Large model-based health care service intelligent question and answer method and system

Abstract

The invention provides a health-care service intelligent question-answering method and system based on a large model, and relates to the technical field of information retrieval. The method comprises the steps of obtaining health description information input by a user, extracting symptom expression, occurrence time and a duration period, screening symptom influence factor data, calculating a health state change range, and obtaining a health state change interval. According to the method, the health state change interval is constructed by extracting the symptom information in the health description and combining the time, the duration period and the influence factor data, so that the dynamic change of the health state is quantified, and the health state assessment fineness is improved. The change rate of the symptom intensity is calculated based on the time sequence data, the symptom fluctuation mode is identified, and the affected symptom category is screened, so that the prediction of the disease development trend is more accurate. Based on similarity calculation of multi-time-point symptoms and disease courses and in combination with disease co-occurrence frequency data, the reasonability of symptom matching can be enhanced, and deviation in disease identification is reduced.

Description

Technical Field

[0001] The present invention relates to the field of information retrieval technology, and in particular to a large-model-based intelligent question-answering method and system for health care services. Background Art

[0002] The field of information retrieval technology includes related methods and systems for classifying, organizing, storing and querying massive amounts of data. Its core content is to optimize the data search process and improve the accuracy of information matching through natural language processing, machine learning, deep learning and other means. This technical field is widely used in search engines, database management systems, intelligent question-answering systems, recommendation systems, etc. In modern information technology, information retrieval not only involves traditional retrieval methods based on keyword matching, but also includes intelligent retrieval based on semantic analysis and personalized recommendations based on user behavior analysis. With the development of big data and artificial intelligence technology, information retrieval technology continues to improve and evolve towards a more intelligent and precise direction.

[0003] Among them, the intelligent question-answering method for health care services based on large models refers to the use of pre-trained deep learning models to process and understand users' natural language input and provide accurate question-answering services in the field of health care services. The subject of this patent covers semantic analysis of health care service-related data, knowledge graph construction, intention recognition, and answer generation. Specific methods include a semantic matching mechanism based on a neural network model to understand user query intentions, combining medical and health databases to build professional knowledge graphs to support question and answer content generation, using context association methods to optimize the consistency of multiple rounds of conversations, using text summarization methods to extract and summarize health consultation content, and combining user portrait analysis with historical interaction data to optimize personalized question-answering services.

[0004] Existing information retrieval technologies primarily rely on keyword-based matching for search. Even when combined with natural language processing and machine learning to optimize the retrieval process, they still lack the ability to perceive dynamic changes in health status, making it impossible to accurately identify symptom trends, resulting in biased symptom matching. While semantic analysis-based retrieval can improve matching accuracy to a certain extent, it lacks in-depth modeling of symptom co-occurrence relationships and is prone to missing related symptoms, limiting the accuracy of question-and-answer systems in health consultations. Furthermore, personalized recommendations are optimized based on user historical behavior and are difficult to incorporate dynamic adjustments to health status, resulting in health recommendations that fail to fully adapt to users' actual needs. Traditional intelligent question-and-answer systems are susceptible to insufficient contextual constraints in multi-round conversations, resulting in poor continuity and consistency, impacting user experience. For health consultation scenarios, existing methods struggle to effectively integrate health records with individual health status, resulting in a lack of targeted health intervention guidance and impacting the practicality of health management. Summary of the Invention

[0005] In order to solve the technical problems existing in the prior art, the embodiment of the present invention provides a method and system for intelligent question answering of health care services based on a large model. The technical solution is as follows:

[0006] A large-scale model-based intelligent question-answering method for health care services includes the following steps:

[0007] S1: Obtain the health description information input by the user, extract the symptom description, occurrence time, and duration, filter the data of symptom influencing factors, calculate the range of health status changes, and obtain the health status change interval;

[0008] S2: extracting the health status time series data in the health status change interval, calculating the symptom intensity change rate, extracting the symptom fluctuation pattern, screening the affected symptom categories, evaluating the symptom development trend, and obtaining the symptom change trend analysis results;

[0009] S3: Based on the symptom change trend analysis results, calculate the similarity between the symptoms at multiple time points and the course of the disease, and calculate the matching weight based on the symptom co-occurrence frequency statistics to obtain the symptom matching correlation;

[0010] S4: calling the symptom matching correlation, comparing symptom records, analyzing symptom coverage, determining missed symptoms based on the symptom co-occurrence probability library, screening symptom categories that meet health care needs, evaluating symptom matching credibility, and obtaining a revised symptom candidate set;

[0011] S5: Based on the modified disease candidate set, match the health records to screen the health intervention guidance, call the user's health status data to compare the adaptability, and obtain the health intelligent question and answer content.

[0012] Optionally, the health status change interval includes symptom description, occurrence time, duration, health status change range, and symptom influencing factor data; the symptom change trend analysis results include symptom intensity change rate, symptom fluctuation pattern, affected symptom category, and symptom development trend; the symptom matching correlation includes the similarity between symptoms and disease course at multiple time points, symptom co-occurrence frequency statistics, and matching weights; the revised symptom candidate set includes symptom matching correlation, symptom record comparison results, symptom coverage analysis, missed symptom judgment results, symptom categories that meet health care needs, and symptom matching credibility; the health care intelligent question and answer content includes revised symptom candidate set, health care record matching results, health care intervention guidance, and user health status data comparison adaptability.

[0013] Optionally, the specific steps for obtaining the health description information input by the user, extracting the symptom description, occurrence time, and duration period, filtering the data of symptom influencing factors, and calculating the health status change range are as follows:

[0014] S101: Obtain health description information input by the user, extract symptom description, occurrence time and duration, match symptom dictionary with time rules, filter and parse time series information, break it down into occurrence time and duration range, and obtain symptom time series data;

[0015] S102: Based on the symptom time series data, factors influencing symptom changes are screened, symptom descriptions are matched with associated influencing factors, data of each category is classified, the proportion of each factor in symptom changes is calculated, and the data is matched with the corresponding time series information to obtain symptom influencing factor data;

[0016] S103: Call the symptom influencing factor data, combine it with the symptom time series data, calculate the range of health status changes, normalize the symptom duration, intervention time and symptom intensity, analyze the health status change trend, and obtain the health status change interval.

[0017] Optionally, the specific steps of extracting the health status time series data in the health status change interval, calculating the symptom intensity change rate, extracting the symptom fluctuation pattern, screening the affected symptom categories, and evaluating the symptom development trend to obtain the symptom change trend analysis results are as follows:

[0018] S201: Calling the health status time series data in the health status change interval, extracting the symptom intensity value, calculating the change rate of adjacent time nodes, screening the change trend of continuous time periods, and generating the symptom intensity change rate;

[0019] S202: Based on the symptom intensity change rate, calculate the fluctuation amplitudes of multiple time periods, determine whether the symptom changes have periodic characteristics, classify the continuous change intervals, screen the fluctuation patterns that meet the amplitude threshold and time span requirements, calculate the symptom fluctuation frequency and deviation, generate a fluctuation trend data set, and obtain the symptom fluctuation pattern;

[0020] S203: calling the symptom fluctuation pattern, screening the affected symptom categories, calculating the fluctuation frequency and change trend, evaluating the future trend of the symptom categories, screening the main influencing factors, and obtaining the symptom change trend analysis results.

[0021] Optionally, for calculating the fluctuation amplitude A for multiple time periods k , using the formula:

[0022]

[0023] Among them, I i,k represents the symptom intensity value at the i-th time point in the k-th time period, represents the average value of symptom intensity at all time points in the kth time period, T i,krepresents the time interval from the i-th time point to the previous time point in the k-th time period, n represents the total number of time points in the k-th time period, m represents the number of time points in the previous time period, and I j,k-1 represents the symptom intensity value in the previous time period k-1, r represents the number of high fluctuation points in special screening, Represents the average intensity of high volatility points in the kth time period, T p,k Represents the time interval of high fluctuation points in the kth time period, V k Represents the average offset of all fluctuation points in the kth time period.

[0024] Optionally, based on the symptom change trend analysis results, the similarity between the symptoms and the course of the disease at multiple time points is calculated, and the matching weight is calculated based on the symptom co-occurrence frequency statistics to obtain the symptom matching correlation. The specific steps are as follows:

[0025] S301: Based on the symptom change trend analysis results, extract symptom feature data at multiple time points, analyze the similarity of symptom intensity, duration and fluctuation pattern, screen continuous change intervals, calculate the similarity deviation range, and obtain symptom time similarity;

[0026] S302: Invoking the symptom time similarity, screening high-frequency co-occurring symptom categories, analyzing the correlation of symptoms at each stage of the disease course, matching symptoms that meet the co-occurrence frequency threshold, classifying symptom co-occurrence patterns, calculating the co-occurrence probability of symptom categories in multiple time periods, and obtaining symptom co-occurrence matching weights;

[0027] S303: Based on the symptom co-occurrence matching weight, screening symptom categories whose matching weight exceeds a weight threshold, classifying the association intervals between symptoms and symptoms, and obtaining the symptom matching association degree.

[0028] Optionally, the specific steps of calling the symptom matching correlation, comparing symptom records, analyzing symptom coverage, determining missed symptoms based on the symptom co-occurrence probability library, screening symptom categories that meet health care needs, and evaluating symptom matching credibility to obtain a revised symptom candidate set are as follows:

[0029] S401: calling the symptom matching correlation, comparing symptom records, calculating the coverage of symptoms in each symptom, screening symptom categories whose matching degree exceeds the symptom matching coverage threshold, and obtaining symptom coverage data;

[0030] S402: Based on the symptom coverage data, screen out missed symptoms, calculate the co-occurrence ratio of symptoms and symptoms, match the screening weights of symptom categories, classify the symptom types that meet the health care needs, calculate the matching confidence interval of the screened symptoms, and obtain the health care symptom screening results;

[0031] S403: Call the health care disease screening result, calculate the disease matching credibility, screen the matching disease categories, calculate the disease credibility interval, screen the matching diseases whose credibility exceeds the set credibility threshold, and generate a modified disease candidate set.

[0032] Optionally, the specific steps for modifying the candidate set of symptoms, matching health records to screen health intervention guidance, calling user health status data to compare adaptability, and obtaining health intelligent question and answer content are as follows:

[0033] S501: Based on the modified symptom candidate set, matching health care records, extracting health care intervention guidance information that matches the symptom category, screening health care information that matches the symptom characteristics, classifying and arranging intervention content, calculating the matching degree, and obtaining health care intervention matching data;

[0034] S502: Calling the health care intervention matching data, comparing the user's health status data, screening health care information that matches the current health status, calculating the adaptability weight of each health care information, classifying the health care information under each health status, analyzing the weight distribution of the adaptability of the health care information, adjusting the screening range, and obtaining health status adaptation information;

[0035] S503: Based on the health status adaptation information, extract the health care intervention information to calculate the adaptation weight, classify the health care guidance content, adjust the matching weight of the health care questions and answers, and generate health care intelligent question and answer content.

[0036] Optionally, for calculating the adaptability weight W of each health care information i,j , using the formula:

[0037]

[0038] Among them, R p,j represents the health information feedback score of the pth user in the jth health condition, represents the mean of all users’ health information feedback scores under the jth health condition, T p,j represents the time interval of the pth user feedback in the jth health condition, Q represents the number of users in the health condition data of the previous stage, R q,j represents the health information feedback score of the qth user in the jth health condition, R q,j-1 represents the health information feedback score of the qth user in the j-1th health condition, M represents the number of user behavior matches in the current health condition, C m,j represents the behavioral indicator data of the mth user in the jth health condition, represents the mean of all user behavior indicator data in the jth health state, S represents the number of selected high-volatility users, represents the mean score of health information feedback of high-fluctuation users under the jth health condition, Ts,j represents the feedback time interval of the sth high-fluctuation user in the jth health condition, V j represents the average offset of all high-fluctuation users under the jth health condition, D s,j represents the symptom intensity data of the sth high-fluctuation user in the jth health condition, D s,j-1 represents the symptom intensity data of the sth high-fluctuation user in the j-1th health condition, U j Represents the mean symptom change of all high-fluctuation users under the jth health condition.

[0039] A large-scale model-based intelligent question-answering system for health care services, comprising:

[0040] The health data extraction module obtains the health description information input by the user, extracts the symptom description, occurrence time, and duration period, filters the data of symptom influencing factors, calculates the range of health status changes, and obtains the health status change interval;

[0041] The symptom trend analysis module extracts the health status time series data in the health status change interval, calculates the symptom intensity change rate, extracts the symptom fluctuation pattern, screens the affected symptom categories, evaluates the symptom development trend, and obtains the symptom change trend analysis results;

[0042] The symptom association calculation module calculates the similarity between the symptoms and the course of the disease at multiple time points based on the symptom change trend analysis results, and calculates the matching weight based on the symptom co-occurrence frequency statistics to obtain the symptom matching correlation;

[0043] The symptom screening optimization module calls the symptom matching correlation, compares symptom records, analyzes symptom coverage, determines missed symptoms based on the symptom co-occurrence probability library, screens symptom categories that meet health care needs, evaluates symptom matching credibility, and obtains a revised symptom candidate set;

[0044] The health care guidance matching module modifies the candidate set of symptoms, matches health care records to screen health care intervention guidance, calls user health status data to compare adaptability, and obtains health care intelligent question and answer content.

[0045] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0046] By extracting symptom information from health descriptions and combining it with data on time, duration, and influencing factors, a health status change interval is constructed, quantifying the dynamic changes in health status and improving the precision of health status assessment. Symptom trend analysis calculates the rate of change in symptom intensity based on time series data, identifies symptom fluctuation patterns, and selects affected symptom categories, enabling more accurate predictions of symptom development trends. Computing the similarity of symptoms and disease course at multiple time points, combined with symptom co-occurrence frequency data, enhances the rationality of symptom matching and reduces bias in symptom identification. Missing symptom analysis is performed on matched symptoms, and comparison is performed against a symptom co-occurrence probability database for more comprehensive symptom identification. Matching confidence assessment optimizes symptom selection and enhances the reliability of identification results. Health records are matched against a modified set of symptom candidates, and adaptive health status comparison is performed to enable precise delivery of health intervention guidance and more personalized health Q&A content. Through a series of precise calculations, data matching, and intelligent analysis, health Q&A results are more closely aligned with the user's health status, optimizing personalized health guidance and improving the accuracy and intelligence of health services. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0048] Figure 1 is a flow chart of the method of the present invention;

[0049] Figure 2 This is a detailed flow chart of step S1 of the present invention;

[0050] Figure 3 This is a detailed flow chart of step S2 of the present invention;

[0051] Figure 4 This is a detailed flow chart of step S3 of the present invention;

[0052] Figure 5 This is a detailed flow chart of step S4 of the present invention;

[0053] Figure 6 This is a detailed flow chart of step S5 of the present invention;

[0054] Figure 7 It is a system module diagram of the present invention. DETAILED DESCRIPTION

[0055] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0056] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as an "exemplary" in the present invention should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of the word "exemplary" is intended to present concepts in a concrete manner. Furthermore, in the embodiments of the present invention, "and / or" can mean both or either of the two.

[0057] In the embodiments of the present invention, the terms "image" and "picture" may be used interchangeably. It should be noted that, when the distinction between them is not emphasized, their intended meanings are the same. The terms "of," "corresponding," and "corresponding" may be used interchangeably. It should be noted that, when the distinction between them is not emphasized, their intended meanings are the same.

[0058] In the embodiments of the present invention, sometimes a subscript such as W1 may be written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0059] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0060] See also Figure 1 , an embodiment of the present invention provides a health care service intelligent question answering method based on a large model, comprising the following steps:

[0061] S1: Obtain the health description information entered by the user, extract the symptom description, onset time, and duration through the semantic parsing capability of the big model, and when screening the data of symptom influencing factors, perform normalization processing in combination with the symptom dictionary of the big model, calculate the range of health status changes, and obtain the health status change interval;

[0062] S2: Extract health status time series data within the health status change interval, use the big model's deep understanding of time series patterns to calculate the rate of change of symptom intensity, extract symptom fluctuation patterns based on the big model's fluctuation pattern recognition algorithm, screen the affected symptom categories, evaluate symptom development trends, and obtain symptom change trend analysis results;

[0063] S3: Based on the results of symptom change trend analysis, the cross-modal similarity calculation framework of the large model is used to analyze the similarity between symptoms and disease courses at multiple time points. The co-occurrence map of symptoms in the large model knowledge base is combined to count the high-frequency associated symptoms. The matching weight is calculated through the dynamic weight allocation of the large model to obtain the symptom matching correlation degree.

[0064] S4: Invoke the symptom matching correlation, analyze the symptom coverage based on the large model's coverage detection algorithm, determine the missing symptoms based on the symptom co-occurrence probability library, screen the symptom categories that meet the health care needs, evaluate the symptom matching credibility, and apply the large model's credibility assessment model to verify the symptom matching quality, and obtain the revised symptom candidate set;

[0065] S5: Based on the modified candidate set of symptoms, the RAG technology driven by the big model is used to search the health care knowledge base, match the health care records to screen the health care intervention guidance, call the user's health status data through the adaptability evaluation algorithm of the big model to compare the adaptability, use the text generation technology of the big model to construct health care suggestions in natural language form, combine the prompt engineering to optimize the question and answer logic, and obtain health care intelligent question and answer content.

[0066] The health status change interval includes symptom description, occurrence time, duration, health status change range, and symptom influencing factor data. The symptom change trend analysis results include the symptom intensity change rate, symptom fluctuation pattern, affected symptom category, and symptom development trend. The symptom matching correlation includes the similarity between symptoms and disease course at multiple time points, symptom co-occurrence frequency statistics, and matching weights. The revised symptom candidate set includes symptom matching correlation, symptom record comparison results, symptom coverage analysis, missed symptom judgment results, symptom categories that meet health care needs, and symptom matching credibility. The health care intelligent Q&A content includes the revised symptom candidate set, health care record matching results, health care intervention guidance, and user health status data comparison adaptability.

[0067] See also Figure 2 , obtain the health description information input by the user, extract the symptom description, occurrence time, and duration, filter the data of symptom influencing factors, calculate the range of health status changes, and obtain the specific steps of the health status change interval as follows:

[0068] S101: Obtain health description information input by the user, extract symptom description, occurrence time and duration, match symptom dictionary with time rules, filter and parse time series information, break it down into occurrence time and duration range, and obtain symptom time series data;

[0069] Obtain the health description information entered by the user, parse the text content, split the description information into sentences, traverse each sentence, extract the symptom description, occurrence time and duration period involved in it in turn, standardize the extracted text information, use the symptom dictionary to map the symptom description, determine the specific symptom name, and use the time rule library to parse the time words in the sentence for time information, such as "yesterday", "a week ago", "lasted for three days", etc., which correspond to the occurrence time and duration respectively, and convert them into standard time format. For example, it occurs at 12:30 on February 20, 2025 and lasts for 72 hours. Through time series decomposition, multiple Continuous symptom data at each time point is collected. If only a relative description of the symptom onset time is given, the specific time point is inferred based on the context. For example, the specific date corresponding to "the day before yesterday" needs to be calculated by subtracting two days from the current date. All extracted time series data are stored in the form of timestamps, and symptom information is recorded using standard codes, such as ICD-10 codes. Ultimately, symptom time series data is obtained. The semantic role labeling (SRL) technology of the large model is used to identify symptom entities, time modifiers, and degree adverbs in user descriptions, construct structured symptom tuples, and use the domain-adapted BERT-BiLSTM-CRF model to improve the accuracy of medical term recognition.

[0070] S102: Based on the symptom time series data, factors influencing symptom changes are screened, symptom descriptions are matched with associated influencing factors, data for each category is categorized, the proportion of each factor in symptom changes is calculated, and the data is matched with the corresponding time series information to obtain symptom influencing factor data;

[0071] Based on symptom time series data and combined with the knowledge distillation capabilities of large models, symptom association patterns are learned from massive health records, and a dynamic influencing factor weight matrix is established. Graph neural networks (GNNs) are then used to model the complex associations between symptoms, environment, and behavior. Factors that may affect symptom changes are screened, and time series data is matched with user input to extract possible related factors, such as diet, environmental changes, and medication status. Factors before and after symptom onset are categorized and stored. For example, medication information is stored as multidimensional data such as "drug name, dosage, and time of administration," and environmental changes are stored as information such as "temperature, humidity, and air quality." A time window method is used to calculate the frequency of each factor before and after symptom changes. For example, the proportion of times a user's symptoms worsen in high humidity environments relative to all records is considered an influencing factor if it exceeds 60%. The contribution ratio of different factors is calculated, for example, the proportion of humidity's impact on symptom worsening = (number of symptom worsenings in high humidity environments / total number of samples) × 100%, ultimately resulting in symptom influencing factor data.

[0072] S103: Calling symptom influencing factor data, combining it with symptom time series data, calculating the range of health status changes, normalizing symptom duration, intervention time, and symptom intensity, analyzing the health status change trend, and obtaining the health status change interval;

[0073] Symptom influencing factor data is called up and combined with symptom time series data to calculate the range of health status changes. First, symptom duration is normalized and all symptom durations are mapped to the range of 0-1. For example, if a symptom with a maximum duration of 120 hours lasts for 30 hours, the normalized value is 30 / 120 = 0.25. Similarly, the intervention duration and symptom intensity are normalized. For example, if the intervention duration is set to a maximum of 48 hours and a single intervention lasts for 12 hours, the normalized value is 12 / 48 = 0.25. The trend of health status change is calculated, and a sliding window method is used to analyze symptom changes. For example, using a 5-day window, the number of symptom exacerbations within each window is counted. If the symptom exacerbates more than twice within three consecutive windows, the health status is considered to have deteriorated. After all data are calculated, the health status change range is obtained. Based on the probabilistic inference engine of the large model, different health state transition paths are simulated, and confidence intervals are generated through Monte Carlo Tree Search (MCTS). The long-term dependency modeling capabilities of Transformer-XL are integrated to capture periodic changes.

[0074] See also Figure 3 The specific steps for extracting health status time series data in the health status change interval, calculating the symptom intensity change rate, extracting the symptom fluctuation pattern, screening the affected symptom categories, and evaluating the symptom development trend to obtain the symptom change trend analysis results are as follows:

[0075] S201: Calling the health status time series data in the health status change interval, extracting the symptom intensity value, calculating the change rate of adjacent time nodes, screening the change trend of continuous time periods, and generating the symptom intensity change rate;

[0076] Applying the Time-aware Attention mechanism of the large model, a feature extractor of multi-scale time windows (hours / days / weeks) is constructed. The nonlinear evolution of symptom intensity is captured by the TCN temporal convolutional network, and the symptom intensity value in the time series is extracted. The symptom intensity value at each time point is converted from the symptom description entered by the user. For example, the cough intensity can be quantified based on frequency and duration. The value range is set to 0-10, where 0 means no symptoms and 10 means the most severe case. The change rate of adjacent time nodes is calculated by dividing the intensity difference between the two time points by the time interval. For example, the symptom intensity at time points t1 and t2 is s1 respectively. =5 and s2=8, time interval Δt=t2-t1=3 hours, then the rate of change Δs / Δt=(8-5) / 3=1 unit / hour, screen the change trend of continuous time periods, and judge the trend of continuous strengthening, weakening or stability of symptoms according to the sign consistency of the change rates of adjacent time points. For example, if the intensity values of four consecutive time points t1, t2, t3, and t4 are 5, 6, 7, and 8, it is determined to be a trend of symptom strengthening. On the contrary, if the intensity values are 8, 7, 6, and 5, it is determined to be a trend of symptom weakening. When the rate of change is within the range of ±0.5, it is classified as a symptom stable period, and finally the symptom intensity change rate is generated.

[0077] S202: Based on the symptom intensity change rate, calculate the fluctuation amplitudes for multiple time periods, determine whether the symptom changes have periodic characteristics, classify the continuous change intervals, screen the fluctuation patterns that meet the amplitude threshold and time span requirements, calculate the symptom fluctuation frequency and deviation, generate a fluctuation trend data set, and obtain the symptom fluctuation pattern;

[0078] Based on the rate of change of symptom intensity, calculate the fluctuation amplitude of multiple time periods, determine whether the symptom change has periodic characteristics, classify the continuous change intervals, screen the fluctuation mode that meets the amplitude threshold and time span requirements, and calculate the symptom fluctuation frequency and deviation, where the fluctuation amplitude A k , using the formula:

[0079]

[0080] Calculate the fluctuation characteristic value A k , generate a fluctuation trend data set and obtain the symptom fluctuation pattern;

[0081] Among them, A k Represents the fluctuation amplitude of the kth time period, I i,k represents the symptom intensity value at the i-th time point in the k-th time period, represents the average value of symptom intensity at all time points in the kth time period, T i,krepresents the time interval from the i-th time point to the previous time point in the k-th time period, n represents the total number of time points in the k-th time period, m represents the number of time points in the previous time period, and I j,k-1 represents the symptom intensity value in the previous time period k-1, r represents the number of high fluctuation points in special screening, Represents the average intensity of high volatility points in the kth time period, T p,k Represents the time interval of high fluctuation points in the kth time period, V k Represents the average offset of all fluctuation points in the kth time period.

[0082] Detailed explanation of the formula and the process of formula calculation and derivation:

[0083] This formula is used to calculate the fluctuation amplitude A within a certain time period k k The calculation method comprehensively considers the change in symptom intensity within the current time period, the impact of the previous time period, and the shift in the peak fluctuation point. The numerator in the formula measures the deviation in symptom intensity, while the denominator is normalized by the time interval and the change in symptoms between the previous and next time periods, making the fluctuation amplitude calculation more accurate.

[0084] Parameter setting and value acquisition

[0085] Symptom Intensity I i,k :

[0086] This parameter represents the symptom intensity value at the i-th time point within the k-th time period, and is usually obtained through physiological monitoring equipment or patient self-assessment scales (such as NRS scores).

[0087] For example, in time period k, the symptom intensity data points of a patient are 5, 7, and 6, corresponding to I 1,k =5, I 2,k =7, I 3,k =6.

[0088] Time period average strength

[0089] This parameter calculates the average of all symptom intensities in the current time period k:

[0090]

[0091] For example, if time period k contains 3 data points (I 1,k =5, I 2,k =7, I 3,k =6), then:

[0092]

[0093] Time interval T i,k :

[0094] This parameter represents the time interval between adjacent data points in the kth time period and is calculated using the timestamp of the monitoring device.

[0095] For example, if time point i occurs at 10:00 and time point i+1 occurs at 10:15, then T i,k =15 minutes.

[0096] Difference in the mean intensity of the previous period

[0097] This calculates the mean of the absolute intensity differences between all data points in the current time period and all data points in the previous time period:

[0098]

[0099] For example, there are two data points I in the previous time period k-1 1,k-1 =4, I 2,k-1 =5, then for time period k, I 1,k =5:

[0100]

[0101] Average strength of high volatility points

[0102] Calculation formula:

[0103]

[0104] For example, in time period k, select two high volatility points (I 1,k =5, I 2,k =7), then:

[0105]

[0106] Fluctuation offset V k :

[0107] This item is used to measure the degree of deviation from the high volatility point and is calculated as follows:

[0108]

[0109] For example, if r = 2, and I 1,k =5, I 2,k =7, then:

[0110]

[0111] Substitute the above values into the formula to calculate the fluctuation amplitude A k :

[0112] Calculate the numerator and denominator of each term: For the first term: I 1,k =5, T 1,k =15,

[0113] I 2,k =7;

[0114]

[0115] I 3,k =6;

[0116]

[0117] Find the sum of squares:

[0118] (0.0645) 2 +(0.0645) 2 +0 2 =0.00416+0.00416+0=0.00832;

[0119] For the second item:

[0120] I 1,k =5, T 1,k =15, V k =1;

[0121]

[0122] I 2,k =7;

[0123]

[0124] Sum:

[0125] 0.0625+0.0625=0.125;

[0126] Final calculation:

[0127]

[0128] The results show that the symptom fluctuation amplitude in this time period is 0.3652, indicating that the degree of change in symptom intensity is relatively small within this time period. Further judgment can be made in combination with the fluctuation frequency, and ultimately used to screen fluctuation patterns that meet the amplitude threshold and time span requirements. A mixture density network (MDN) is developed to quantify the uncertainty of symptom fluctuations, and a typical fluctuation pattern library is generated in combination with a variational autoencoder (VAE). Contrastive learning is used to enhance pattern discrimination capabilities.

[0129] S203: Invoke the symptom fluctuation model, screen the affected symptom categories, calculate the fluctuation frequency and change trend, evaluate the future trend of the symptom categories, screen the main influencing factors, and obtain the symptom change trend analysis results;

[0130] A multi-task learning framework was constructed to simultaneously predict symptom trends and inflection point probabilities. The LSTM-Transformer hybrid architecture was integrated to balance long-term and short-term dependencies with global pattern capture. If the fluctuation frequency f of a symptom across multiple time periods is ≥ 0.05 times / hour, the symptom is considered to be rapidly changing. If the fluctuation frequency f is ≤ 0.01 times / hour, the symptom is considered to be slowly changing. Future trends of symptom categories were assessed, and extrapolated predictions were made using the trends of the most recent n time periods. For example, if symptom intensity showed an upward trend over the past 72 hours, it was predicted that symptoms would likely worsen within the next 24 hours. Key influencing factors were screened, and their contribution rates were calculated. This ratio of the number of occurrences of the influencing factors to the total number of fluctuations was used. For example, if humidity changes occurred 60 times out of 100 fluctuations, the contribution rate of humidity is 60 / 100 = 60%. This ultimately yielded the symptom trend analysis results.

[0131] See also Figure 4 Based on the results of symptom change trend analysis, the similarity between symptoms and disease course at multiple time points is calculated. According to the statistical data of symptom co-occurrence frequency and matching weight, the specific steps to obtain the symptom matching correlation are as follows:

[0132] S301: Based on the symptom change trend analysis results, extract symptom feature data at multiple time points, analyze the similarity of symptom intensity, duration, and fluctuation pattern, screen the continuous change interval, calculate the similarity deviation range, and obtain the symptom time similarity;

[0133] Based on the results of symptom change trend analysis, a multimodal contrastive learning framework was designed to map symptom time series data and standard disease course descriptions into a unified semantic space. A hybrid measurement method of cosine similarity and dynamic time warping (DTW) was used to extract symptom feature data at multiple time points, identify representative time points in the time series, and extract symptom intensity I, duration T, and fluctuation pattern W from the symptom data. Symptom intensity is calculated based on user subjective scores or physiological monitoring indicators, duration is the time interval from the start to the end of the symptom, and fluctuation pattern is determined by the rate of change of intensity. The symptom features at each time point are standardized and stored in vector form. For example, the feature vectors of a symptom at time points t1, t2, and t3 are (6, 12, 0.3), (7, 15, 0.4), and (5, 10, 0.2), respectively. The first dimension represents symptom intensity I, the second dimension represents duration T, and the third dimension represents fluctuation pattern W. The similarity between different time points is calculated using the Euclidean distance formula: Where i and j are time point numbers. For example, the similarity between time points t1 and t2 is calculated as: Filter the continuous change interval and set the similarity threshold D th =4, if the Euclidean distance D between adjacent time points ij ≤D th , it is regarded as a continuous change interval, and the similarity deviation range is calculated, and the similarity mean μ is used. D and standard deviation σ D , define the deviation range: [μ D -σ D ,μ D +σ D ], if the similarity at a certain time point exceeds this range, it is considered that the symptom characteristics at that time point have abnormal fluctuations. If μ is calculated at multiple time points, D =3.2,σ D =0.5, then the normal variation range is [2.7, 3.7]. If the similarity calculated at a certain time point is 4.5, it is judged as an abnormal fluctuation, and the symptom time similarity is finally obtained.

[0134] S302: Invoke symptom temporal similarity, screen for frequently co-occurring symptom categories, analyze the relevance of symptoms at each stage of the disease, match symptoms that meet the co-occurrence frequency threshold, classify symptom co-occurrence patterns, calculate the co-occurrence probability of symptom categories in multiple time periods, and obtain symptom co-occurrence matching weights;

[0135] Based on the large model, a dynamic disease knowledge graph is constructed. The topological features of disease nodes are learned through the GraphSAGE algorithm. The GAT attention mechanism is developed to quantify the correlation strength between diseases. The temporal similarity of symptoms is called, and high-frequency co-occurring disease categories are screened. The symptom time periods with high temporal similarity are extracted. The number of co-occurrences of different diseases in similar time points is counted, and the co-occurrence frequency is calculated. For example, if symptoms A and B co-occur M = 40 times in a time window N = 100, the co-occurrence frequency calculation formula is: Analyze the correlation between symptoms at each stage of the disease course, divide the disease course into early, middle and late stages, and calculate the co-occurrence ratio of symptoms in each stage, such as the co-occurrence frequency F in the early stage of the disease early =0.3, mid-term co-occurrence frequency F mid =0.5, late co-occurrence frequency F late =0.2, matching symptoms that meet the co-occurrence frequency threshold, setting the threshold F th =0.25, filter out the co-occurrence frequency F AB ≥F thThe co-occurrence pattern of symptoms is classified, and the symptoms with high co-occurrence frequency are divided into different categories, such as upper respiratory tract infection related symptoms (cough, runny nose, sore throat) are classified into one category, and gastrointestinal tract related symptoms (abdominal pain, nausea, vomiting) are classified into one category. The co-occurrence probability of the symptom category in multiple time periods is calculated, such as the co-occurrence probability of symptom category C in N T = M appears in 50 time periods T =30 times, then the formula for calculating the co-occurrence probability is: Finally, the disease co-occurrence matching weight is obtained.

[0136] S303: Based on the co-occurrence matching weight of the symptom, screening the symptom categories whose matching weight exceeds the weight threshold, classifying the correlation intervals between the symptoms and the symptom, and obtaining the symptom matching correlation degree;

[0137] Introduce reinforcement learning mechanism (PPO algorithm) to dynamically adjust matching weights based on user feedback. Construct a dual-tower structure model to separate symptom representation learning and weight decision-making processes, filter out disease categories whose matching weights exceed the weight threshold, and set the weight threshold W. th =0.3, filter out the co-occurrence matching weight W X ≥W th Symptoms, such as the matching weight W of disease D D =0.45, the disease is selected, the correlation interval between symptoms and symptoms is classified, and the time range of the symptoms is recorded. For example, if the symptom D co-occurs from time t1 to t5, the correlation interval is [t1, t5]. The matching degree R between symptoms and symptoms is calculated using the co-occurrence ratio of symptoms and symptoms. The calculation formula is as follows: Among them, M E is the number of times symptom E and disease D co-occur, N E is the total number of observations, such as M j =30,N E =100, then we get: Finally, the disease matching correlation is obtained.

[0138] See also Figure 5 , call the symptom matching correlation, compare symptom records, analyze the symptom coverage, determine the missing symptoms based on the symptom co-occurrence probability library, screen the symptom categories that meet the health care needs, evaluate the symptom matching credibility, and obtain the modified symptom candidate set. The specific steps are as follows:

[0139] S401: Calling the symptom matching correlation, comparing symptom records, calculating the coverage of symptoms in each symptom, screening symptom categories whose matching degree exceeds the symptom matching coverage threshold, and obtaining symptom coverage data;

[0140] Call the symptom matching correlation, compare the symptom records, extract the symptom data corresponding to all symptom categories, calculate the coverage of each symptom in different symptoms, and use the coverage calculation formula: Among them C i,j Representative symptoms S i In disease D j The coverage in S i,j For disease D j Related symptoms include symptom S i The number of j For disease D j The total number of related symptoms. For example, if a disease D1 is associated with 20 symptoms, and a symptom S5 appears 12 times in the disease record, then calculate The screening match exceeds the disease matching coverage threshold C th The disease category is set as C th =0.5, such as the calculated value C of a disease D2 7,2 =0.55 exceeds the threshold, it is selected into the matching list, and finally the symptom coverage data is obtained. The medical diagnosis rules are encoded into the coverage evaluation model through knowledge distillation.

[0141] S402: Based on the symptom coverage data, screen out missed symptoms, calculate the co-occurrence ratio of symptoms and symptoms, match the screening weights of symptom categories, classify the symptom types that meet the health care needs, calculate the matching confidence interval of the screened symptoms, and obtain the health care symptom screening results;

[0142] Based on the symptom coverage data, missing symptoms were screened, symptoms that were missing in the matching symptoms but frequently appeared in similar symptoms were extracted, and the co-occurrence ratio of symptoms and symptoms was calculated using the formula Among them, P i,j Representative symptoms S i In disease D j The co-occurrence ratio in O i,j For disease D j Symptoms appear in S i The number of times, N j is the total number of observations of the disease. For example, among 200 cases of disease D3, symptom S9 appears 150 times. The screening weights of matching disease categories are calculated, and the matching confidence intervals of screening diseases are calculated using the mean μ P and standard deviation σ P Calculate the confidence interval [μ P -z·σ P ,μ P +z·σ P ], where z is the standard score for setting the confidence level, if μ P =0.6,σ P=0.1, z=1.96 (95% confidence level), then the confidence interval is calculated as [0.6-1.96×0.1,0.6+1.96×0.1]=[0.404,0.796]. The symptom data whose disease categories are within the confidence interval are screened out, and finally the health care disease screening results are obtained. The missing symptoms are predicted in combination with the disease prototype library, and the conditional dependency relationship between symptoms is modeled using the Bayesian network.

[0143] S403: Call the health care disease screening results, calculate the disease matching credibility, screen the matching disease categories, calculate the disease credibility interval, screen the matching diseases whose credibility exceeds the set credibility threshold, and generate a modified disease candidate set;

[0144] Construct a multi-objective ranking model (NDCG optimization), balance the disease matching degree and the availability of health care resources, use distillation ranking learning (DistillRank) to improve the generalization ability in small sample scenarios, calculate the credible interval of the matching disease category, and use the confidence calculation formula Where T j Representative symptoms D j The matching confidence, M j is the number of observations that meet the matching criteria for the disease, N j is the total number of observations. For example, if 90 cases of disease D4 meet the matching criteria among 150 cases, then The screening credibility exceeds the set credibility threshold T th Matching symptoms, set T yh =0.55. If the symptom D5 is calculated as T5=0.58, which exceeds the threshold, the symptom will enter the modified symptom candidate set, and finally the modified symptom candidate set will be generated.

[0145] See also Figure 6 ,According to the modified disease candidate set, matching health care records to screen health care intervention guidance, calling user health status data to compare adaptability, and obtaining health care intelligent question and answer content, the specific steps are as follows:

[0146] S501: Based on the modified symptom candidate set, matching health care records, extracting health care intervention guidance information that matches the symptom category, screening health care information that matches the symptom characteristics, classifying and organizing intervention content, calculating the matching degree, and obtaining health care intervention matching data;

[0147] Based on the modified symptom candidate set, matching health care records, extracting health care intervention guidance information that meets the symptom category, screening health care information corresponding to symptom characteristics, retrieving all intervention plans from the health care database, and classifying them according to symptom type. For example, intervention information for digestive system diseases includes diet adjustment, exercise guidance, psychological adjustment, etc. Screening health care information that meets symptom characteristics, extracting specific measures involving symptom relief and recovery management, such as gastritis patients are suitable for low-fat and easily digestible foods, and avoidance of irritating foods. Classify and organize intervention content, and divide intervention measures into categories such as daily diet, exercise plan, and traditional Chinese medicine conditioning. Calculate the degree of matching using the matching degree calculation formula: Among them, M i,j Representative symptoms D j and intervention information G j The matching degree, N i,j For disease D i Applicable intervention G j The number of T j = is the total number of intervention options for the disease. For example, if there are 20 intervention options for disease D1, 12 of which meet the specific characteristics of the disease, then Finally, the health care intervention matching data was obtained by constructing a multi-level index structure. The first level was based on the precise matching of disease codes, the second level used ColBERT vector retrieval, and the third level optimized the results through large-scale model reordering (listwise ranking).

[0148] S502: Calling health care intervention matching data, comparing the user's health status data, screening health care information that matches the current health status, calculating the adaptability weight of each health care information, classifying the health care information under each health status, analyzing the weight distribution of the health care information adaptability, adjusting the screening range, and obtaining health status adaptation information;

[0149] Call the health care intervention matching data, compare the user's health status data, screen the health care information that matches the current health status, calculate the adaptability weight of each health care information, and classify the health care information under each health status. The adaptability weight of each health care information is calculated using the formula:

[0150]

[0151] Calculate the weight of health care information adaptability, analyze the weight distribution of health care information adaptability, adjust the screening range, and obtain health status adaptation information;

[0152] Among them, W i,j represents the adaptability weight of health care intervention i under health status j, R p,j represents the health information feedback score of the pth user in the jth health condition, represents the mean of all users’ health information feedback scores under the jth health condition, Tp,j represents the time interval of the pth user feedback in the jth health condition, Q represents the number of users in the health condition data of the previous stage, R q,j represents the health information feedback score of the qth user in the jth health condition, R q,j-1 represents the health information feedback score of the qth user in the j-1th health condition, M represents the number of user behavior matches in the current health condition, C m,j represents the behavioral indicator data of the mth user in the jth health condition, represents the mean of all user behavior indicator data in the jth health state, S represents the number of selected high-volatility users, represents the mean score of health information feedback of high-fluctuation users under the jth health condition, T s,j represents the feedback time interval of the sth high-fluctuation user in the jth health condition, V j represents the average offset of all high-fluctuation users under the jth health condition, D s,j represents the symptom intensity data of the sth high-fluctuation user in the jth health condition, D s,j-1 represents the symptom intensity data of the sth high-fluctuation user in the j-1th health condition, U j represents the mean symptom change of all high-fluctuation users under the jth health condition;

[0153] Detailed explanation of the formula and the process of formula calculation and derivation:

[0154] This formula calculates the adaptability weight of wellness information across different health conditions. It normalizes user wellness feedback, behavioral data, and symptom changes to comprehensively measure the adaptability of wellness interventions. The numerator accounts for fluctuations in all users' wellness feedback and behavioral indicators, while the denominator calculates the differences in feedback and symptom changes for users with high volatility, ensuring that the adaptability weight comprehensively considers the impact of feedback stability, behavioral consistency, and symptom fluctuations.

[0155] Parameter value acquisition:

[0156] User health care feedback score R p,j The data is obtained through users' ratings of health care intervention programs or questionnaire surveys. The rating range is set between 0 and 100. User 1 gave a feedback of 85, user 2 gave a feedback of 78, and user 3 gave a feedback of 90.

[0157] Average health and wellness feedback Calculate the mean of all users' health feedback scores under the current health status:

[0158]

[0159] Assume P = 3, then:

[0160]

[0161] User feedback time interval T p,j Calculated by the timestamps of user data submission, the interval for user 1 is 5 days, for user 2 it is 6 days, and for user 3 it is 4 days.

[0162] User feedback fluctuations in the previous stage Calculate the average difference between the current feedback value of all users and the feedback value of the previous stage: Set Q = 3, and the user feedback of the previous stage is user 1: 80, user 2: 75, and user 3: 85, then:

[0163]

[0164] User behavior indicator C m,j Quantified by indicators such as number of steps, eating habits, and exercise frequency, User 1: 7000 steps, User 2: 6500 steps, User 3: 7200 steps.

[0165] User behavior average Calculate the mean of behavioral indicators:

[0166]

[0167] High volatility user feedback mean Select users with large feedback fluctuations for average calculation, assuming that the feedback fluctuations of users 1 and 3 are large:

[0168]

[0169] High Fluctuation User Symptom Intensity D s,j Collected through health monitoring equipment, User 1: Symptom intensity 5, User 2: Symptom intensity 6, User 3: Symptom intensity 4.

[0170] Symptom intensity in the previous stage D s,j-1 According to historical data, User 1 has symptom intensity of 6, User 2 has symptom intensity of 7, and User 3 has symptom intensity of 5.

[0171] Mean symptom change U j Calculate the mean change in symptom intensity for all users:

[0172]

[0173] Substitute the values into the calculation:

[0174] Calculate the numerator:

[0175]

[0176]

[0177] Calculate the denominator:

[0178]

[0179] Calculate the fitness weight:

[0180]

[0181] The results show that the adaptability weight of the health care intervention under the current health status is 208.7. The higher the value, the more suitable the intervention is for the current health status of the user. Subsequently, the health care intervention can be screened by setting the adaptability weight threshold to ensure that the intervention plan with high matching degree is recommended first.

[0182] S503: Based on the health status adaptation information, extract the health care intervention information to calculate the adaptation weight, classify the health care guidance content, adjust the matching weight of the health care question and answer, and generate the health care intelligent question and answer content;

[0183] The Retrieval-Augmented Generation (RAG) architecture is used to integrate structured medical guidelines with unstructured case libraries. A chain-of-verification mechanism is designed to automatically detect the medical compliance of generated content. For example, for patients with chronic diseases, intervention measures such as dietary adjustment, exercise rehabilitation, and psychological support are selected, and the fitness weight is calculated using the formula: Among them A i,j Represents health status H i and intervention plan G j The fitness weight, V i,j Health status H i The number of cases that are suitable for this intervention plan, Q j is the total number of intervention options. For example, if a health condition H4 is suitable for 30 intervention options, and 18 of them meet the adaptation criteria, then Categorize health care guidance content and divide each intervention plan into different health status categories, such as low-carb diet and high-fiber food conditioning for diabetic patients, adjust the matching weight of health care questions and answers, and set the matching threshold A th , such as setting A th =0.55, screen out the plans with matching degree higher than the threshold, such as an intervention plan calculated as A 5,j =0.58 exceeds the threshold, it will be included in the final health care intelligent question and answer database, and finally generate health care intelligent question and answer content.

[0184] See also Figure 7 , an intelligent question-answering system for health care services based on a large model, the system includes:

[0185] The health data extraction module obtains the health description information input by the user, extracts the symptom description, occurrence time, and duration period, filters the data of symptom influencing factors, calculates the range of health status changes, and obtains the health status change interval;

[0186] The symptom trend analysis module extracts the health status time series data in the health status change interval, calculates the symptom intensity change rate, extracts the symptom fluctuation pattern, screens the affected symptom categories, evaluates the symptom development trend, and obtains the symptom change trend analysis results;

[0187] The symptom association calculation module calculates the similarity between symptoms and disease course at multiple time points based on the symptom change trend analysis results, and calculates the matching weight based on the symptom co-occurrence frequency statistics to obtain the symptom matching correlation;

[0188] The symptom screening and optimization module calls the symptom matching correlation, compares symptom records, analyzes symptom coverage, determines missed symptoms based on the symptom co-occurrence probability library, screens symptom categories that meet health care needs, evaluates symptom matching credibility, and obtains a revised symptom candidate set;

[0189] The health care guidance matching module modifies the candidate set of symptoms, matches health care records to screen health care intervention guidance, calls user health status data to compare adaptability, and obtains health care intelligent question and answer content.

[0190] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A large-scale model-based intelligent question-answering method for health care services, characterized in that: The following steps are involved: S1: Obtain the health description information input by the user, extract the symptom description, occurrence time, and duration, filter the data of symptom influencing factors, calculate the range of health status changes, and obtain the health status change interval; S2: extracting the health status time series data in the health status change interval, calculating the symptom intensity change rate, extracting the symptom fluctuation pattern, screening the affected symptom categories, evaluating the symptom development trend, and obtaining the symptom change trend analysis results; S3: Based on the symptom change trend analysis results, calculate the similarity between the symptoms at multiple time points and the course of the disease, and calculate the matching weight based on the symptom co-occurrence frequency statistics to obtain the symptom matching correlation; S4: calling the symptom matching correlation, comparing symptom records, analyzing symptom coverage, determining missed symptoms based on the symptom co-occurrence probability library, screening symptom categories that meet health care needs, evaluating symptom matching credibility, and obtaining a revised symptom candidate set; S5: Based on the modified disease candidate set, match the health records to screen the health intervention guidance, call the user's health status data to compare the adaptability, and obtain the health intelligent question and answer content.

2. The intelligent question-answering method for health care services based on a large model according to claim 1 is characterized by: The health status change interval includes symptom description, occurrence time, duration, health status change range, and symptom influencing factor data. The symptom change trend analysis results include symptom intensity change rate, symptom fluctuation pattern, affected symptom category, and symptom development trend. The symptom matching correlation includes the similarity between symptoms and disease course at multiple time points, symptom co-occurrence frequency statistics, and matching weights. The revised symptom candidate set includes symptom matching correlation, symptom record comparison results, symptom coverage analysis, missed symptom judgment results, symptom categories that meet health care needs, and symptom matching credibility. The health care intelligent question and answer content includes revised symptom candidate set, health care record matching results, health care intervention guidance, and user health status data comparison adaptability.

3. The large-model-based intelligent question-answering method for health care services according to claim 1 is characterized by: The specific steps for obtaining the health description information entered by the user, extracting the symptom description, occurrence time, and duration, filtering the data of symptom influencing factors, and calculating the range of health status changes are as follows: S101: Obtain health description information input by the user, extract symptom description, occurrence time and duration, match symptom dictionary with time rules, filter and parse time series information, break it down into occurrence time and duration range, and obtain symptom time series data; S102: Based on the symptom time series data, factors influencing symptom changes are screened, symptom descriptions are matched with associated influencing factors, data of each category is classified, the proportion of each factor in symptom changes is calculated, and the data is matched with the corresponding time series information to obtain symptom influencing factor data; S103: Call the symptom influencing factor data, combine it with the symptom time series data, calculate the range of health status changes, normalize the symptom duration, intervention time and symptom intensity, analyze the health status change trend, and obtain the health status change interval.

4. The intelligent question-answering method for health care services based on a large model according to claim 1 is characterized by: The specific steps of extracting the health status time series data in the health status change interval, calculating the symptom intensity change rate, extracting the symptom fluctuation pattern, screening the affected symptom categories, evaluating the symptom development trend, and obtaining the symptom change trend analysis results are as follows: S201: Calling the health status time series data in the health status change interval, extracting the symptom intensity value, calculating the change rate of adjacent time nodes, screening the change trend of continuous time periods, and generating the symptom intensity change rate; S202: Based on the symptom intensity change rate, calculate the fluctuation amplitudes of multiple time periods, determine whether the symptom changes have periodic characteristics, classify the continuous change intervals, screen the fluctuation patterns that meet the amplitude threshold and time span requirements, calculate the symptom fluctuation frequency and deviation, generate a fluctuation trend data set, and obtain the symptom fluctuation pattern; S203: calling the symptom fluctuation pattern, screening the affected symptom categories, calculating the fluctuation frequency and change trend, evaluating the future trend of the symptom categories, screening the main influencing factors, and obtaining the symptom change trend analysis results.

5. The intelligent question-answering method for health care services based on a large model according to claim 1 is characterized in that: For calculating the fluctuation range A for multiple time periods k , using the formula: Among them, I i,k represents the symptom intensity value at the i-th time point in the k-th time period, represents the average value of symptom intensity at all time points in the kth time period, T i,k represents the time interval from the i-th time point to the previous time point in the k-th time period, n represents the total number of time points in the k-th time period, m represents the number of time points in the previous time period, and I j,k-1 represents the symptom intensity value in the previous time period k-1, r represents the number of high fluctuation points in special screening, Represents the average intensity of high volatility points in the kth time period, T p,k Represents the time interval of high fluctuation points in the kth time period, V k Represents the average offset of all fluctuation points in the kth time period.

6. The large-model-based intelligent question-answering method for health care services according to claim 1 is characterized by: Based on the symptom change trend analysis results, the similarity between symptoms and disease course at multiple time points is calculated. The specific steps for obtaining the symptom matching correlation are as follows: S301: Based on the symptom change trend analysis results, extract symptom feature data at multiple time points, analyze the similarity of symptom intensity, duration and fluctuation pattern, screen continuous change intervals, calculate the similarity deviation range, and obtain symptom time similarity; S302: Invoking the symptom time similarity, screening high-frequency co-occurring symptom categories, analyzing the correlation of symptoms at each stage of the disease course, matching symptoms that meet the co-occurrence frequency threshold, classifying symptom co-occurrence patterns, calculating the co-occurrence probability of symptom categories in multiple time periods, and obtaining symptom co-occurrence matching weights; S303: Based on the symptom co-occurrence matching weight, screening symptom categories whose matching weight exceeds a weight threshold, classifying the association intervals between symptoms and symptoms, and obtaining the symptom matching association degree.

7. The large-model-based intelligent question-answering method for health care services according to claim 1 is characterized by: The specific steps of calling the symptom matching correlation, comparing symptom records, analyzing symptom coverage, determining missed symptoms based on the symptom co-occurrence probability library, screening symptom categories that meet health care needs, and evaluating symptom matching credibility to obtain a revised symptom candidate set are as follows: S401: calling the symptom matching correlation, comparing symptom records, calculating the coverage of symptoms in each symptom, screening symptom categories whose matching degree exceeds the symptom matching coverage threshold, and obtaining symptom coverage data; S402: Based on the symptom coverage data, screen out missed symptoms, calculate the co-occurrence ratio of symptoms and symptoms, match the screening weights of symptom categories, classify the symptom types that meet the health care needs, calculate the matching confidence interval of the screened symptoms, and obtain the health care symptom screening results; S403: Call the health care disease screening result, calculate the disease matching credibility, screen the matching disease categories, calculate the disease credibility interval, screen the matching diseases whose credibility exceeds the set credibility threshold, and generate a modified disease candidate set.

8. The large-model-based intelligent question-answering method for health care services according to claim 1 is characterized by: The specific steps for correcting the candidate set of symptoms, matching health records to screen health intervention guidance, and comparing the user's health status data to determine adaptability are as follows: S501: Based on the modified symptom candidate set, matching health care records, extracting health care intervention guidance information that matches the symptom category, screening health care information that matches the symptom characteristics, classifying and arranging intervention content, calculating the matching degree, and obtaining health care intervention matching data; S502: Calling the health care intervention matching data, comparing the user's health status data, screening health care information that matches the current health status, calculating the adaptability weight of each health care information, classifying the health care information under each health status, analyzing the weight distribution of the adaptability of the health care information, adjusting the screening range, and obtaining health status adaptation information; S503: Based on the health status adaptation information, extract the health care intervention information to calculate the adaptation weight, classify the health care guidance content, adjust the matching weight of the health care questions and answers, and generate health care intelligent question and answer content.

9. The large-model-based intelligent question-answering method for health care services according to claim 1, characterized in that: For calculating the adaptability weight W of each health care information i,j , using the formula: Among them, R p,j represents the health information feedback score of the pth user in the jth health condition, represents the mean of all users’ health information feedback scores under the jth health condition, T p,j represents the time interval of the pth user feedback in the jth health condition, Q represents the number of users in the health condition data of the previous stage, R q,j represents the health information feedback score of the qth user in the jth health condition, R q,j-1 represents the health information feedback score of the qth user in the j-1th health condition, M represents the number of user behavior matches in the current health condition, C m,j represents the behavioral indicator data of the mth user in the jth health condition, represents the mean of all user behavior indicator data in the jth health state, S represents the number of selected high-volatility users, represents the mean score of health information feedback of high-fluctuation users under the jth health condition, T s,j represents the feedback time interval of the sth high-fluctuation user in the jth health condition, V j represents the average offset of all high-fluctuation users under the jth health condition, D s,j represents the symptom intensity data of the sth high-fluctuation user in the jth health condition, D s,j-1 represents the symptom intensity data of the sth high-fluctuation user in the j-1th health condition, U j Represents the mean symptom change of all high-fluctuation users under the jth health condition.

10. A health care service intelligent question-answering system based on a large model, characterized by: The system is implemented according to the large model-based intelligent question-answering method for health care services according to any one of claims 1 to 9, wherein the system comprises: The health data extraction module obtains the health description information input by the user, extracts the symptom description, occurrence time, and duration period, filters the data of symptom influencing factors, calculates the range of health status changes, and obtains the health status change interval; The symptom trend analysis module extracts the health status time series data in the health status change interval, calculates the symptom intensity change rate, extracts the symptom fluctuation pattern, screens the affected symptom categories, evaluates the symptom development trend, and obtains the symptom change trend analysis results; The symptom association calculation module calculates the similarity between the symptoms and the course of the disease at multiple time points based on the symptom change trend analysis results, and calculates the matching weight based on the symptom co-occurrence frequency statistics to obtain the symptom matching correlation; The symptom screening optimization module calls the symptom matching correlation, compares symptom records, analyzes symptom coverage, determines missed symptoms based on the symptom co-occurrence probability library, screens symptom categories that meet health care needs, evaluates symptom matching credibility, and obtains a revised symptom candidate set; The health care guidance matching module modifies the candidate set of symptoms, matches health care records to screen health care intervention guidance, calls user health status data to compare adaptability, and obtains health care intelligent question and answer content.

Images