Intelligent triage and guide analysis method, device and system based on large language model

Abstract

The invention relates to the technical field of digital data processing, in particular to an intelligent triage and guide analysis method, device and system based on a large language model.The method comprises the steps that symptom information of a patient is collected in a text input or voice input mode, and the symptom information and patient identity information are associated; cleaning the collected symptom information and extracting a symptom keyword group; matching the extracted symptom keywords with data in a knowledge graph to determine a to-be-selected disease group; generating a corresponding recommended department table according to the to-be-selected disease group, and arranging the departments according to the priority; and correcting the arrangement sequence of the departments based on the historical treatment records of the patients. Through the steps, the symptom information of the patient can be efficiently collected and processed, and accurate and personalized medical suggestions can be provided. According to the method, the efficiency and the quality of medical services are greatly improved, and meanwhile, the medical experience of the patient is also improved.

Description

Technical Field

[0001] The present invention relates to the technical field of digital data processing, and particularly to an intelligent triage and guiding analysis method, device, and system based on a large language model. Background Art

[0002] Triage and guiding is an important link in medical and health services, aiming to improve the efficiency of medical services and the patient's medical experience. By asking the patient's basic information such as symptoms and medical history, triage and guiding can help determine what type of medical service the patient needs and guide them to the appropriate department or doctor for treatment. This process not only helps to reasonably allocate medical resources, reduce the patient's waiting time in line, but also ensures that emergency and critically ill patients receive timely treatment.

[0003] The existing triage and guiding mainly adopts the method of filling in online by oneself or asking the corresponding medical staff manually offline. Among them, filling in online has relatively high requirements for individuals and is not very practical for some elderly people and children with weak behavioral abilities. Therefore, the method of manual inquiry needs to be adopted more, which increases the work burden. Summary of the Invention

[0004] The purpose of the present invention is to provide an intelligent triage and guiding analysis method, device, and system based on a large language model, aiming to automatically identify diseases and match departments in combination with the patient's symptom description, thereby improving work efficiency.

[0005] To achieve the above object, in the first aspect, the present invention provides an intelligent triage and guiding analysis method based on a large language model, including collecting the patient's symptom information through text input or voice input, and associating the symptom information with the patient identity information;

[0006] Cleaning the collected symptom information and extracting symptom keyword groups;

[0007] Matching the extracted symptom keywords with the data in the knowledge graph to determine a group of candidate diseases;

[0008] Generating a corresponding recommended department list according to the group of candidate diseases and arranging these departments according to priority;

[0009] Revising the arrangement order of the departments based on the patient's historical medical records.

[0010] Among them, the specific steps of collecting the patient's symptom information through text input or voice input and associating the symptom information with the patient identity information include:

[0011] Obtaining the first text information input by the patient through text;

[0012] Obtain the voice information input by the patient and convert the voice information into second text information;

[0013] Merge the first text information and the second text information into symptom information and associate it with the patient identity information.

[0014] Among them, the specific steps of cleaning the collected symptom information and extracting symptom keyword groups include:

[0015] Remove the irrelevant information in the collected symptom information to obtain screening information;

[0016] Split the screening information into phrases to be matched;

[0017] Use a pre-defined medical term dictionary to directly extract keywords from the phrases to be matched through string matching to obtain symptom keyword groups.

[0018] Among them, the specific steps of matching the extracted keywords with the data in the knowledge graph to determine the group of candidate diseases include:

[0019] Generate a knowledge graph based on the symptom descriptions, disease names, and relevant department information of diseases;

[0020] Group the symptom keyword groups provided by the patient to obtain multiple keyword combinations to be matched, and each keyword combination has at least two keywords;

[0021] Match each keyword group to be matched in the knowledge graph to obtain the disease names corresponding to each keyword combination;

[0022] Count the frequencies of the occurrences of each disease name and sort the diseases in descending order of the frequencies to obtain the group of candidate diseases.

[0023] Among them, the specific steps of generating a knowledge graph based on the symptom descriptions, disease names, and relevant department information of diseases include:

[0024] Clean the collected symptom descriptions, disease names, and relevant department information to remove duplicate information to obtain initial information;

[0025] Use named entity recognition technology to extract core entities in the initial information, and the core entities include disease names, symptom descriptions, and department information;

[0026] Use a machine learning model to generate the relationships between the core entities;

[0027] Import the core entities and their corresponding relationship data into a graph database, establish the links between the entities and their relationships, and generate a knowledge graph.

[0028] Among them, the specific steps of arranging and grouping the symptom keyword groups provided by the patient to obtain multiple keyword combinations to be matched, where each keyword combination has at least two keywords include:

[0029] Convert all the collected symptom keywords into a unified format;

[0030] Group the symptom keyword groups based on the body systems to which the symptoms belong as the main grouping basis to obtain multiple keyword combinations to be matched.

[0031] Among them, the specific steps of matching each keyword group to be matched in the knowledge graph to obtain the disease name corresponding to each keyword combination include:

[0032] Generate an extended phrase for each keyword group to be matched;

[0033] Match the corresponding disease nodes in the knowledge graph based on the extended phrase;

[0034] Obtain the corresponding disease name based on the disease node.

[0035] Among them, the specific steps of generating a corresponding recommended department list according to the disease group to be selected and arranging these departments according to the priority include:

[0036] Obtain the corresponding department list based on the disease group to be selected;

[0037] Merge the same departments;

[0038] Arrange the departments based on the frequency of disease matching in the merged departments.

[0039] In a second aspect, the present invention also provides an intelligent triage and guiding analysis device based on a large language model, including a symptom collection module, a keyword group acquisition module, a disease group to be selected matching module, a department arrangement module, and a department correction module;

[0040] The symptom collection module is used to collect the symptom information of the patient by means of text input or voice input, and associate the symptom information with the patient identity information;

[0041] The keyword group acquisition module is used to clean the collected symptom information and extract the symptom keyword group;

[0042] The disease group to be selected matching module is used to match the extracted symptom keywords with the data in the knowledge graph to determine the disease group to be selected;

[0043] The department arrangement module is used to generate a corresponding recommended department list according to the disease group to be selected and arrange these departments according to the priority;

[0044] The department correction module is used to correct the arrangement order of departments based on the patient's historical medical records.

[0045] In a third aspect, the present invention also provides an intelligent triage and guidance analysis system based on a large language model, including the intelligent triage and guidance analysis device based on a large language model described above.

[0046] The intelligent triage and guidance analysis method, device and system based on a large language model of the present invention allow patients to describe their symptoms by means of text input or voice input. Whether through a mobile application, website or other digital platforms, users can conveniently submit their health conditions. While collecting symptom information, the system requires users to provide basic identity information (such as name, contact information, etc.) in order to associate these symptoms with specific patients. Perform preliminary data cleaning on all symptom descriptions collected from patients. Remove unnecessary punctuation marks, special characters and formatting errors to ensure the consistency and accuracy of the data. Use advanced natural language processing techniques (NLP), especially named entity recognition (NER) technology, to extract key symptom phrases from the cleaned symptom information. In addition, synonym replacement and context understanding are also performed to ensure the comprehensiveness and accuracy of the extracted keywords. Convert the extracted keywords into a unified format for subsequent processing and analysis. Utilize a pre-constructed medical knowledge graph, which contains a wide range of diseases, symptoms and related information. This knowledge graph serves as a basic database to support the disease matching process. Adopt an efficient string matching algorithm or machine learning model to compare the extracted symptom keywords with the data in the knowledge graph to identify relevant diseases. This process takes into account the combination patterns of symptoms and their association strengths with different diseases. Generate a group of candidate diseases containing multiple potential diseases according to the matching degree, and calculate the corresponding matching degree scores for each disease for subsequent sorting and recommendation. Based on the group of candidate diseases, query the corresponding diagnosis and treatment department information for each disease to generate a preliminary department list. Check and merge duplicate department entries to avoid the same department being listed multiple times due to multiple diseases. Arrange the departments in descending order according to the matching frequencies of the departments with the diseases in the group of candidate diseases and other factors (such as urgency, specialty expertise, etc.) to form a final recommended department list. Access and analyze the patient's electronic health record (EHR) to understand their past medical history, treatment effects and any special preferences. Based on the findings in the historical records, appropriately adjust the originally set department sorting weights. For example, increase the weights of the departments that have successfully treated the patient's current or similar symptoms.

[0047] Through the above steps, the intelligent triage and diagnosis analysis method based on the large language model can not only efficiently collect and process patients' symptom information, but also provide accurate and personalized medical advice. This method greatly improves the efficiency and quality of medical services, and at the same time improves patients' medical experience, enabling them to obtain the required medical services more quickly. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0049] Figure 1 is a flowchart of the intelligent triage and diagnosis analysis method based on the large language model of the present invention.

[0050] Figure 2 is a flowchart of collecting patients' symptom information by text input or voice input and associating the symptom information with the patients' identity information in the present invention.

[0051] Figure 3 is a flowchart of cleaning the collected symptom information and extracting symptom keyword groups in the present invention.

[0052] Figure 4 is a flowchart of matching the extracted keywords with the data in the knowledge graph to determine the candidate disease groups in the present invention.

[0053] Figure 5 is a flowchart of generating a knowledge graph based on the symptom descriptions, disease names, and relevant department information of diseases in the present invention.

[0054] Figure 6 is a flowchart of arranging and grouping the symptom keyword groups provided by the patient to obtain multiple keyword combinations to be matched, and each keyword combination has at least two keywords in the present invention.

[0055] Figure 7 is a flowchart of matching each keyword group to be matched in the knowledge graph to obtain the disease names corresponding to each keyword combination in the present invention.

[0056] Figure 8 is a flowchart of generating a corresponding recommended department table based on the candidate disease groups and arranging these departments according to the priority in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0057] Embodiments of the present invention will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where like or similar reference numerals denote like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are intended to explain the present invention and should not be construed as limiting the present invention.

[0058] First Embodiment

[0059] Please refer to Figures 1 to 8 , the present invention provides an intelligent triage and guidance analysis method based on a large language model, including:

[0060] S101 Collect the symptom information of the patient by means of text input or voice input, and associate the symptom information with the patient identity information;

[0061] The specific steps include:

[0062] S201 Obtain the first text information input by the patient through text input;

[0063] Provide the patient with an intuitive and easy-to-operate user interface that allows them to directly enter their symptom descriptions on the interface. This can be achieved through a mobile application, a web form, or a self-service terminal. The patient, according to their own situation, enters the relevant symptom information in the provided text box. To help the patient express their condition more accurately, the system can provide some guiding questions or a list of common symptoms for selection. Once the patient has completed the input, the system will conduct a preliminary check on the text information, such as checking whether it is empty and whether it contains obvious formatting errors, etc.

[0064] S202 Obtain the voice information input by the patient and convert the voice information into second text information;

[0065] For patients who prefer to use voice input, the system should provide a voice input function. This requires integrating efficient speech recognition technology to ensure that spoken language can be accurately converted into text. Then, natural language processing (NLP) technology is used to further optimize the converted text, such as correcting recognition errors and adjusting word order, etc., to ensure that the finally generated text information is as close as possible to the patient's original intention.

[0066] S203 Merge the first text information and the second text information into symptom information and associate it with the patient identity information.

[0067] Combine the first text information obtained from text input and the second text information obtained from voice input. If there are duplicate or conflicting contents between the two, manual or automated review and correction are required. The symptom information of each patient needs to be associated with their personal identity information (such as name, ID number, contact information, etc.). This step usually needs to follow strict privacy protection regulations to ensure the security and confidentiality of patient information. The sorted symptom information and personal identity information are securely stored in a central database.

[0068] S102 Clean the collected symptom information and extract symptom keyword groups;

[0069] The specific steps include:

[0070] S301 Remove the irrelevant information from the collected symptom information to obtain screening information;

[0071] Perform preliminary data cleaning on all text or text information converted from voice collected from patients. This includes removing unnecessary punctuation marks, special characters, and any formatting issues that interfere with subsequent analysis. Use a predefined stop word list (such as common words like "of", "is", etc. that are not directly helpful for symptom description) to filter out these non-critical words to reduce noisy data. Use a medical term dictionary or thesaurus to identify and replace synonyms or near-synonyms of symptom descriptions in the text to ensure uniformity and accuracy. For example, "headache" and "cephalalgia" should be regarded as the same symptom. With the help of natural language processing (NLP) technology, especially context analysis tools, identify sentences or phrases that seemingly irrelevant but actually imply important health information, and decide whether to retain them according to the actual situation.

[0072] S302 Split the screening information into phrases to be matched;

[0073] Divide the symptom description text after cleaning into smaller units according to logical meaning, such as sentences or phrases, for subsequent processing. This step helps to focus on specific symptom descriptions rather than the entire narrative background.

[0074] Further divide each sentence or phrase into independent phrases. Here, a phrase refers to a group of words that can express a complete meaning, such as "continuous fever", "worsening cough at night", etc. Standardize the divided phrases to ensure that they meet the predefined format requirements, such as converting all to lowercase for unified management and comparison.

[0075] S303 Use a predefined medical term dictionary to directly extract keywords from the phrases to be matched through string matching to obtain symptom keyword groups.

[0076] Prepare a medical term dictionary that comprehensively covers various diseases and their related symptoms. The dictionary should include standard medical terms, their common variants, and abbreviations.

[0077] Select a suitable string matching algorithm for automated keyword extraction. Commonly used algorithms include naive string matching, KMP algorithm, regular expression matching, etc. The specific selection depends on actual requirements and performance considerations.

[0078] Use the selected algorithm to compare each phrase to be matched with the entries in the medical term dictionary one by one, and find the keywords that are exactly or partially matched. For partially matched cases, it can be determined whether to include them in the final keyword group according to the similarity threshold.

[0079] Perform final optimization on the extracted keywords, including but not limited to merging highly relevant keywords (such as "high fever" and "fever"), removing redundant information, and adjusting the keyword order to make it more logical.

[0080] Through the above steps, accurate symptom keyword groups can be effectively extracted from the patient's symptom description, providing strong support for further medical diagnosis. This method not only improves the efficiency of data processing but also enhances the accuracy of understanding the patient's condition.

[0081] S103 Match the extracted symptom keywords with the data in the knowledge graph to determine the candidate disease group;

[0082] The specific steps include:

[0083] S401 Generate a knowledge graph based on the symptom description, disease name, and related department information of the disease;

[0084] The specific steps include:

[0085] S501 Clean the collected symptom description, disease name, and related department information to remove duplicate information and obtain the initial information;

[0086] Collect information on symptom description, disease name, and related departments from multiple sources (such as electronic health records, medical literature, clinical guidelines, etc.). Clean the collected raw data, which includes removing duplicate records, correcting spelling mistakes, unifying formats (for example, converting all text to lowercase or standard abbreviations), deleting irrelevant characters or fields, etc. Ensuring the consistency and accuracy of the data is the key at this stage. Identify and delete duplicate data entries to avoid biases in subsequent analysis.

[0087] S502 Use named entity recognition technology to extract core entities from the initial information, where the core entities include disease names, symptom descriptions, and department information;

[0088] Select appropriate natural language processing (NLP) tools or platforms that should have powerful named entity recognition (NER) capabilities to accurately extract key entities from unstructured text. Then clarify the core entity categories to be recognized, such as disease names, symptom descriptions, and department information. For each category, further subdivision is also required, such as classifying symptoms according to the system (cardiovascular, digestive system, etc.). Apply NER technology to automatically annotate the core entities in the document. Adjust the model parameters or train a custom model according to the specific situation to improve the recognition accuracy.

[0089] S503 Use a machine learning model to generate relationships between core entities;

[0090] Determine the types of relationships that exist between different types of entities, such as "a certain disease presents certain symptoms", "a certain disease is treated by a certain department", etc. Then select a suitable relationship extraction model (such as a deep learning-based relationship extraction model) and train it using the labeled dataset. If there is enough labeled data, it can be directly trained; otherwise, some data needs to be manually labeled first as the training set. Finally, use the trained model to predict relationships for the unlabeled data and find potential connections between entities.

[0091] S504 Import the core entities and their corresponding relationship data into a graph database to establish links between entities and their relationships and generate a knowledge graph.

[0092] Select a suitable graph database according to the project requirements (such as Neo4j, TigerGraph, etc.). Such databases are especially suitable for storing and querying data with complex associations. Write scripts or use graphical interface tools to import the core entities and the relationship data between them into the selected graph database. During this process, pay attention to optimizing the data structure to improve query efficiency. Create nodes in the graph database to represent each entity, and use edges to represent the relationships between entities. Ensure that each edge accurately reflects the specific relationship type between entities. After completing the preliminary construction of the knowledge graph, conduct verification checks to ensure that there are no incorrect connections or missing important information. Continuously adjust and optimize according to the actual usage to improve the quality and practicality of the knowledge graph.

[0093] S402 Group the symptom keyword groups provided by the patient to obtain multiple keyword combinations to be matched, and each keyword combination has at least two keywords;

[0094] The specific steps include:

[0095] S601 Convert all collected symptom keywords into a unified format;

[0096] First, standardize all symptom keywords collected from patients. This includes converting all text to lowercase, removing punctuation, special characters, and extra spaces to ensure consistency in subsequent processing. Use pre-defined medical terminology dictionaries or thesaurus to identify and replace synonyms or near-synonyms in symptom descriptions. For example, "headache" and "headache" should be considered the same symptom. This can reduce data dispersion problems caused by differences in expression. For some common symptom expressions, normalize them into standard medical terms. For example, convert "heart beats fast" into "tachycardia" to improve the professionalism and accuracy of the data.

[0097] S602 groups the symptom keyword groups according to the body system to which the symptoms belong as the main grouping basis to obtain a plurality of keyword combinations to be matched.

[0098] Decide on a categorization framework based on the different systems of the human body, such as cardiovascular, respiratory, digestive, nervous, etc. This helps organize information based on the primary body part that the symptoms affect.

[0099] According to the above classification framework, the standardized symptom keywords were initially assigned to the corresponding body system categories. For example, "chest pain" and "palpitations" were classified as cardiovascular system; "cough" and "shortness of breath" were classified as respiratory system.

[0100] Within each body system category, further check whether there are cases where only a single symptom keyword is included. If there is only one symptom keyword in a grouping, try to meet the requirement of containing at least two keywords by using association rule learning or other machine learning techniques to find other symptoms that often co-occur with the symptom and add them to the current grouping. After completing the preliminary grouping, conduct a comprehensive review of the results to ensure that each keyword combination not only meets the body system classification criteria, but also actually contains at least two symptom keywords. At the same time, the practical significance and clinical relevance of these combinations need to be evaluated, and appropriate adjustments should be made if necessary.

[0101] S403 matches each keyword group to be matched in the knowledge graph to obtain the disease name corresponding to each keyword combination;

[0102] The specific steps include:

[0103] S701 generates an extended phrase for each keyword group to be matched;

[0104] Using a pre - defined medical term dictionary or thesaurus, find all synonyms, near - synonyms, and variant forms for each keyword. For example, "fever" can be expanded to "have a fever", "high fever", etc. Based on medical knowledge and data mining techniques, identify other symptoms closely related to the original keyword and add them to the expanded phrase. For instance, if the keyword is "cough", then "expectoration", "shortness of breath", etc. are also related symptoms. Consider the context information in the symptom description. For certain specific expressions (such as "cough that gets worse at night"), try to break it down and supplement more specific descriptions (such as "night", "gets worse") to form a more comprehensive expanded phrase.

[0105] S702 Match the corresponding disease nodes in the knowledge graph based on the expanded phrases;

[0106] According to the expanded phrases, design corresponding query statements or algorithms to search the knowledge graph. This usually involves complex graph query languages (such as Cypher for Neo4j databases) or API calls to accurately locate the disease nodes related to these symptoms. Since the diagnosis of some diseases requires multiple symptoms to be considered, in addition to direct matching, multi - step queries need to be performed to explore the connection paths between different symptoms through intermediate nodes (such as other symptoms, risk factors, etc.). Calculate a matching score for each match. This score can be calculated based on multiple factors, including but not limited to the frequency of symptom occurrence, the strength of the association between symptoms, and the degree of literature support. This can help distinguish potential matching results and preferentially display the optimal diseases.

[0107] S703 Obtain the corresponding disease names based on the disease nodes.

[0108] Once the disease nodes that match the expanded phrases are determined, relevant disease names and other useful information can be extracted from them, such as the typical symptoms of the disease, common treatment methods, preventive measures, etc. Check all the extracted disease names, remove duplicates, and merge those cases that, although have different names, actually refer to the same disease (such as different naming habits or translation differences). Organize the final result list, sort it according to the matching score or other importance indicators, and display it to the user. Ensure that the results are easy to understand and provide sufficient background information to help doctors make further judgments or suggestions.

[0109] S404 Count the frequency of occurrence of each disease name and sort the diseases in descending order of frequency to obtain a group of candidate diseases.

[0110] After completing step S703, we have obtained a set of disease names that match the patient's symptom keyword combination. This data contains detailed information about each potential disease. Prepare a data structure (such as a hash table or dictionary) to store each disease name and its corresponding frequency of occurrence. This structure will serve as the basis for subsequent statistics.

[0111] Traverse all the disease names matched from the knowledge graph. For each disease name, check whether it already exists in our statistical structure. If it does, increase the frequency count corresponding to the disease name. If it does not exist, add it to the statistical structure and set the initial frequency to 1.

[0112] Depending on the actual situation, it is necessary to give higher weights to certain types of matches. For example, if a symptom is a typical manifestation of a certain disease, then when this symptom is identified, the frequency count of the disease can be appropriately increased.

[0113] Choose a suitable sorting algorithm to sort the disease names. Common sorting algorithms include quick sort, merge sort, etc. Most programming language standard libraries provide built-in sorting functions that can be used directly. Based on the frequency data obtained previously, all disease names are sorted in descending order. This means that the disease with the highest frequency will be at the front of the list, and the diseases with lower frequency will be arranged in order.

[0114] The final list of candidate disease groups is generated based on the sorting results. This list not only shows the names of each disease, but also their frequency information, which helps doctors quickly understand which diseases are the best diagnostic options. In addition to the basic disease name and frequency, you can also consider adding more relevant information to each disease, such as typical symptom descriptions, common treatment options, preventive measures, etc., to provide doctors with more decision support.

[0115] S104 generates a corresponding recommended department table according to the disease group to be selected, and arranges these departments according to priority;

[0116] The specific steps include:

[0117] S801 obtains a corresponding department list based on the disease group to be selected;

[0118] Ensure that the knowledge graph or database contains information about each disease and the corresponding departments responsible for its diagnosis and treatment. This typically includes, but is not limited to, internal medicine, surgery, pediatrics, etc. For the group of candidate diseases obtained in step S404, traverse each disease name and extract the department information related to that disease from the knowledge graph or relevant database. This step needs to handle many-to-many relationships, that is, one disease is treated by multiple departments, and one department is also responsible for treating multiple diseases. Based on the above matching results, generate a preliminary list of departments. Each entry in this list should contain at least two pieces of information: the department name and the disease name it is associated with.

[0119] S802 Merge the same departments;

[0120] Check the preliminarily generated list of departments, identify and merge the repeatedly occurring departments. For example, if the "Department of Respiratory Medicine" appears multiple times in the list (because it is responsible for treating multiple candidate diseases), it should be merged into a single entry. For each merged department, calculate the total number of times it appears in the list, that is, how many candidate diseases it is associated with. This step helps to determine the priority of each department in the subsequent process.

[0121] S803 Arrange the departments based on the frequency of disease matching in the merged departments.

[0122] Determine that the main criterion for sorting is the frequency of matching between the department and the diseases in the candidate disease group. That is, departments associated with more candidate diseases will be given higher priority.

[0123] In addition to the basic frequency, other factors can be considered to adjust the priority of the departments. For example: some diseases are more urgent or severe than others, so the departments associated with them can be given appropriate higher priority. The professional expertise or specific technical capabilities of the departments are also one of the factors affecting the sorting.

[0124] Use the quicksort algorithm to sort the departments in descending order based on the set criteria, ensuring that the most relevant departments are ranked at the front. Finally, generate an ordered list of recommended departments, listing all the departments involved and their relevant information (such as address, phone number, business hours, etc.), and clearly marking the priority of each department. This form can be directly provided to patients or doctors as a reference to help them make more informed choices.

[0125] S105 Correct the sorting order of the departments based on the patient's historical medical records.

[0126] First, it is necessary to securely extract the complete historical medical records of the patient from the hospital's information system or electronic health record (EHR). These records typically include, but are not limited to, past diagnosis results, treatment experiences, surgical history, and departments visited. When handling patient data, relevant privacy protection regulations (such as GDPR or HIPAA) must be strictly adhered to ensure the security and confidentiality of all personal information.

[0127] Carefully analyze the patient's medical history to identify key information related to the current symptoms. For example, if the patient has a chronic medical condition, the department responsible for managing that chronic disease is a priority; or if the patient received effective treatment in the same department previously, that department is also a recommended choice. Check the previous treatment effects of the patient. If a certain department provided an effective treatment plan for the patient in the past and the patient was satisfied with its services, it can be considered a good choice for the current condition.

[0128] Based on the findings in the historical medical records, adjust the originally set department sorting weights. For example, increase the weights of departments that have successfully treated the patient's current or similar symptoms, and reduce or exclude departments with no relevant experience or poor effects. According to the new weight factors, use an appropriate algorithm to re - sort the department list. This step involves complex calculation logic to ensure that the sorting reflects both the importance of the current symptoms and the patient's personal medical background. Considering some special situations, such as the patient being allergic to specific drugs or having a poor response to certain treatment methods, special attention should be paid to avoid recommending departments or treatment plans that cause adverse reactions during sorting.

[0129] After completing the sorting, medical professionals review the results to ensure that the adjusted department recommendations are both in line with medical logic and practical. Provide the final department recommendation list to the patient and establish a feedback mechanism so that the patient can report any discomfort or questions. This can not only improve patient satisfaction but also provide a basis for future service improvement.

[0130] Second Embodiment

[0131] The present invention also provides an intelligent triage and guiding analysis device based on a large - language model, including a symptom collection module, a keyword group acquisition module, a candidate disease matching module, a department arrangement module, and a department correction module; the symptom collection module is used to collect the patient's symptom information through text input or voice input and associate the symptom information with the patient identity information; the keyword group acquisition module is used to clean the collected symptom information and extract symptom keyword groups; the candidate disease matching module is used to match the extracted symptom keywords with the data in the knowledge graph to determine a candidate disease group;

[0132] The department arrangement module is used to generate a corresponding recommended department list according to the disease group to be selected, and arrange these departments according to the priority; the department correction module is used to correct the arrangement order of departments based on the patient's historical medical records.

[0133] The symptom collection module efficiently collects the patient's symptom information through text input or voice input. Whether through a mobile application, website or other digital platforms, users can conveniently describe their health problems. While collecting symptom information, the system will require users to provide basic identity information (such as name, ID number, etc.) to associate the symptoms with specific patients, ensuring the accuracy and personalization of subsequent medical services.

[0134] The keyword group acquisition module cleans the symptom descriptions collected from the patient, removing unnecessary punctuation marks, special characters and formatting errors to ensure data consistency and accuracy. Using natural language processing techniques, especially named entity recognition (NER), key symptom phrases are extracted from the cleaned symptom information. This step also includes advanced functions such as synonym replacement and context understanding to ensure that the extracted keywords are as comprehensive and accurate as possible.

[0135] The disease group to be selected matching module utilizes a pre-constructed medical knowledge graph, which contains a wide range of diseases, symptoms and related information. This graph serves as a basic database to support the disease matching process. Then, an efficient string matching algorithm or machine learning model is adopted to compare the extracted symptom keywords with the data in the knowledge graph to identify relevant diseases. This process takes into account the combination patterns of symptoms and their association strengths with different diseases. A disease group to be selected containing multiple potential diseases is generated according to the matching degree, and a corresponding matching degree score is calculated for each disease for subsequent sorting and recommendation.

[0136] Based on the disease group to be selected, the department arrangement module queries the department information corresponding to each disease to generate a preliminary department list. Check and merge duplicate department entries to avoid the same department being listed multiple times due to multiple diseases. The departments are sorted in descending order according to the matching frequency of each department with the diseases in the disease group to be selected and other factors (such as urgency, specialty expertise, etc.) to form the final recommended department list.

[0137] The department correction module accesses and analyzes the patient's electronic health record (EHR) to understand their past medical history, treatment effects and any special preferences. Based on the findings in the historical records, the originally set department sorting weights are appropriately adjusted. For example, increase the weights of departments that have successfully treated the patient's current or similar symptoms.

[0138] This intelligent triage and diagnosis analysis device based on large language models can not only greatly improve the efficiency and quality of medical services, but also significantly improve the patient's medical experience, enabling them to obtain more accurate and personalized medical guidance.

[0139] Third Embodiment

[0140] The present invention also provides an intelligent triage and diagnosis analysis system based on large language models, including the intelligent triage and diagnosis analysis device based on large language models described above.

[0141] This system utilizes advanced natural language processing technologies (especially large language models) and medical knowledge graphs. By collecting patients' symptom information and conducting in-depth analysis, it recommends the most suitable diagnosis and treatment departments for patients and disease diagnoses. The overall system design takes into account user experience, data security, and medical professionalism to ensure that the services provided are both accurate and reliable.

[0142] The above-disclosed is only a preferred embodiment of the present invention. Of course, it cannot be used to limit the scope of rights of the present invention. Those of ordinary skill in the art can understand the implementation of all or part of the above processes, and equivalent changes made according to the claims of the present invention still fall within the scope covered by the invention.

Claims

1. Intelligent triage and guiding analysis method based on large language models, It is characterized in that it includes: collecting the symptom information of the patient through text input or voice input, and associating the symptom information with the patient identity information; cleaning the collected symptom information and extracting symptom keyword groups; matching the extracted symptom keywords with the data in the knowledge graph to determine the candidate disease groups; generating a corresponding recommended department list according to the candidate disease groups, and arranging these departments according to the priority; correcting the arrangement order of the departments based on the patient's historical medical records.

2. The intelligent triage and guiding analysis method based on the large language model according to claim 1, characterized in that the specific steps of collecting the symptom information of the patient through text input or voice input and associating the symptom information with the patient identity information include: obtaining the first text information input by the patient through text; obtaining the voice information input by the patient, and converting the voice information into the second text information; merging the first text information and the second text information into symptom information and associating it with the patient identity information.

3. The intelligent triage and guiding analysis method based on the large language model according to claim 2, characterized in that the specific steps of cleaning the collected symptom information and extracting symptom keyword groups include: removing the irrelevant information in the collected symptom information to obtain the screened information; splitting the screened information into phrases to be matched; using a pre-defined medical term dictionary, directly extracting keywords from the phrases to be matched through string matching to obtain symptom keyword groups.

4. The intelligent triage and guiding analysis method based on the large language model according to claim 3, characterized in that the specific steps of matching the extracted keywords with the data in the knowledge graph to determine the candidate disease groups include: generating a knowledge graph based on the symptom descriptions, disease names and relevant department information of diseases; grouping the symptom keyword groups provided by the patient to obtain multiple keyword combinations to be matched, and each keyword combination has at least two keywords; matching each type of keyword group to be matched in the knowledge graph to obtain the disease names corresponding to each keyword combination; counting the frequencies of the occurrences of each disease name and arranging the diseases in descending order of the frequencies of occurrence to obtain the candidate disease groups.

5. The intelligent triage and guiding analysis method based on the large language model according to claim 4, characterized in that the specific steps of generating a knowledge graph based on the symptom descriptions, disease names and relevant department information of diseases include: cleaning the collected symptom descriptions, disease names and relevant department information to remove duplicate information to obtain the initial information; using named entity recognition technology to extract core entities in the initial information, and the core entities include disease names, symptom descriptions and department information; using a machine learning model to generate the relationships between the core entities; importing the core entities and their corresponding relationship data into a graph database, establishing the links between the entities and their relationships, and generating a knowledge graph.

6. The intelligent triage and guiding analysis method based on the large language model according to claim 5, characterized in that The specific steps of arranging and grouping the symptom keyword groups provided by the patient to obtain multiple keyword combinations to be matched, where each keyword combination has at least two keywords are as follows: Convert all collected symptom keywords into a unified format; Group the symptom keyword groups based on the body system to which the symptoms belong as the main grouping basis to obtain multiple keyword combinations to be matched.

7. The intelligent triage and guiding analysis method based on a large language model according to claim 6, wherein The specific steps of matching each keyword combination to be matched in the knowledge graph to obtain the disease name corresponding to each keyword combination are as follows: Generate an extended phrase for each keyword combination to be matched; Match the corresponding disease nodes in the knowledge graph based on the extended phrase; Obtain the corresponding disease name based on the disease nodes.

8. The intelligent triage and guiding analysis method based on a large language model according to claim 7, wherein The specific steps of generating a corresponding recommended department table according to the disease group to be selected and arranging these departments according to the priority are as follows: Obtain the corresponding department list based on the disease group to be selected; Merge the same departments; Arrange the departments based on the frequency of disease matching in the merged departments.

9. The intelligent triage and guiding analysis device based on the large language model is characterized in that, It includes a symptom collection module, a keyword group acquisition module, a disease to be selected matching module, a department arrangement module, and a department correction module; The symptom collection module is used to collect the patient's symptom information by means of text input or voice input, and associate the symptom information with the patient identity information; The keyword group acquisition module is used to clean the collected symptom information and extract the symptom keyword groups; The disease to be selected matching module is used to match the extracted symptom keywords with the data in the knowledge graph to determine the disease group to be selected; The department arrangement module is used to generate a corresponding recommended department table according to the disease group to be selected and arrange these departments according to the priority; The department correction module is used to correct the arrangement order of the departments based on the patient's historical medical records.

10. An intelligent triage and guidance analysis system based on a large language model, characterized in that, It includes an intelligent triage and guiding analysis device according to claim 9.

Images