question inquiry corresponding to the target user.

[0084] In this embodiment, after the doctor agent asks questions to the examinee, the examinee answers based on his / her own health problems. After obtaining the examinee's answer A_k, the health information set H_dyn is updated.

[0085] The host agent is further responsible for managing the entire interactive question-and-answer process and generating an inquiry stop instruction to end the physical examination interactive inquiry process.

[0086] Specifically, the host agent determines whether the currently collected information (i.e., the target user's information completeness) is sufficient to diagnose the target user's corresponding disease list based on the currently collected health records and the examinee's answers, the predefined prompt, and the domain-related knowledge of disease i extracted through GraphRAG. If it is insufficient, and the current dialogue round is less than or equal to 10, the doctor agent can be prompted to continue asking questions about disease i; if the collected information is sufficient for diagnosis, or the dialogue round exceeds 10, an inquiry stop instruction is generated to terminate the physical examination question inquiry corresponding to the target user, and based on the existing health information sets H_dyn and H_base, the probability P(i) that the patient has disease i is output.

[0087] For disease i, if P(i)+DR(i) is greater than the preset probability threshold (such as 0.8), then disease i is determined to be a risk disease. The host agent can prompt the doctor agent to conduct multiple rounds of dialogue inquiries on the next disease and repeat the above process until all relevant questions corresponding to the disease are completed, ending the inquiry process.

[0088] Step S30, generating target symptom information and target risk factors based on the dynamic health information, and updating the candidate disease information list based on the target symptom information and target risk factors to obtain a target disease information list;

[0089] In this embodiment, the doctor agent combines the examinee's health information (H_base and H_dyn) and the extracted domain context knowledge, uses a predefined Prompt template to analyze the examinee's potential symptoms and risk factors, and generates a list of candidate high-risk symptoms and risk factors.

[0090] Specifically, the basic health information H_base, dynamic health information H_dyn of the examinee and the domain context knowledge extracted from the GraphRAG method are filled into the Prompt template together, so as to provide comprehensive data support for subsequent analysis.

[0091] The large model outputs a structured candidate high-risk symptom list Cq_Sym (i.e., target symptom information) and a risk factor list Cq_Risk (i.e., target risk factors):

[0092] Cq_Sym: high-risk symptoms, such as chest pain, dizziness, shortness of breath, etc.

[0093] Cq_Risk: Risk factors, such as staying up late, overeating, long-term smoking, etc.

[0094] According to the latest target symptom information and target risk factors, on the basis of the candidate disease information list, the probability of the examinee suffering from other diseases is further calculated to obtain the target disease information list.

[0095] Step S40: Generate target physical examination item information corresponding to the target user based on the physical examination item recommendation model and the target disease information list.

[0096] In this embodiment, the physical examination item recommendation model generates a personalized physical examination item list suitable for the target user as the target physical examination item information based on the above target disease information list and the predefined disease-physical examination item table. The reason for recommending each physical examination item can also be further clarified (i.e., the disease and risk factor corresponding to the physical examination item are further displayed).

[0097] Furthermore, after step S40, the method further includes:

[0098] Obtaining the physical examination result corresponding to the target user and the question feedback information corresponding to the list of related questions by the target user;

[0099] Based on the basic health information, the candidate disease information list, the related question list, the physical examination results and the question feedback information, the target large language model and / or the physical examination item recommendation model is optimized.

[0100] In this embodiment, the target large language model is further prioritized by the reflector agent, combined with the physical examination results corresponding to the target user and the question feedback information corresponding to the list of related questions by the target user. After the doctor agent asks a question, the reflector agent evaluates the rationality and efficiency of the question based on the predefined prompt and the patient's health information:

[0101] 1. Reasonableness assessment: used to indicate whether the assessment questions are consistent with the health record information of the current examinee, ensuring that the questions are relevant and reasonable. For example, if the examinee has no relevant medical history, is it inappropriate to ask questions about certain symptoms?

[0102] 2. Efficiency evaluation: used to indicate whether the evaluation question can effectively reduce information uncertainty and whether the question can help quickly determine disease risks. If the question is redundant or inefficient, the reflector agent can further feedback improvement suggestions.

[0103] 3. Question improvement: This means that if the evaluation determines that the doctor agent’s question is not reasonable or efficient enough, the reflector agent will make optimization suggestions to prompt the doctor agent to regenerate the question.

[0104] This embodiment provides a method for recommending physical examination items based on a large language model, the method generates the initial risk factors of the target user based on the basic health information of the target user, and generates a candidate disease information list corresponding to the initial risk factors based on a preset disease factor association table; based on the target large language model, generates a list of related questions corresponding to the initial risk factors and the candidate disease information list, and performs interactive physical examination question inquiries to the target user based on the list of related questions to obtain the dynamic health information corresponding to the target user; based on the dynamic health information, generates target symptom information and target risk factors, and based on the target symptom information and target risk factors, updates the candidate disease information list to obtain the target disease information list; based on the physical examination item recommendation model and the target disease information list, generates the target physical examination item information corresponding to the target user. In the above manner, the present application first screens out the initial risk factors through the basic health information of the target user, and determines the candidate disease information list of the target user according to the initial risk factors. Generate a list of related questions corresponding to the initial risk factors and the candidate disease information list through the target large language model, conduct interactive physical examination questions and answers with the target user according to the list of related questions, and mine the dynamic health information of the target user according to the question and answer results. Combined with dynamic health information, the target user's symptom information and health risk factors are further mined to further enrich the target user's disease information list. Finally, based on the target disease information list, personalized physical examination item recommendations are implemented for the target user. Through interactive physical examination questions, the efficiency of collecting dynamic health information of the target user is improved, thereby improving the efficiency of physical examination item recommendations. The target user's basic health information is combined with dynamic health information to comprehensively determine the target user's potential disease information. Based on the comprehensive target disease information list, physical examination item information that is more suitable for the target user is obtained, which improves the accuracy of physical examination item information recommendations.

[0105] Furthermore, before obtaining the domain context information corresponding to each disease information in the candidate disease information list and the domain context information corresponding to each risk factor in the initial risk factors based on the target knowledge graph, the method further includes:

[0106] Collect disease and health related information, and divide the disease and health related information into text units of various preset units;

[0107] Based on the target large language model, extracting entities corresponding to each symptom, risk factor and disease in each text unit, and extracting corresponding relationships between each symptom and each disease;

[0108] Based on each entity and each corresponding relationship, construct a homogeneous undirected weighted graph, wherein the nodes and edges in the homogeneous undirected weighted graph represent each entity and each corresponding relationship of each entity, respectively, and the weight of each edge represents a normalized count of detected relationship instances;

[0109] The subgraphs corresponding to different text units with the same entity in the homogeneous undirected weighted graph are merged, and the merged homogeneous undirected weighted graph is hierarchically clustered to obtain the communities

[0110] The report-style summaries corresponding to each community are extracted, and the target knowledge graph is generated based on each community and its corresponding report-style summaries.

[0111] In this embodiment, disease and health related information is first collected, such as data from medical literature, electronic health records, clinical trial reports, and public health databases, etc. Based on the above data, a medical knowledge base is constructed using GraphRAG indexing technology.

[0112] It is understandable that the collected data is preprocessed in advance, including text cleaning, removal of irrelevant information and terminology standardization.

[0113] Disease and health related information is divided into a series of text units (TextUnits), which are the basic units for subsequent analysis. For example, the unit size is 300 tokens.

[0114] Furthermore, in order to avoid truncation of sentences and paragraphs, a sequential block method can be adopted, introducing a sliding window to provide context for each block.

[0115] A large language model (LLM) is used to extract all entities, relations, and key statements from different text units. A key statement is a statement that describes the interaction between the subject entity and the object entity, such as (smoking may cause lung cancer, subject: smoking, object: lung cancer, action type: cause / lead). Combined with predefined prompts and examples, the extraction target is clearly defined, including entity categories (such as diseases, symptoms, treatment methods, etc.) and relations, and then a large language model is used to output the extraction results in a formalized structure.

[0116] A homogeneous undirected weighted graph is constructed using the extracted entities and relations. The nodes in the homogeneous undirected weighted graph are used to represent each entity, and the edges in the homogeneous undirected weighted graph are used to represent the corresponding relationship between each entity. The weight of the edge represents the standardized count of the detected relationship instance. The subgraphs corresponding to different text units are merged according to the same entity (i.e., the name and type are consistent).

[0117] In order to further improve the efficiency of recommending physical examination items and ensure the connectivity within the community, after the merger is completed, the Leiden algorithm is used to perform hierarchical clustering on the merged homogeneous undirected weighted graph to form different communities, making the community division more reasonable.

[0118] It is understandable that the merged homogeneous undirected weighted graph can also be hierarchically clustered using the traditional Louvain algorithm.

[0119] For each community, a summary of the community and its members is generated from the bottom up. A report-style summary is created for each community in the Leiden hierarchy. The above summary facilitates understanding of the global structure and semantics of the dataset.

[0120] The target knowledge graph is generated based on each community and its corresponding report-style summary, and the constructed knowledge graph is stored in the graph database Neo4j for subsequent query and analysis.

[0121] Furthermore, before step S40, the method further includes:

[0122] Initialize the weight parameters and bias parameters of each network layer in the initial model, and generate the survival probability parameters of each network layer based on preset rules;

[0123] Based on the survival probability parameters of each network layer, a target network layer and an optimized network layer are determined in each network layer, wherein the target network layer is a network layer to be trained, and the optimized network layer is a network layer for which training needs to be skipped;

[0124] Based on the preset training sample set, initial weight parameters and initial bias parameters, the optimization network layer is skipped, and the target network layer is trained until the preset convergence condition is reached, and the trained model is obtained as the physical examination item recommendation model.

[0125] In this embodiment, in order to prevent the physical examination item recommendation model from overfitting or underfitting during training, thereby causing the model training time to be too long, a random depth mechanism is provided based on the survival probability corresponding to each network layer in the model. Based on the random depth mechanism, some network layers can be randomly skipped during the training process, reducing the amount of forward and backward propagation calculations during model training, thereby reducing the computational complexity during model training; and only part of the network layer is used in each training, so that the training time is significantly shortened while ensuring the convergence of the model, preventing the model from overfitting, and enhancing the model's robustness to abnormal data and noise data. In addition, a label generation model is obtained through training based on the random depth mechanism. Under the premise of ensuring model performance, the model training efficiency and label generation accuracy can be further improved.

[0126] Specifically, the network parameters and the survival probability of each layer are initialized first. Then, in each forward propagation process, it is randomly decided whether to skip the layer according to the survival probability of each layer. For the network layers participating in the training, the parameters are updated by forward and backward propagation as usual. For the network layers not participating in the training, the parameters are not updated, thereby realizing multiple iterative training of the model to ensure that the model converges and achieves the optimal performance.

[0127] Exemplarily, the step S10 specifically includes:

[0128] Initializing the weight parameters and bias parameters of each network layer based on the He initialization method or the Xavier initialization method to obtain the initial weight parameters and the initial bias parameters;

[0129] Based on a linear decreasing strategy, the survival probability parameters of each network layer are generated, wherein the level of the network layer is inversely proportional to the corresponding survival probability parameter value.

[0130] In this embodiment, by introducing a random depth mechanism in the training process, efficient training of a deep neural network is achieved, specifically including:

[0131] 1. Initialize the network parameters, that is, initialize the weight parameters and bias parameters in each layer of the random deep neural network. Specifically, the model parameters of the random deep neural network can be initialized by using methods such as He initialization or Xavier initialization to ensure the stability and effectiveness of the training.

[0132] Second, initialize the survival probability, that is, assign an initial survival probability p to each layer of the network l .

[0133] Specifically, a linear decreasing strategy can be adopted, that is, the level of the network layer is inversely proportional to the corresponding survival probability parameter value. By adopting a linear decreasing survival probability design, the effective depth of the network during the training process is guaranteed, while avoiding overly complex calculations.

[0134] That is, the survival probability is higher near the input layer, while the survival probability is lower near the output layer. For example, the survival probability can be set as:

[0135]

[0136] Among them, l is the current network layer, L is the total number of network layers, and p L It is the survival probability of the output layer in the previous round. The initial value can be taken in the first round. The value is generally low, such as less than 50%.

[0137] Specifically, a preset probability threshold can be set. When the survival probability parameter is not less than the preset probability threshold, the network layer can be recorded as the target network layer to train the network layer. When the survival probability parameter is less than the preset probability threshold, the network layer can be recorded as the optimized network layer to skip the network layer.

[0138] The initial model is a random deep network model with initial weight parameters and initial bias parameters as the current model parameters. In each forward propagation process, the basic information of the contacts in the preset label sample set is input into each target network layer in the initial random deep network model, that is, the output results of each target network layer skip the next layer of the optimization network layer and directly pass it to the next layer of the target network layer. After completing the training of all target network layers in the random deep network model, it is determined whether the current random deep network model has reached the convergence condition. If not, the target network layer in the random deep network model is iteratively trained again, and the optimization network layer is still skipped until the random deep network model reaches the preset convergence condition. Stop model training to obtain the trained label generation model.

[0139] Furthermore, the determining of the target network layer and optimizing the network layer at each network layer based on the survival probability parameters of each network layer specifically includes:

[0140] Obtain any value within a preset value range as a random parameter of a network layer;

[0141] Compare the survival probability parameter of each network layer with its corresponding random parameter;

[0142] Based on the comparison results of each network layer, the network layer whose survival probability parameter is not less than the random parameter is determined as the target network layer, and the network layer whose survival probability parameter is less than the random parameter is determined as the optimized network layer.

[0143] First, the disease information samples in the physical examination item recommendation sample set are input into the first-level network layer.

[0144] Then, each network layer except the first-level network layer in the random deep network model is processed layer by layer, and each network layer of the initialized random deep network model is forward propagated:

[0145] For each network layer l, generate a random number r between [0,1] l , as random parameters of each level of network layer;

[0146] The random parameters r of each network layer are l The corresponding survival probability parameter p l To compare:

[0147] 1. If r l≤p l , then this layer participates in the calculation and performs normal forward propagation:

[0148] x l+1 =f l (x l ,W l ,b l )

[0149] Among them, x l is the input of the lth layer, W l and b l are the weight and bias of the lth layer, respectively, and f l is the activation function of this layer.

[0150] 2. If r l >p l , then skip this layer and pass the current input directly to the next layer:

[0151] x l+1 =x l

[0152] In the above manner, this embodiment generates corresponding random parameters for each level of network layer to ensure the randomness of skipping the layer, and then compares the random parameters of the layer with its corresponding survival probability parameters, and determines whether to skip the layer during training based on the comparison results. Thus, the randomness of the optimized network layer is ensured based on the random parameters, thereby avoiding the fixed probability parameter threshold leading to a fixed optimized network layer, and further improving the accuracy of the prediction results of the random deep network model.

[0153] After completing a round of forward propagation of the model, back propagation is performed to update the parameters of the network layers involved in the forward propagation. Specifically:

[0154] 1. Calculate the recommended physical examination item sample corresponding to the disease information sample (i.e., the target value y true ) and the loss function of the current recommended physical examination item label (network output y):

[0155] According to the network output y and target value y true Calculate the loss function L(y,y true ), using cross entropy loss or mean squared error loss.

[0156] 2. Back propagation layer by layer:

[0157] Starting from the output layer, the gradient is calculated forward layer by layer.

[0158] For each layer l, if the layer participates in the forward propagation (i.e., r l ≤p l ), the gradient is calculated normally and the parameters are updated:

[0159]

[0160] Among them, η is the preset learning rate.

[0161] If the layer is skipped (i.e. r l >p l ), no parameter update is performed and the gradient is directly passed to the previous layer:

[0162]

[0163] In the above manner, this embodiment will iterate multiple times on the entire training set based on the above forward propagation and back propagation steps until the loss function converges or reaches a preset number of training rounds. During each iteration, the survival probability p of each network layer in this round is l Keep it unchanged to ensure the stability and effect of model training.

[0164] See also Figure 4 , Figure 4 This is a schematic diagram of the functional modules of a physical examination item recommendation device based on a large language model provided in an embodiment of the present application.

[0165] like Figure 4 As shown, the physical examination item recommendation device 400 based on the large language model includes:

[0166] The information list generating module 410 is used to generate the initial risk factors of the target user based on the basic health information of the target user, and to generate a candidate disease information list corresponding to the initial risk factors based on a preset disease factor association table;

[0167] The dynamic information generation module 420 is used to generate a list of related questions corresponding to the initial risk factors and the candidate disease information list based on the target large language model, and perform interactive physical examination question inquiries to the target user based on the list of related questions to obtain dynamic health information corresponding to the target user;

[0168] An information list updating module 430 is used to generate target symptom information and target risk factors based on the dynamic health information, and to update the candidate disease information list based on the target symptom information and target risk factors to obtain a target disease information list;

[0169] The physical examination item recommendation module 440 is used to generate target physical examination item information corresponding to the target user based on the physical examination item recommendation model and the target disease information list.

[0170] Furthermore, the dynamic information generation module 420 specifically includes:

[0171] A context information acquisition unit, configured to acquire, based on a target knowledge graph, domain context information corresponding to each disease information in the candidate disease information list and domain context information corresponding to each risk factor in the initial risk factors;

[0172] A related list generating unit, used for inputting the candidate disease information and its corresponding domain context information, the initial risk factor and its corresponding domain context information into the target large language model to obtain a candidate symptom list and a candidate risk factor list;

[0173] a confidence score calculation unit, configured to calculate, based on the target large language model, a first confidence score corresponding to each symptom in the candidate symptom list and a second confidence score corresponding to each risk factor in the candidate risk factor list;

[0174] a risk factor determination unit, configured to determine at least one relevant symptom in the candidate symptom list and at least one relevant risk factor in the candidate risk factor list based on a preset score threshold, the first confidence score, and the second confidence score;

[0175] The question list generating unit is used to generate the relevant question list based on each relevant symptom and each relevant risk factor.

[0176] Furthermore, the physical examination item recommendation device 400 based on a large language model further includes a knowledge graph generation module, and the knowledge graph generation module is used to:

[0177] Collect disease and health related information, and divide the disease and health related information into text units of various preset units;

[0178] Based on the target large language model, extracting entities corresponding to each symptom, risk factor and disease in each text unit, and extracting corresponding relationships between each symptom and each disease;

[0179] Based on each entity and each corresponding relationship, construct a homogeneous undirected weighted graph, wherein the nodes and edges in the homogeneous undirected weighted graph represent each entity and each corresponding relationship of each entity, respectively, and the weight of each edge represents a normalized count of detected relationship instances;

[0180] Merging subgraphs corresponding to different text units with the same entity in the homogeneous undirected weighted graph, and performing hierarchical clustering on the merged homogeneous undirected weighted graph to obtain various communities;

[0181] The report-style summaries corresponding to each community are extracted, and the target knowledge graph is generated based on each community and its corresponding report-style summaries.

[0182] Furthermore, the physical examination item recommendation device 400 based on a large language model further includes:

[0183] The probability parameter generation module is used to initialize the weight parameters and bias parameters of each network layer in the initial model, and generate the survival probability parameters of each network layer based on preset rules;

[0184] A network layer classification module, used to determine a target network layer and an optimized network layer in each network layer based on the survival probability parameters of each network layer, wherein the target network layer is a network layer to be trained, and the optimized network layer is a network layer for which training needs to be skipped;

[0185] The label model training module is used to skip the optimization network layer based on a preset training sample set, initial weight parameters and initial bias parameters, and train the target network layer until a preset convergence condition is reached to obtain a trained model as the physical examination item recommendation model.

[0186] Furthermore, the network layer classification module specifically includes:

[0187] A random parameter generating unit, used to obtain any value within a preset value range as a random parameter of a network layer;

[0188] A network parameter comparison unit, used to compare the survival probability parameter of each network layer with its corresponding random parameter;

[0189] The network classification unit is used to determine, based on the comparison results of each network layer, the network layer whose survival probability parameter is not less than the random parameter as the target network layer, and determine the network layer whose survival probability parameter is less than the random parameter as the optimized network layer.

[0190] Furthermore, the dynamic information generation module 420 specifically includes:

[0191] a completeness calculation unit, configured to calculate the information completeness of the target user based on the basic health information and the dynamic health information of the target user;

[0192] The inquiry process control unit is used to generate an inquiry stop instruction to terminate the physical examination question inquiry corresponding to the target user when the dialogue round corresponding to the target user exceeds a preset number threshold or the information completeness corresponding to the target user exceeds a preset ratio threshold.

[0193] Furthermore, the physical examination item recommendation device 400 based on a large language model further includes a model optimization module, which is used to:

[0194] Obtaining the physical examination result corresponding to the target user and the question feedback information corresponding to the list of related questions by the target user;

[0195] Based on the basic health information, the candidate disease information list, the related question list, the physical examination results and the question feedback information, the target large language model and / or the physical examination item recommendation model is optimized.

[0196] It should be noted that those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described device and each module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0197] The above-mentioned device can be implemented in the form of a computer program. Figure 5 Runs on the computer device shown.

[0198] See also Figure 5 , Figure 5 1 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device may be a server.

[0199] See also Figure 5 The computer device includes a processor, a memory and a network interface connected through a system bus, wherein the memory may include a non-volatile storage medium and an internal memory.

[0200] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any method for recommending physical examination items based on a large language model.

[0201] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0202] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any physical examination item recommendation method based on a large language model.

[0203] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0204] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0205] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0206] Based on the basic health information of the target user, generate the initial risk factors of the target user, and based on the preset disease factor association table, generate a list of candidate disease information corresponding to the initial risk factors;

[0207] Based on the target large language model, a list of related questions corresponding to the initial risk factors and the candidate disease information list is generated, and based on the list of related questions, an interactive physical examination question inquiry is performed to the target user to obtain dynamic health information corresponding to the target user;

[0208] generating target symptom information and target risk factors based on the dynamic health information, and updating the candidate disease information list based on the target symptom information and target risk factors to obtain a target disease information list;

[0209] Based on the physical examination item recommendation model and the target disease information list, target physical examination item information corresponding to the target user is generated.

[0210] In one embodiment, when the processor is used to process the target large language model to generate the initial risk factors and the list of related questions corresponding to the candidate disease information list, it is also used to implement:

[0211] Based on the target knowledge graph, obtaining domain context information corresponding to each disease information in the candidate disease information list and domain context information corresponding to each risk factor in the initial risk factors;

[0212] Inputting the candidate disease information and its corresponding domain context information, the initial risk factor and its corresponding domain context information into the target large language model to obtain a candidate symptom list and a candidate risk factor list;

[0213] Based on the target large language model, calculating a first confidence score corresponding to each symptom in the candidate symptom list and a second confidence score corresponding to each risk factor in the candidate risk factor list;

[0214] Based on a preset score threshold, the first confidence score, and the second confidence score, determining at least one relevant symptom in the candidate symptom list and determining at least one relevant risk factor in the candidate risk factor list;

[0215] Based on the respective relevant symptoms and the respective relevant risk factors, the list of relevant questions is generated.

[0216] In one embodiment, the processor is used to process the target knowledge graph, and before obtaining the domain context information corresponding to each disease information in the candidate disease information list and the domain context information corresponding to each risk factor in the initial risk factor, it is also used to implement:

[0217] Collect disease and health related information, and divide the disease and health related information into text units of various preset units;

[0218] Based on the target large language model, extracting entities corresponding to each symptom, risk factor and disease in each text unit, and extracting corresponding relationships between each symptom and each disease;

[0219] Based on each entity and each corresponding relationship, construct a homogeneous undirected weighted graph, wherein the nodes and edges in the homogeneous undirected weighted graph represent each entity and each corresponding relationship of each entity, respectively, and the weight of each edge represents a normalized count of detected relationship instances;

[0220] Merging subgraphs corresponding to different text units with the same entity in the homogeneous undirected weighted graph, and performing hierarchical clustering on the merged homogeneous undirected weighted graph to obtain various communities;

[0221] The report-style summaries corresponding to each community are extracted, and the target knowledge graph is generated based on each community and its corresponding report-style summaries.

[0222] In one embodiment, the processor is used to process the physical examination item recommendation model and the target disease information list, and before generating the target physical examination item information corresponding to the target user, it is also used to implement:

[0223] Initialize the weight parameters and bias parameters of each network layer in the initial model, and generate the survival probability parameters of each network layer based on preset rules;

[0224] Based on the survival probability parameters of each network layer, a target network layer and an optimized network layer are determined in each network layer, wherein the target network layer is a network layer to be trained, and the optimized network layer is a network layer for which training needs to be skipped;

[0225] Based on the preset training sample set, initial weight parameters and initial bias parameters, the optimization network layer is skipped, and the target network layer is trained until the preset convergence condition is reached, and the trained model is obtained as the physical examination item recommendation model.

[0226] In one embodiment, the processor is used to process the survival probability parameters based on each network layer, and when each network layer determines the target network layer and optimizes the network layer, it is also used to implement:

[0227] Obtain any value within a preset value range as a random parameter of a network layer;

[0228] Compare the survival probability parameter of each network layer with its corresponding random parameter;

[0229] Based on the comparison results of each network layer, the network layer whose survival probability parameter is not less than the random parameter is determined as the target network layer, and the network layer whose survival probability parameter is less than the random parameter is determined as the optimized network layer.

[0230] In one embodiment, when the processor is used to process the interactive physical examination question query to the target user based on the relevant question list to obtain the dynamic health information corresponding to the target user, it is also used to implement:

[0231] Calculating the information completeness of the target user based on the basic health information and the dynamic health information of the target user;

[0232] When the number of dialogue turns corresponding to the target user exceeds a preset number threshold or the information completeness corresponding to the target user exceeds a preset ratio threshold, an inquiry stop instruction is generated to terminate the physical examination question inquiry corresponding to the target user.

[0233] In one embodiment, the processor is used to process the physical examination item recommendation model and the target disease information list, and after generating the target physical examination item information corresponding to the target user, it is also used to implement:

[0234] Obtaining the physical examination result corresponding to the target user and the question feedback information corresponding to the list of related questions by the target user;

[0235] Based on the basic health information, the candidate disease information list, the related question list, the physical examination results and the question feedback information, the target large language model and / or the physical examination item recommendation model is optimized.

[0236] A computer-readable storage medium is also provided in an embodiment of the present application, wherein the computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and the processor executes the program instructions to implement any one of the physical examination item recommendation methods based on a large language model provided in the embodiments of the present application.

[0237] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (Smart Medi a Card, SMC), a secure digital (Secure Digital, SD) card, a flash memory card (Flash Card), etc., equipped on the computer device.

[0238] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for recommending physical examination items based on a large language model, characterized in that: The method for recommending physical examination items comprises the following steps: Based on the basic health information of the target user, generate the initial risk factors of the target user, and based on the preset disease factor association table, generate a list of candidate disease information corresponding to the initial risk factors; Based on the target large language model, a list of related questions corresponding to the initial risk factors and the candidate disease information list is generated, and based on the list of related questions, an interactive physical examination question inquiry is performed to the target user to obtain dynamic health information corresponding to the target user; generating target symptom information and target risk factors based on the dynamic health information, and updating the candidate disease information list based on the target symptom information and target risk factors to obtain a target disease information list; Based on the physical examination item recommendation model and the target disease information list, target physical examination item information corresponding to the target user is generated.

2. The method for recommending physical examination items based on a large language model according to claim 1, characterized in that: The generating of the initial risk factors and the list of related questions corresponding to the candidate disease information list based on the target large language model specifically includes: Based on the target knowledge graph, obtaining domain context information corresponding to each disease information in the candidate disease information list and domain context information corresponding to each risk factor in the initial risk factors; Inputting the candidate disease information and its corresponding domain context information, the initial risk factor and its corresponding domain context information into the target large language model to obtain a candidate symptom list and a candidate risk factor list; Based on the target large language model, calculating a first confidence score corresponding to each symptom in the candidate symptom list and a second confidence score corresponding to each risk factor in the candidate risk factor list; Based on a preset score threshold, the first confidence score, and the second confidence score, determining at least one relevant symptom in the candidate symptom list and determining at least one relevant risk factor in the candidate risk factor list; Based on the respective relevant symptoms and the respective relevant risk factors, the list of relevant questions is generated.

3. The method for recommending physical examination items based on a large language model as claimed in claim 2, characterized in that: Before acquiring the domain context information corresponding to each disease information in the candidate disease information list and the domain context information corresponding to each risk factor in the initial risk factors based on the target knowledge graph, the method further includes: Collect disease and health related information, and divide the disease and health related information into text units of various preset units; Based on the target large language model, extracting entities corresponding to each symptom, risk factor and disease in each text unit, and extracting corresponding relationships between each symptom and each disease; Based on each entity and each corresponding relationship, construct a homogeneous undirected weighted graph, wherein the nodes and edges in the homogeneous undirected weighted graph represent each entity and each corresponding relationship of each entity, respectively, and the weight of each edge represents a normalized count of detected relationship instances; The subgraphs corresponding to different text units with the same entity in the homogeneous undirected weighted graph are merged, and the merged homogeneous undirected weighted graph is hierarchically clustered to obtain the communities The report-style summaries corresponding to each community are extracted, and the target knowledge graph is generated based on each community and its corresponding report-style summaries.

4. The method for recommending physical examination items based on a large language model according to claim 1, characterized in that: Before generating the target physical examination item information corresponding to the target user based on the physical examination item recommendation model and the target disease information list, the method further includes: Initialize the weight parameters and bias parameters of each network layer in the initial model, and generate the survival probability parameters of each network layer based on preset rules; Based on the survival probability parameters of each network layer, a target network layer and an optimized network layer are determined in each network layer, wherein the target network layer is a network layer to be trained, and the optimized network layer is a network layer for which training needs to be skipped; Based on the preset training sample set, initial weight parameters and initial bias parameters, the optimization network layer is skipped, and the target network layer is trained until the preset convergence condition is reached, and the trained model is obtained as the physical examination item recommendation model.

5. The method for recommending physical examination items based on a large language model as claimed in claim 4, characterized in that: The determining of the target network layer and optimizing the network layer at each network layer based on the survival probability parameters of each network layer specifically includes: Obtain any value within a preset value range as a random parameter of a network layer; Compare the survival probability parameter of each network layer with its corresponding random parameter; Based on the comparison results of each network layer, the network layer whose survival probability parameter is not less than the random parameter is determined as the target network layer, and the network layer whose survival probability parameter is less than the random parameter is determined as the optimized network layer.

6. The method for recommending physical examination items based on a large language model according to claim 1, characterized in that: The performing of interactive physical examination question inquiry to the target user based on the relevant question list to obtain the dynamic health information corresponding to the target user specifically includes: Calculating the information completeness of the target user based on the basic health information and the dynamic health information of the target user; When the number of dialogue turns corresponding to the target user exceeds a preset number threshold or the information completeness corresponding to the target user exceeds a preset ratio threshold, an inquiry stop instruction is generated to terminate the physical examination question inquiry corresponding to the target user.

7. The method for recommending physical examination items based on a large language model according to any one of claims 1 to 6, characterized in that: After generating the target physical examination item information corresponding to the target user based on the physical examination item recommendation model and the target disease information list, the method further includes: Obtaining the physical examination result corresponding to the target user and the question feedback information corresponding to the list of related questions by the target user; Based on the basic health information, the candidate disease information list, the related question list, the physical examination results and the question feedback information, the target large language model and / or the physical examination item recommendation model is optimized.

8. A physical examination item recommendation device based on a large language model, characterized in that: The physical examination item recommendation device based on the large language model includes: An information list generating module, used to generate initial risk factors of the target user based on the basic health information of the target user, and to generate a candidate disease information list corresponding to the initial risk factors based on a preset disease factor association table; A dynamic information generation module, for generating a list of related questions corresponding to the initial risk factors and the candidate disease information list based on a target large language model, and performing interactive physical examination question inquiries to the target user based on the list of related questions to obtain dynamic health information corresponding to the target user; An information list updating module, used to generate target symptom information and target risk factors based on the dynamic health information, and update the candidate disease information list based on the target symptom information and target risk factors to obtain a target disease information list; The physical examination item recommendation module is used to generate target physical examination item information corresponding to the target user based on the physical examination item recommendation model and the target disease information list.

9. A computer device, characterized in that: The computer device includes a processor, a memory, and a physical examination item recommendation program based on a large language model stored in the memory and executable by the processor, wherein when the physical examination item recommendation program based on a large language model is executed by the processor, the steps of the physical examination item recommendation method based on a large language model as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a physical examination item recommendation program based on a large language model, wherein when the physical examination item recommendation program based on a large language model is executed by a processor, the steps of the physical examination item recommendation method based on a large language model as described in any one of claims 1 to 7 are implemented.

Images