Constipation prediction system based on artificial intelligence large model

Abstract

The present invention discloses a constipation prediction system based on a large artificial intelligence model, which relates to the field of medical and health information technology. In order to solve the problem that it is difficult to fully capture the complex risk factors of constipation, resulting in limited feature coverage of the prediction model, no group classification of users and dynamic assessment of risk levels, resulting in the problem that the intervention plan is universal but difficult to meet the precise management needs of different groups; the present invention integrates multi-source data such as electronic medical records and physiological parameters through a multimodal data processing module to achieve refined classification of user groups, combines knowledge graph technology to analyze feature causal relationships, and improves the accuracy of constipation prediction. The intelligent prediction module dynamically adapts the model according to the group risk level, outputs probability distribution and generates a visual causal path report. The dynamic decision-making module is based on the knowledge graph matching strategy set, adjusts the intervention plan through doctor-patient collaboration, and uses the effect feedback mechanism to optimize the strategy in real time to achieve personalized health management.

Description

Technical Field

[0001] The present invention relates to the field of medical health information technology, and in particular to a constipation prediction system based on an artificial intelligence large model. Background Art

[0002] Traditional constipation assessment relies on single-dimensional data or doctor experience, and lacks integrated analysis and dynamic management of multimodal data, resulting in insufficient prediction accuracy and low personalization of intervention plans. For example, the patent application with publication number CN114081446A discloses a constipation prediction method, which uses a weight sensor to collect weight change data during the user's toileting process, and then classifies the data according to different defecation durations. The analysis parameters of constipation are calculated based on different classification results: weight stability is used to determine whether the user has a sense of tension in defecation, and then the weight stability is used as a parameter to calculate the stability of the defecation period, and then the status of the user's entire defecation process is analyzed to determine whether there are a large number of defecation difficulty nodes, and finally compared with historical data. Compare and analyze the defecation cycle data within a week to infer the changing trend of users' defecation.

[0003] Although the above patent can accurately predict and determine whether there is a possibility of constipation, the following problems still exist:

[0004] Relying solely on weight data from toileting scenarios without integrating the user's overall physiological characteristics makes it difficult to fully capture the complex risk factors of constipation, resulting in limited feature coverage of the prediction model.

[0005] Users are not classified into groups or dynamically assessed for risk levels. All user data are analyzed using a unified standard, making it impossible to perform differentiated modeling for specific risk factors of people with different physiological characteristics. This results in intervention plans that are highly universal but lack personalization, making it difficult to meet the precise management needs of different groups. Summary of the Invention

[0006] The purpose of the present invention is to provide a constipation prediction system based on a large artificial intelligence model, and provide corresponding prevention and treatment recommendations to patients or medical professionals based on the prediction results, helping medical professionals to more accurately identify constipation risks, thereby achieving early intervention and treatment, and realizing refined classification of user groups. Through collaborative adjustment of intervention plans between doctors and patients, the quality of life of patients can be improved, so as to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] Constipation prediction system based on artificial intelligence large model, including:

[0009] a multimodal data processing module configured to acquire electronic medical data related to constipation based on a multi-source data interface, preprocess the acquired electronic medical data, identify anomalies in the preprocessed electronic medical data, screen out valid medical data, and extract features from the valid medical data;

[0010] An intelligent prediction module is configured to perform vector processing based on the extraction results of valid medical data, generate a multimodal feature vector, and input it into the corresponding prediction model to predict constipation and generate a visual assessment report of the causal path of constipation risk;

[0011] The dynamic decision-making module is configured to match the corresponding strategy set from the strategy library based on the constipation prediction results and the constipation field knowledge graph, generate the user's constipation improvement plan, and evaluate the intervention effect of the constipation improvement plan.

[0012] Furthermore, before the multimodal data processing module obtains the electronic medical data, it also includes:

[0013] Establish a data transmission link with the smart terminal through the network communication protocol to collect the user's physiological parameters in real time, including the user's multi-dimensional physiological indicators and special data;

[0014] Build a user health archive, determine the user's physiological characteristics based on the collected physiological parameters, and classify user groups with similar physiological characteristics based on the cluster analysis model;

[0015] Combining users' real-time monitoring data with health status evolution trends, the user group classification results are adaptively adjusted and user groups are dynamically divided.

[0016] Furthermore, dynamic segmentation of user groups also includes:

[0017] Obtaining the user physiological characteristic data corresponding to each user group classification result, determining the physiological indicators of the user physiological characteristic data, calculating the coefficient of variation of each user's physiological characteristic, and constructing the corresponding physiological characteristic label;

[0018] Extract key physiological indicators from user physiological characteristic data, determine the weight of each physiological indicator, calculate the comprehensive risk index of each user group, conduct quantitative assessment of each user group, and determine the risk level of each user group;

[0019] Count the actual constipation incidence rates within each initial risk level range for different physiological characteristic labels, analyze the correlation between the corresponding user physiological characteristics and constipation risk, and identify key risk characteristics and risk contribution;

[0020] A dynamic threshold adjustment strategy is formulated based on the correlation between user physiological characteristics and constipation risk, and the risk level threshold of user groups with each physiological characteristic label is adjusted according to the dynamic threshold adjustment strategy.

[0021] Furthermore, the coefficient of variation of each user's physiological characteristics is calculated, including:

[0022] Retrieving the user's physiological characteristic data and historical risk identification records;

[0023] Determine whether the currently collected physiological characteristic data of the user is the first time the physiological characteristic data of the user is collected based on the historical risk identification record corresponding to the user;

[0024] When the user is collecting physiological characteristic data for the first time, the deviation rate of each physiological characteristic data from its corresponding physiological standard data range is extracted;

[0025] Obtaining an average value of the deviation rate according to the deviation rate of each physiological characteristic data from its corresponding physiological standard data range, and using the average value of the deviation rate as the coefficient of variation of the user's physiological characteristics;

[0026] If this is not the first time that the user is collecting physiological characteristic data, the weight of the historical risk identification record corresponding to each physiological characteristic data is retrieved;

[0027] Obtaining a weighted median corresponding to each physiological feature according to the user physiological feature data and its corresponding weight;

[0028] Comparing the weighted median corresponding to each physiological feature with a preset median threshold, and screening out the number of physiological features that exceed the preset median threshold;

[0029] When the number of physiological characteristics exceeding the preset median threshold does not exceed the preset reference number value, obtaining an average value of the deviation rate using all physiological characteristic data included in the currently collected user physiological characteristic data, and using the average value of the deviation rate as the coefficient of variation of the user's physiological characteristics;

[0030] When the number of physiological characteristics exceeding the preset median threshold exceeds a preset number reference value, the coefficient of variation of the user's physiological characteristic data is obtained using the weighted median corresponding to each physiological characteristic.

[0031] Furthermore, the coefficient of variation of the user's physiological characteristic data is obtained using the weighted median corresponding to each physiological characteristic, including:

[0032] Retrieve the weighted median corresponding to each physiological feature;

[0033] Retrieve the weight value of each physiological feature that appears in historical risk identification records;

[0034] Obtaining a standard deviation of the weight values ​​of each physiological characteristic appearing in the historical risk identification records using the weight values ​​of each physiological characteristic appearing in the historical risk identification records;

[0035] Obtaining the weight change impact intensity corresponding to each physiological feature using the standard deviation of the weight values ​​of each physiological feature appearing in the historical risk identification records;

[0036] The coefficient of variation corresponding to the current user's physiological characteristic data is obtained by combining the weight change influence strength corresponding to each physiological characteristic with the weighted median corresponding to each physiological characteristic;

[0037] Furthermore, the multimodal data processing module includes:

[0038] Data acquisition unit, configured as:

[0039] Obtain the patient's constipation-related medical history, medication records, and symptom description text through the electronic medical record system interface;

[0040] Clean and standardize the acquired electronic medical data, determine the data fluctuation range of each type of data based on the data type of the electronic medical data, and generate basic verification rules;

[0041] Performing data verification on the acquired electronic medical data based on basic verification rules, and marking the electronic medical data that does not meet the verification rules as abnormal data based on the verification results;

[0042] Submit the marked abnormal data to the manual review queue, remove the confirmed abnormal data based on the review results, and generate a medical data set that meets the quality standards as valid medical data;

[0043] The feature extraction unit is configured to extract features from valid medical data, construct a Bayesian network to analyze the causal relationship between features, and identify the combined effects between features.

[0044] Furthermore, the multimodal data processing module further includes:

[0045] The knowledge graph construction unit is configured as follows:

[0046] Extract structured and unstructured data from electronic medical records, perform semantic analysis on unstructured data, identify medical entities and entity categories, perform data mapping on structured data, and establish associations with medical entities;

[0047] Identify the semantic relationships between medical entities, and construct an entity relationship set based on the identification results. Then, determine the temporal relationship corresponding to the entities based on each sub-association relationship in the entity relationship set, and construct a knowledge graph in the field of constipation.

[0048] Based on the constipation domain knowledge graph, the temporal relationships corresponding to the entities are analyzed, the triggering factors in the temporal relationship chain of the entities are identified, a subset of the constipation domain knowledge graph is generated, and the correlation coefficient between the user's physiological characteristics and the constipation risk is calculated.

[0049] Furthermore, the correlation coefficient between the user's physiological characteristics and the risk of constipation is calculated, including:

[0050] Extract the knowledge graph in the field of constipation;

[0051] Retrieving a standardized physiological characteristic value corresponding to each physiological characteristic of the user;

[0052] Retrieve the user's corresponding time series risk score; wherein, the time series risk score is obtained by accumulating the time series relationship chain of triggering factors in the knowledge graph;

[0053] Retrieve the number of risk triggers for the user and the average number of triggers for all users;

[0054] performing a ratio process on the number of risk triggering factors of the user and the average number of triggering factors of all users to obtain the ratio between the number of risk triggering factors of the user and the average number of triggering factors of all users;

[0055] The ratio between the number of risk trigger factors of the user and the average number of trigger factors of all users is identified as the time series chain confidence weight corresponding to the user;

[0056] The correlation coefficient between the user's physiological characteristics and the constipation risk is obtained by using the standardized physiological characteristic value, the temporal risk score and the temporal chain confidence weight corresponding to each user's physiological characteristic.

[0057] The correlation coefficient between the user's physiological characteristics and the risk of constipation is obtained by the following formula:

[0058] ;

[0059] Among them, C i represents the correlation coefficient between the physiological characteristics of the i-th user and the risk of constipation; U represents the total number of users; ω u represents the confidence weight of the time series chain corresponding to the u-th user; s ui represents the normalized physiological feature value corresponding to the i-th physiological feature of the u-th user; r u represents the temporal risk score corresponding to the u-th user. u represents the subscript number;

[0060] Furthermore, the intelligent prediction module includes:

[0061] a model determination unit configured to match corresponding constipation prediction models for user groups of different risk levels based on the dynamic classification results of user groups, and to adjust the matched constipation prediction model according to the latest user group classification level when the comprehensive risk index of the user group exceeds a preset fluctuation threshold;

[0062] A constipation prediction unit is configured to input the feature extraction results of the valid data into the corresponding constipation prediction model to perform constipation prediction and output the probability distribution data of constipation occurring in the user in the next few days;

[0063] The risk assessment unit is configured to adjust the risk assessment threshold based on the constipation field knowledge graph, and perform risk assessment based on the user's constipation prediction results. At the same time, a constipation risk causal path diagram is generated based on a subset of the constipation field knowledge graph.

[0064] Furthermore, the dynamic decision module includes:

[0065] a strategy matching unit configured to match a strategy set corresponding to the risk level of the user's constipation prediction result and the physiological feature label according to the constructed constipation domain knowledge graph, and to prioritize the strategies in the strategy set according to the risk level of the user's constipation prediction result;

[0066] The doctor-patient collaborative decision-making unit is configured to obtain and parse the doctor-patient interaction text, identify the doctor's adjustment needs and recommended parameters, obtain the patient's feedback needs, and adjust the strategy set to generate the user's constipation improvement plan;

[0067] The intervention effect feedback unit is configured to continuously collect physiological data of the user after the user implements the constipation improvement plan, evaluate the intervention effect of the constipation improvement plan, and adjust the unimplemented strategies in the constipation improvement plan according to the intervention effect.

[0068] Furthermore, the strategy matching unit further includes:

[0069] Based on the results of dynamic segmentation of user groups, the typical feature-strategy association rules of each user group are extracted from the constipation knowledge graph to generate corresponding strategy templates.

[0070] Establish a group strategy effect comparison matrix, analyze the intervention effect response differences of different user groups based on the feedback data obtained by the intervention effect feedback unit, and adjust the weights of each strategy combination in the strategy set;

[0071] Based on the physiological characteristic labels of the user group and the risk level of the user's constipation prediction results, a prevention strategy plan is generated, and timely prevention recommendations are pushed to the corresponding user group in combination with environmental characteristics.

[0072] Compared with the prior art, the present invention has the following beneficial effects:

[0073] The multimodal data processing module integrates multi-source data such as electronic medical records and physiological parameters to achieve refined classification of user groups. The knowledge graph technology is combined to analyze the causal relationship of features and improve the accuracy of constipation prediction. The intelligent prediction module dynamically adapts the model according to the group risk level, outputs the probability distribution and generates a visual causal path report to enhance the interpretability of medical decision-making. The dynamic decision-making module is based on the knowledge graph matching strategy set, adjusts the intervention plan through doctor-patient collaboration, and uses the effect feedback mechanism to optimize the strategy in real time to form a closed-loop management. It pushes timely prevention suggestions based on environmental characteristics to achieve personalized health management. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 This is a module diagram of the constipation prediction system based on the artificial intelligence large model of the present invention. DETAILED DESCRIPTION

[0075] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0076] To address the technical issues of difficulty in fully capturing the complex risk factors of constipation, resulting in limited feature coverage of the prediction model, failure to classify users into groups and dynamically assess risk levels, and the resulting intervention plan being universal but unable to meet the precise management needs of different groups, please refer to Figure 1 , this embodiment provides the following technical solutions:

[0077] Constipation prediction system based on artificial intelligence large model, including:

[0078] a multimodal data processing module configured to acquire electronic medical data related to constipation based on a multi-source data interface, including medical history, medication records, symptom description text, etc., pre-process the acquired electronic medical data, identify anomalies in the pre-processed electronic medical data, filter out valid medical data, and perform feature extraction on the valid medical data;

[0079] The intelligent prediction module is configured to perform vectorization processing based on the results of effective medical data extraction, generate multimodal feature vectors containing text semantic features, temporal physiological indicator features, and structured medical indicator features, and input them into the corresponding prediction model to predict constipation, thereby accurately predicting the user's constipation risk and generating a visual assessment report of the causal path of constipation risk, including:

[0080] The model determination unit is configured to match corresponding constipation prediction models for user groups of different risk levels based on the results of dynamic user group segmentation. When the comprehensive risk index of a user group exceeds a preset fluctuation threshold, the matched constipation prediction model is adjusted according to the latest user group segmentation level. For example, a deep neural network combined with a long short-term memory network (LSTM) is used for high-risk groups to capture the temporal changes of complex physiological indicators; a lightweight random forest model is used for low-risk groups to balance prediction accuracy and computational efficiency.

[0081] A constipation prediction unit is configured to input the feature extraction results of the valid data into the corresponding constipation prediction model to perform constipation prediction and output the probability distribution data of constipation occurring in the user in the next few days;

[0082] a risk assessment unit configured to adjust a risk assessment threshold based on a constipation domain knowledge graph, perform risk assessment based on a user's constipation prediction result, and generate a constipation risk causal path diagram based on a subset of the constipation domain knowledge graph;

[0083] The dynamic decision-making module is configured to match the corresponding strategy set from the strategy library based on the constipation prediction results and the constipation field knowledge graph, generate the user's constipation improvement plan, and evaluate the intervention effect of the constipation improvement plan.

[0084] In this embodiment, by integrating multi-source data such as electronic medical records and physiological parameters, combined with cluster analysis and dynamic threshold adjustment strategies, refined classification of user groups and quantitative assessment of risk levels are achieved. Knowledge graph technology effectively identifies key risk factors, enabling the constipation prediction model to dynamically adapt to groups with different risk levels, improving the accuracy of predicted probability distribution, and converting complex medical associations into intuitive graphs to assist doctors and patients in understanding risk factors and mechanisms of action. It can also explore potential triggering factors through temporal relationship analysis, promote medical knowledge discovery and clinical decision support in the field of constipation, build a strategy library based on the knowledge graph, realize automatic matching of prediction results and intervention plans, and dynamically adjust strategy combinations through collaborative interaction between doctors and patients. The intervention effect feedback mechanism supports real-time optimization of plans, and pushes timely prevention suggestions based on environmental characteristics to improve the efficiency and personalization of constipation prevention and improvement.

[0085] In this embodiment, before the multimodal data processing module obtains the electronic medical data, it also includes:

[0086] Through network communication protocols, data transmission links are established with smart terminals such as smart wearable devices and smart sanitary ware to collect users' physiological parameters in real time, including multi-dimensional physiological indicators such as the user's heart rate, blood pressure, body movement frequency, intestinal electrical signals, as well as special data such as defecation time and stool characteristics detected by smart toilets, ensuring comprehensive and real-time data collection;

[0087] Build a user health archive, determine the user's physiological characteristics based on the collected physiological parameters, and classify user groups with similar physiological characteristics based on the cluster analysis model;

[0088] Combining real-time user monitoring data with health status evolution trends, adaptively adjust user group classification results and dynamically divide user groups;

[0089] Obtaining the user physiological characteristic data corresponding to each user group classification result, determining the physiological indicators of the user physiological characteristic data, calculating the coefficient of variation of each user's physiological characteristic, and constructing the corresponding physiological characteristic label;

[0090] In this embodiment, physiological indicators include continuous physiological indicators and discrete physiological indicators. Continuous physiological indicators (such as heart rate and intestinal electrical signal amplitude) are divided into five intervals and assigned labels of "very low", "low", "medium", "high", and "very high" respectively. For discrete physiological indicators (such as stool trait categories), the category with the highest frequency in the group is used as the representative label. The stability of the physiological characteristics within the group is evaluated by calculating the coefficient of variation of the characteristics. If the coefficient of variation is less than a threshold (such as 0.2), the "stable" label is added; otherwise, the "fluctuating" label is added.

[0091] Extract key physiological indicators from user physiological characteristic data, determine the weight of each physiological indicator, calculate the comprehensive risk index of each user group, conduct quantitative assessment of each user group, and determine the risk level of each user group;

[0092] Count the actual constipation incidence rates within each initial risk level range for different physiological characteristic labels, analyze the correlation between the corresponding user physiological characteristics and constipation risk, and identify key risk characteristics and risk contribution;

[0093] Develop a dynamic threshold adjustment strategy based on the correlation between user physiological characteristics and constipation risk, and adjust the risk level threshold for user groups with different physiological characteristic tags according to the dynamic threshold adjustment strategy;

[0094] In this embodiment, if the actual constipation incidence rate of a certain physiological characteristic label group in the initial low-risk interval exceeds a preset threshold (such as 30%), the low-risk threshold corresponding to the group will be adjusted downward by a certain proportion (such as 10%); if the actual constipation incidence rate in the high-risk interval is lower than the preset standard (such as 50%), the high-risk threshold will be adjusted upward (such as 15%).

[0095] In this embodiment, users are classified according to physiological characteristics through a cluster analysis model, and group divisions are dynamically adjusted in combination with real-time data. At the same time, physiological data are converted into feature labels through methods such as coefficient of variation to achieve personalized risk stratification management; key physiological indicators are extracted to determine weights to calculate a comprehensive risk index for quantitative evaluation, and the risk level threshold is dynamically adjusted based on the actual incidence of constipation to improve the timeliness and reliability of the prediction model; the correlation between physiological characteristics and the incidence of constipation is explored to provide a basis for intelligent prediction and dynamic decision-making modules, realizing full-chain innovation-driven development from data collection to precise intervention, which is different from traditional constipation assessment and intervention methods.

[0096] Specifically, the coefficient of variation of each user's physiological characteristics is calculated, including:

[0097] Retrieving the user's physiological characteristic data and historical risk identification records;

[0098] Determine whether the currently collected physiological characteristic data of the user is the first time the physiological characteristic data of the user is collected based on the historical risk identification record corresponding to the user;

[0099] When the user is collecting physiological characteristic data for the first time, the deviation rate of each physiological characteristic data from its corresponding physiological standard data range is extracted;

[0100] Obtaining an average value of the deviation rate according to the deviation rate of each physiological characteristic data from its corresponding physiological standard data range, and using the average value of the deviation rate as the coefficient of variation of the user's physiological characteristics;

[0101] If this is not the first time that the user is collecting physiological characteristic data, the weight of the historical risk identification record corresponding to each physiological characteristic data is retrieved;

[0102] Obtaining a weighted median corresponding to each physiological feature according to the user physiological feature data and its corresponding weight;

[0103] Comparing the weighted median corresponding to each physiological feature with a preset median threshold, and screening out the number of physiological features that exceed the preset median threshold;

[0104] When the number of physiological characteristics exceeding the preset median threshold does not exceed the preset reference number value, obtaining an average value of the deviation rate using all physiological characteristic data included in the currently collected user physiological characteristic data, and using the average value of the deviation rate as the coefficient of variation of the user's physiological characteristics;

[0105] When the number of physiological characteristics exceeding the preset median threshold exceeds a preset number reference value, the coefficient of variation of the user's physiological characteristic data is obtained using the weighted median corresponding to each physiological characteristic.

[0106] The technical effect of the above technical solution is: first, retrieve the user's physiological characteristic data and historical risk identification records to determine whether the currently collected physiological characteristic data is the first time collected by the user. If it is the first time collected, calculate the deviation rate of each physiological characteristic data from the corresponding physiological standard data range, and then calculate the average value of the deviation rate, and use it as the coefficient of variation of the user's physiological characteristics. If it is not the first time collected, retrieve the weight of the historical risk identification record corresponding to each physiological characteristic data, and calculate the weighted median corresponding to each physiological characteristic based on this, compare it with the preset median threshold, and count the number of physiological characteristics that exceed the threshold. When the number exceeding the threshold does not exceed the preset number reference value, calculate the average value of the deviation rate of all currently collected physiological characteristic data as the coefficient of variation; if it exceeds, calculate the coefficient of variation of the user's physiological characteristic data based on the weighted median corresponding to each physiological characteristic.

[0107] Different calculation strategies are employed based on whether the user is collecting physiological characteristic data for the first time. For initial data collection, the average deviation rate is calculated using the standard data range as a reference to quickly establish a baseline variation index for the user's physiological characteristics. For non-initial data collection, the coefficient of variation is calculated by incorporating historical risk record weights and taking into account the user's individual history. This ensures that the coefficient of variation calculation more closely reflects the user's actual physiological state changes, improving data accuracy and applicability. For non-initial data collection, the weighted median is calculated by incorporating historical risk identification record weights to highlight the impact of significant historical risk data on the current assessment. Compared to simple mean calculation, the weighted median more robustly reflects data trends, effectively eliminates outlier interference, and accurately assesses the degree of change in the user's physiological characteristics, providing a reliable basis for risk assessment. A median threshold and a reference number are set to identify and filter physiological characteristic data exceeding the median threshold. When the number of data exceeding the threshold is small, the coefficient of variation is still calculated using the average deviation rate of the current data to ensure calculation stability. When the number of data exceeding the threshold is large, the weighted median calculation is used instead, enhancing the ability to capture abnormal physiological characteristic changes, dynamically adapting to different physiological characteristic fluctuations, and improving risk identification sensitivity. The precisely calculated coefficient of variation can more accurately quantify the degree of abnormality and changing trends of the user's physiological characteristics. Whether it is the first time to establish a user's physiological data model or subsequent continuous monitoring, this calculation method can effectively identify potential risks and avoid misjudgment or omission of risks due to unreasonable data calculations. It provides reliable data support for applications such as health monitoring and disease early warning, and improves the reliability and effectiveness of system risk identification. The coefficient of variation is calculated based on the user's own historical data and the standard data range, fully considering the uniqueness and dynamic changes of individual physiological characteristics. Different users have different physiological baselines. This method can perform personalized calculations for each user, adapt to the physiological differences between individuals, provide users with more targeted health assessments and risk warnings, and enhance the applicability of the system to diverse user groups.

[0108] Specifically, the coefficient of variation of the user's physiological characteristic data is obtained using the weighted median corresponding to each physiological characteristic, including:

[0109] Retrieve the weighted median corresponding to each physiological feature;

[0110] Retrieve the weight value of each physiological feature that appears in historical risk identification records;

[0111] Obtaining a standard deviation of the weight values ​​of each physiological characteristic appearing in the historical risk identification records using the weight values ​​of each physiological characteristic appearing in the historical risk identification records;

[0112] Obtaining the weight change impact intensity corresponding to each physiological feature using the standard deviation of the weight values ​​of each physiological feature appearing in the historical risk identification records;

[0113] The influence strength of the weight change is obtained by the following formula:

[0114] ;

[0115] Among them, G represents the impact strength of weight change; n represents the number of times each physiological feature appears in the historical risk identification record; w i+1 and w i Respectively represent the weight values ​​corresponding to the i+1th and ith occurrences in the historical risk identification records; w z represents the corresponding weight median value that appears n times in the historical risk identification record; w σ Indicates the standard deviation of the corresponding weight value in the historical risk identification records for n times; specifically, Calculating the absolute difference between the weight values ​​in two adjacent historical risk identification records is used to measure the fluctuation range of the weight at adjacent moments, reflecting the dynamic differences in weight changes, and is the basis for capturing the changing characteristics of the weight sequence. Multiply the current weight value by the fluctuation range of the adjacent weights, so that the large weight value and large fluctuation contribute more significantly to the overall product, highlighting the impact of changes under high weights. Introducing the intermediate weight value w z, combined with the weight at the current moment, construct an adjustment term related to the central trend of the weight sequence, and balance the relationship between the "current value" and the "historical center" in the weight change. By multiplying the numerator, from the perspective of weight stability and central trend, a constraint and calibration of the cumulative fluctuation characteristics of the numerator is formed. Raise the quotient of the multiplication results of the numerator and denominator to the nth power, and convert the cumulative product result into an averaged index related to the number of records n, eliminating the influence of the number of records n on the magnitude of the result, so that the influence intensity of the physiological characteristic weight change of different recording times can be compared. Existing technologies mostly use a single statistic (such as mean, variance) to measure weight stability. This formula can dynamically capture the continuous change trend of the weight sequence by multiplying the product of adjacent weight differences with the current weight, and more accurately reflect the dynamic characteristics of "continuous fluctuation" or "mutation" of the weight, avoiding the loss of sequence change details by a single statistic. At the same time, the median value and standard deviation of the weight are introduced, and constraints related to the central trend and dispersion degree are constructed in the denominator, so that the impact of weight change not only focuses on the fluctuation amplitude, but also relates to the overall distribution of the weight sequence (such as the central position and the dispersion range). Compared with the method of using only variance or mean, the description of weight change is more comprehensive. The design allows moments with high weights and large fluctuations to contribute more to the multiplication, which meets the requirement of "high weights correspond to higher risk attention, and their changes need to be considered in detail" in practical applications, and improves the ability to identify changes in key weights. By taking the cumulative product to the nth power and converting it into an average indicator, the impact of weight changes of different physiological characteristics (with different number of historical records n) can be directly compared, solving the problem of incomparable results caused by differences in the number of records, and enhancing the versatility of the algorithm in multi-feature and multi-user scenarios. The introduction of the weight standard deviation and median value in the denominator can, to a certain extent, suppress the excessive impact of abnormal weight values ​​on the results. For example, if the adjacent differences caused by a single abnormally high weight value are too large, the constraints of the standard deviation and median value in the denominator (for example, when the standard deviation increases and the median value stabilizes, the denominator as a whole increases) will make the final result more robust and more resistant to interference than simple multiplication or difference statistics.

[0116] The coefficient of variation corresponding to the current user's physiological characteristic data is obtained by combining the weight change influence strength corresponding to each physiological characteristic with the weighted median corresponding to each physiological characteristic;

[0117] The coefficient of variation is obtained by the following formula:

[0118] ;

[0119] Where H represents the coefficient of variation; m represents the number of physiological characteristics; x i Indicates the data value corresponding to the i-th physiological characteristic data in the current user's physiological characteristic data; u i represents the weighted median corresponding to the i-th physiological characteristic data in the current user's physiological characteristic data; Gi represents the influence strength of the weight change corresponding to the i-th physiological characteristic data in the current user's physiological characteristic data; x maxi and x mini They represent the maximum and minimum values ​​of the data in the historical risk identification record corresponding to the i-th physiological characteristic data; u z Indicates the normalized median average value of n physiological characteristic data in the current user's physiological characteristic data. Specifically, Calculate the current physiological characteristic data value x i The influence of weight change on intensity G i After weighting, the weighted median u of the feature i Absolute deviation. This reflects the degree to which the current data (after taking into account the influence of weights) deviates from the historical typical value (weighted median) and is a core indicator for capturing the variation of individual physiological characteristics. The above deviations are normalized using the inverse of the exponential of the extreme value difference of historical data as the denominator. Features with large historical extreme value differences (such as blood pressure that may fluctuate in a large range) have small inverse exponentials, and the deviations are "magnified"; features with small extreme value differences (such as relatively stable body temperature) have large inverse exponentials, and the deviations are "reduced". This allows the variation deviations of different physiological characteristics to be compared on a unified scale, adapting to the differences in the fluctuation ranges of the physiological characteristics themselves. Then, the above normalized deviations of all physiological characteristics are summed up, and the variation information of multiple features is integrated to reflect the degree of variation of the user's overall physiological characteristics. At the same time, The normalized median average u is introduced in z , further calibrate the averaged results. z Reflect the overall central trend of the weighted median of multiple features, convert the variation deviation of multiple features into a coefficient of variation that is compatible with the number of features and the overall median distribution, and quantify the overall variation level of the user's physiological characteristics. Calibrate the number of features and the overall median distribution to avoid significant distortion of the coefficient of variation due to abnormal fluctuations in individual features (such as a single erroneous measurement). When the deviation of a feature is abnormally large, the deviation of the feature in the numerator may be suppressed (if its historical fluctuation is already large) due to the exponential normalization of the historical extreme value difference, or the denominator may be averaged and calibrated to further balance and improve the robustness of the results. Existing technologies often analyze the variation of each physiological feature separately, or simply perform weighted summation, ignoring the differences in the fluctuation range of the feature itself. This formula is passed Normalize the deviations of different features so that the variation contributions of features with different fluctuation ranges, such as blood pressure and heart rate, can be fairly integrated, improving the accuracy of variation assessment in multi-feature scenarios. Combine the current data, historical weighted median, and the impact of weight changes to dynamically calculate the weight of historical risk records (G i Reflection) and current physiological data, historical typical values ​​(u i) fusion, compared with using only current data or a single historical statistic, can more comprehensively capture the "historical-current" variation correlation of physiological characteristics and improve the foresight of risk identification. The normal fluctuation range of physiological characteristics (such as body temperature and blood sugar) varies significantly, and the exponential treatment of historical extreme value differences in the formula automatically adapts to this heterogeneity. For example, if the extreme difference in blood sugar is large (such as in diabetic patients), the contribution of its deviation to the coefficient of variation will be more prominent; if the extreme difference in body temperature is small, the contribution will be more moderate, making the coefficient of variation more in line with the actual characteristics of the physiological characteristics. In the denominator By calibrating the number of features and the overall median distribution, we can avoid significant distortion of the coefficient of variation due to abnormal fluctuations in individual features (such as a single erroneous measurement). When the deviation of a feature is abnormally large, the deviation of the feature in the numerator may be suppressed by exponential normalization of the historical extreme value difference (if its historical fluctuation is already large), or the denominator may be averaged and calibrated to further balance and improve the robustness of the results. In medical and health scenarios, subtle variations in user physiological characteristics can be more accurately identified (such as trend changes in long-term data of chronic patients). Influencing the intensity G through weight changes i By associating historical risk records, the coefficient of variation can reflect "physiological abnormalities under dynamic changes in risk weights". Compared with the traditional coefficient of variation (which only focuses on the degree of data dispersion), it has higher guiding value for disease warning and health management.

[0120] The technical effect of the above technical solution is to obtain the weighted median corresponding to each physiological feature (reflecting the trend of historical data concentration) and the weight value in the historical risk identification record (reflecting the importance of historical data). σ ) measures the fluctuation range of weights, combined with the historical weight value sequence (w i ) and the intermediate value (w z ), use the formula to calculate the weight change impact intensity (G), and quantify the impact of historical weight fluctuations on the current evaluation. i ) and weighted median (u i ) multiplied by the weight change effect strength (G i ), and combined with the extreme values ​​of historical data and the normalized median average (u z ), the coefficient of variation (H) is output through the formula, which reflects the overall variation degree of physiological characteristics.

[0121] The influence strength (G) is calculated by combining the standard deviation of weight values ​​and sequence variation. This allows identification of physiological characteristics with unusual weight fluctuations in historical risk records. For example, if a characteristic's weight has recently fluctuated frequently (with a large standard deviation), its G value will increase, indicating a decrease in the credibility of the characteristic's historical data. This automatically reduces its influence weight in the coefficient of variation calculation to prevent outdated or anomalous historical data from interfering with the current assessment. The weighted median is insensitive to extreme values ​​and better represents the typical state of a user's physiological characteristics than the mean. Combining the weight change strength (G) allows dynamic adjustment of each characteristic's contribution to the coefficient of variation. For example, a higher G value can be assigned to characteristics with stable weights (small standard deviation), giving them a greater impact on the coefficient of variation and improving the robustness of the assessment results. The formula integrates the deviation of current data from the historical median, the range of historical data fluctuations, and the dynamic characteristics of the weights to comprehensively quantify the coefficient of variation from multiple dimensions: current deviation and historical fluctuation range. Normalizing the median to the mean eliminates dimensional differences between characteristics, enabling horizontal comparison of the variations of different physiological indicators. Combined with the strength of weight changes, it can identify user-specific physiological trends (e.g., a sustained increase in a feature's weight accompanied by an increase in the coefficient of variation), providing early warning of potential health risks and providing a more proactive approach than using a single indicator threshold. The weight change impact strength (G) automatically suppresses noise interference in historical data. For example, if a feature's historical weight fluctuates abnormally due to measurement error (e.g., occasional high-weight outliers), its standard deviation increases and its G value decreases, thereby reducing the influence of this feature in the coefficient of variation calculation. This ensures that the assessment results are based on reliable historical data and improves the accuracy of risk identification. For users who are not collecting data for the first time, dynamic analysis of historical weights allows the coefficient of variation to adapt to long-term changes in the baseline of individual physiological characteristics (e.g., an upward shift in the normal blood pressure range due to aging). Compared to fixed standard value assessments, this method avoids misjudgments caused by individual baseline drift and better reflects the evolution of the user's true physiological state.

[0122] In this embodiment, the multimodal data processing module includes:

[0123] Data acquisition unit, configured as:

[0124] Obtain the patient's constipation history, medication records, and symptom description text through the electronic medical record system interface, including information such as bowel movement frequency, stool characteristics, and accompanying symptoms;

[0125] The acquired electronic medical data is cleaned and standardized, and the data fluctuation range of each type of data is determined based on the data type of the electronic medical data. Basic verification rules are generated, such as bowel movement frequency >30 times / day or <0 times / month, and medication dosage exceeding the pharmacopoeial standard range are considered abnormal data;

[0126] Performing data verification on the acquired electronic medical data based on basic verification rules, and marking the electronic medical data that does not meet the verification rules as abnormal data based on the verification results;

[0127] Submit the marked abnormal data to the manual review queue, remove the confirmed abnormal data based on the review results, and generate a medical data set that meets the quality standards as valid medical data;

[0128] A feature extraction unit is configured to extract features from valid medical data using a time-frequency analysis method, construct a Bayesian network to analyze the causal relationship between features, identify the combined effects between features, and identify potential causal paths such as "abnormal intestinal electrical signals → prolonged bowel movement intervals";

[0129] In this embodiment, comprehensive multi-source information related to patient constipation is obtained through the electronic medical record system interface. Basic verification rules are formulated according to the data type, abnormal data is accurately identified and eliminated, and a high-quality and valid medical data set is generated. This lays a solid and reliable data foundation for subsequent analysis, avoids analytical bias caused by data impurities, and deeply mines the characteristics of valid medical data to clearly identify potential causal paths. This can provide in-depth insights into the intrinsic correlations between constipation-related factors, provide a key basis for constipation prediction, and help to more accurately understand and predict the mechanism of constipation.

[0130] The knowledge graph construction unit is configured as follows:

[0131] Extract structured and unstructured data from electronic medical records, perform semantic analysis on unstructured data (such as medical records and doctor's orders), identify medical entities and entity categories, including disease names, symptoms, drug names, and treatment methods, perform data mapping on structured data (such as examination reports and test results), and establish associations with medical entities;

[0132] Identify the semantic relationships between medical entities, such as "medication-treatment-disease" and "symptoms-complications-disease", and construct an entity relationship set based on the recognition results. Based on the sub-associations in the entity relationship set, determine the corresponding temporal relationships of the entities, such as the timeline of "symptom onset-examination implementation-treatment intervention", to build a knowledge graph in the field of constipation;

[0133] Based on the constipation domain knowledge graph, we analyze the temporal relationships corresponding to entities, identify the triggering factors in the temporal relationship chain of entities, such as symptom onset → examination, generate a subset of the constipation domain knowledge graph, and calculate the correlation coefficient between the user's physiological characteristics and constipation risk;

[0134] Specifically, the correlation coefficient between the user's physiological characteristics and the risk of constipation is calculated, including:

[0135] Extract the knowledge graph in the field of constipation;

[0136] Retrieving a standardized physiological characteristic value corresponding to each physiological characteristic of the user;

[0137] Retrieve the user's corresponding time series risk score; wherein, the time series risk score is obtained by accumulating the time series relationship chain of triggering factors in the knowledge graph;

[0138] Retrieve the number of risk triggers for the user and the average number of triggers for all users;

[0139] performing a ratio process on the number of risk triggering factors of the user and the average number of triggering factors of all users to obtain the ratio between the number of risk triggering factors of the user and the average number of triggering factors of all users;

[0140] The ratio between the number of risk trigger factors of the user and the average number of trigger factors of all users is identified as the time series chain confidence weight corresponding to the user;

[0141] The correlation coefficient between the user's physiological characteristics and the constipation risk is obtained by using the standardized physiological characteristic value, the temporal risk score and the temporal chain confidence weight corresponding to each user's physiological characteristic.

[0142] The correlation coefficient between the user's physiological characteristics and the risk of constipation is obtained by the following formula:

[0143] ;

[0144] Where C represents the correlation coefficient between the physiological characteristics of the i-th user and the risk of constipation; U represents the total number of users; ω u represents the confidence weight of the time series chain corresponding to the u-th user; s ui represents the normalized physiological feature value corresponding to the i-th physiological feature of the u-th user; r u represents the temporal risk score corresponding to the u-th user. Specifically, The confidence level of the temporal relationship of a user's risk triggers is determined by the ratio of the number of risk triggers to the average number. The higher the weight, the greater the reference value of the triggering factor temporal relationship for risk analysis. The dimensional differences of different physiological characteristics are eliminated so that each characteristic value can be calculated on a unified scale, representing the standardized level of the i-th physiological characteristic of the u-th user. Based on the accumulation of the temporal relationship chain of trigger factors in the knowledge graph, the integral value of the user's constipation risk caused by temporal trigger factors is quantified. The "temporal confidence weight-standardized physiological characteristics-temporal risk score" association information of all users is integrated to measure the degree of positive correlation between the i-th physiological characteristic and constipation risk through the temporal risk path. The square root of the sum of "square of time series confidence weight - square of standardized physiological characteristics" is actually a quantification of the degree of dispersion of the standardized physiological characteristic value under the weighting of the time series confidence weight (similar to the weighted standard deviation), reflecting the weighted fluctuation level of the i-th physiological characteristic value. Taking the square root of the sum of "square of the time series confidence weight - square of the time series risk integral" quantifies the degree of dispersion of the time series risk integral under the weighting of the time series confidence weight, reflecting the weighted volatility of the time series risk integral. Existing technologies often rely on a single data dimension (such as physiological characteristic statistics) to analyze disease correlation. This method introduces a knowledge graph in the field of constipation. By associating the time series risk integral with the triggering factors, domain knowledge is integrated into the correlation calculation, making the correlation analysis between physiological characteristics and constipation risk more consistent with the disease pathogenesis mechanism and improving the scientific nature of the correlation judgment. The confidence weight of the time series chain is determined by the ratio of the number of risk triggering factors, distinguishing the reliability of the triggering factor time series of different users. For example, users whose number of triggering factors differs greatly from the average number (such as patients with complete long-term records) have a higher weight and contribute more to the correlation calculation, so that the results are more focused on high-confidence user data and enhance the targeted analysis. The integration of standardized physiological characteristics, temporal risk scores, and temporal confidence weights considers both the numerical characteristics of the physiological characteristics themselves and the temporal relationships and confidence levels of risk triggers. Compared to traditional methods that simply associate physiological characteristics with disease labels, this method can more comprehensively capture the complex "physiology-risk-temporal" relationship and improve the accuracy of the correlation coefficient. The weighted discrete degree product of the denominator gives the correlation coefficient standardized mathematical properties. The correlation between different physiological characteristics (such as heart rate and dietary fiber intake) and constipation risk can be directly compared, solving the problem of incomparable correlation results under multiple feature dimensions and facilitating clinical screening of key risk characteristics. The temporal chain confidence weight weights user data to reduce the impact of noise data from low-confidence users (such as those with incomplete records of trigger factors and chaotic temporal sequences).

[0145] The technical effect of the above technical solution is as follows: first, the knowledge graph in the field of constipation is extracted, and then the standardized physiological characteristic value of each physiological characteristic of the user, the corresponding temporal risk score (accumulated from the temporal relationship chain of trigger factors in the knowledge graph), and the number of risk trigger factors of the user and the average number of trigger factors of all users are retrieved, and the temporal chain confidence weight is obtained by comparing the two. Finally, the standardized physiological characteristic value, temporal risk score, and temporal chain confidence weight are substituted into the formula to calculate the correlation coefficient between the user's physiological characteristics and constipation risk. The formula integrates the "temporal confidence weight-standardized physiological characteristic-temporal risk score" correlation information in the numerator, and normalizes the weighted discrete degree of the physiological characteristics and risk score in the denominator to obtain the correlation coefficient. Introducing the knowledge graph in the field of constipation and incorporating the knowledge of the temporal relationship of disease triggering into the calculation allows the correlation between physiological characteristics and constipation risk to fit the disease occurrence mechanism, breaking through the limitations of traditional single data dimension analysis and improving the scientific nature and disease adaptability of correlation judgment. The temporal chain confidence weighting assigns different weights to different user data based on the number of risk triggers. Data from high-confidence users (e.g., those with complete trigger records) contributes more significantly to the calculation, effectively reducing interference from low-quality data, focusing the results on reliable information, and enhancing the targetedness and robustness of the analysis. The method integrates standardized physiological characteristics, temporal risk scores, and temporal confidence weights to comprehensively cover physiological values, risk-time series relationships, and data confidence dimensions. Compared with traditional simple correlation methods, it more accurately captures the complex "physiology-risk-time series" relationship and improves the accuracy of the correlation coefficient in identifying constipation risk characteristics. Denominator normalization standardizes the correlation coefficient, allowing direct comparison of the correlation between different physiological characteristics and constipation risk, facilitating rapid clinical screening of key risk characteristics and optimizing feature priority in diagnosis and treatment decisions. The weighting mechanism and normalization process enable the method to robustly calculate the correlation coefficient even in real-world scenarios with varying user data quality, effectively combating noise and ensuring reliable application in real-world medical data environments, facilitating constipation risk assessment and intervention.

[0146] In this embodiment, unstructured data such as clinical guideline texts are used in conjunction with a named entity recognition and relationship extraction model to construct a word vector representation enhanced by a medical dictionary. Character-level, word-level, and sentence-level features are integrated to identify entities such as diseases, symptoms, drugs, and treatments, construct entity sets, and ensure entity consistency. For example, "hard stool" can be mapped to "abnormal stool characteristics."

[0147] In this embodiment, the knowledge graph construction unit performs deep semantic analysis and association construction on the structured and unstructured data in the electronic medical records, accurately identifies medical entities and categories, constructs a knowledge graph in the field of constipation and a subset of the knowledge graph, realizes the systematic integration and in-depth application of knowledge in the field of constipation, provides rich knowledge support for intelligent prediction and decision-making, and improves the scientificity and accuracy of the system's constipation risk assessment and intervention strategy formulation.

[0148] In this embodiment, the dynamic decision module includes:

[0149] The strategy matching unit is configured to match the strategy set corresponding to the risk level of the user's constipation prediction result and physiological feature label based on the constructed constipation knowledge graph. For example, for high-risk users with "dry stool characteristics + fluctuating intestinal electrical signals", the strategy related to "prokinetic drugs + intestinal microecological regulation" is preferentially matched, and the strategies in the strategy set are prioritized according to the risk level of the user's constipation prediction result;

[0150] Based on the results of dynamic user group segmentation, the typical feature-strategy association rules of each user group are extracted from the constipation knowledge graph to generate corresponding strategy templates. For example, for the "diabetes and constipation" group, dietary fiber types that do not affect blood sugar (such as oligofructose) are preferentially recommended.

[0151] Establish a group strategy effect comparison matrix, analyze the intervention effect response differences of different user groups based on the feedback data obtained by the intervention effect feedback unit, and adjust the weights of each strategy combination in the strategy set;

[0152] Based on the physiological characteristics of the user group and the risk level of the user's constipation prediction results, a prevention strategy is generated to reduce the probability of risk escalation. In combination with environmental characteristics, timely prevention recommendations are pushed to the corresponding user group.

[0153] In this embodiment, medication dosage and exercise frequency are converted into "It is recommended to take 10mg of a certain type of prokinetic drug after breakfast every day for 3 consecutive days" through natural language generation technology. For unstructured content such as dietary recommendations, personalized recipes are generated based on the user's dietary habits and preferences extracted from the medical history text. A high-fiber plant protein combination is designed for vegetarians. For low-risk groups, such as those with the physiological characteristic label "stable + moderate", strategies based on lifestyle adjustments are output, such as "It is recommended to increase daily water intake to 2L and dietary fiber intake to 25g". For high-risk groups, such as those with "long-term medication history + abnormal intestinal electrical signals", a composite strategy including medical intervention is generated, such as "Drug treatment plan A + biofeedback treatment twice a week", with a strategy priority ranking.

[0154] The doctor-patient collaborative decision-making unit is configured to obtain and parse the doctor-patient interaction text, identify the doctor's adjustment needs and recommended parameters, obtain the patient's feedback needs, such as allergies to certain types of drugs, and adjust the strategy set to generate the user's constipation improvement plan. For example, if the doctor recommends changing the exercise frequency from "30 minutes a day" to "30 minutes every other day", the knowledge graph is used to verify whether the reduced exercise frequency in the adjusted plan affects intestinal motility. If the patient responds with feedback that he refuses to use drugs, non-drug alternatives are retrieved from the strategy library;

[0155] The intervention effect feedback unit is configured to continuously collect physiological data of users after they implement the constipation improvement plan, evaluate the intervention effect of the constipation improvement plan, and adjust the unimplemented strategies in the constipation improvement plan based on the intervention effect, forming a "prediction-intervention-feedback" health management closed loop.

[0156] In this embodiment, based on the dynamic division results of user groups, differentiated prevention and treatment strategies are formulated for different cluster groups to provide users with more accurate health management plans, capture the needs of both doctors and patients, and verify and adjust strategies based on the knowledge graph to generate plans that take into account the opinions of doctors and patients. Prevention strategies and suggestions are generated in combination with the environment to achieve personalized customization of the entire process, ensure continuous optimization of the plan, and improve the quality and efficiency of constipation management.

[0157] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. The constipation prediction system based on artificial intelligence large model is characterized by: include: a multimodal data processing module configured to acquire electronic medical data related to constipation based on a multi-source data interface, preprocess the acquired electronic medical data, identify anomalies in the preprocessed electronic medical data, screen out valid medical data, and extract features from the valid medical data; The multimodal data processing module also includes calculating the coefficient of variation of each feature based on the user's physiological characteristic data, constructing the user's physiological characteristic label, and combining the key indicator weights in the physiological characteristic label to quantitatively assess and grade the constipation risk of the user group; An intelligent prediction module is configured to perform vector processing based on the extraction results of valid medical data, generate a multimodal feature vector, and input it into the corresponding prediction model to predict constipation and generate a visual assessment report of the causal path of constipation risk; The dynamic decision-making module is configured to match the corresponding strategy set from the strategy library based on the constipation prediction results and the constipation field knowledge graph, generate the user's constipation improvement plan, and evaluate the intervention effect of the constipation improvement plan.

2. The constipation prediction system based on the artificial intelligence large model according to claim 1, characterized in that: Before the multimodal data processing module obtains electronic medical data, it also includes: Establish a data transmission link with the smart terminal through the network communication protocol to collect the user's physiological parameters in real time, including the user's multi-dimensional physiological indicators and special data; Build a user health archive, determine the user's physiological characteristics based on the collected physiological parameters, and classify user groups with similar physiological characteristics based on the cluster analysis model; Combining users' real-time monitoring data with health status evolution trends, the user group classification results are adaptively adjusted and user groups are dynamically divided.

3. The constipation prediction system based on the artificial intelligence large model according to claim 2, characterized in that: And dynamically divide user groups, including: Obtaining the user physiological characteristic data corresponding to each user group classification result, determining the physiological indicators of the user physiological characteristic data, calculating the coefficient of variation of each user's physiological characteristic, and constructing the corresponding physiological characteristic label; Extract key physiological indicators from user physiological characteristic data, determine the weight of each physiological indicator, calculate the comprehensive risk index of each user group, conduct quantitative assessment of each user group, and determine the risk level of each user group; Count the actual constipation incidence rates within each initial risk level range for different physiological characteristic labels, analyze the correlation between the corresponding user physiological characteristics and constipation risk, and identify key risk characteristics and risk contribution; A dynamic threshold adjustment strategy is formulated based on the correlation between user physiological characteristics and constipation risk, and the risk level threshold of user groups with each physiological characteristic label is adjusted according to the dynamic threshold adjustment strategy.

4. The constipation prediction system based on the artificial intelligence large model according to claim 3, characterized in that: Calculate the coefficient of variation of each user's physiological characteristics, including: Retrieving the user's physiological characteristic data and historical risk identification records; Determine whether the currently collected physiological characteristic data of the user is the first time the physiological characteristic data of the user is collected based on the historical risk identification record corresponding to the user; When the user is collecting physiological characteristic data for the first time, the deviation rate of each physiological characteristic data from its corresponding physiological standard data range is extracted; Obtaining an average value of the deviation rate according to the deviation rate of each physiological characteristic data from its corresponding physiological standard data range, and using the average value of the deviation rate as the coefficient of variation of the user's physiological characteristics; If this is not the first time that the user is collecting physiological characteristic data, the weight of the historical risk identification record corresponding to each physiological characteristic data is retrieved; Obtaining a weighted median corresponding to each physiological feature according to the user physiological feature data and its corresponding weight; Comparing the weighted median corresponding to each physiological feature with a preset median threshold, and screening out the number of physiological features that exceed the preset median threshold; When the number of physiological characteristics exceeding the preset median threshold does not exceed the preset reference number value, obtaining an average value of the deviation rate using all physiological characteristic data included in the currently collected user physiological characteristic data, and using the average value of the deviation rate as the coefficient of variation of the user's physiological characteristics; When the number of physiological characteristics exceeding the preset median threshold exceeds a preset number reference value, the coefficient of variation of the user's physiological characteristic data is obtained using the weighted median corresponding to each physiological characteristic.

5. The constipation prediction system based on artificial intelligence large model according to claim 4, characterized in that: The coefficient of variation of the user's physiological characteristic data is obtained using the weighted median corresponding to each physiological characteristic, including: Retrieve the weighted median corresponding to each physiological feature; Retrieve the weight value of each physiological feature that appears in historical risk identification records; Obtaining a standard deviation of the weight values ​​of each physiological characteristic appearing in the historical risk identification records using the weight values ​​of each physiological characteristic appearing in the historical risk identification records; Obtaining the weight change impact intensity corresponding to each physiological feature using the standard deviation of the weight values ​​of each physiological feature appearing in the historical risk identification records; The coefficient of variation corresponding to the current user's physiological characteristic data is obtained by using the weight change influence strength corresponding to each physiological characteristic and the weighted median corresponding to each physiological characteristic.

6. The constipation prediction system based on artificial intelligence large model according to claim 3, characterized in that: Multimodal data processing module, including: Data acquisition unit, configured as: Obtain the patient's constipation-related medical history, medication records, and symptom description text through the electronic medical record system interface; Clean and standardize the acquired electronic medical data, determine the data fluctuation range of each type of data based on the data type of the electronic medical data, and generate basic verification rules; Performing data verification on the acquired electronic medical data based on basic verification rules, and marking the electronic medical data that does not meet the verification rules as abnormal data based on the verification results; Submit the marked abnormal data to the manual review queue, remove the confirmed abnormal data based on the review results, and generate a medical data set that meets the quality standards as valid medical data; A feature extraction unit is configured to extract features from valid medical data, construct a Bayesian network to analyze the causal relationship between features, and identify the combined effects between features; The knowledge graph construction unit is configured as follows: Extract structured and unstructured data from electronic medical records, perform semantic analysis on unstructured data, identify medical entities and entity categories, perform data mapping on structured data, and establish associations with medical entities; Identify the semantic relationships between medical entities, and construct an entity relationship set based on the identification results. Then, determine the temporal relationship corresponding to the entities based on each sub-association relationship in the entity relationship set, and construct a knowledge graph in the field of constipation. Based on the constipation domain knowledge graph, the temporal relationships corresponding to the entities are analyzed, the triggering factors in the temporal relationship chain of the entities are identified, a subset of the constipation domain knowledge graph is generated, and the correlation coefficient between the user's physiological characteristics and the constipation risk is calculated.

7. The constipation prediction system based on artificial intelligence large model according to claim 6, characterized in that: Calculate the correlation coefficient between the user's physiological characteristics and the risk of constipation, including: Extract the knowledge graph in the field of constipation; Retrieving a standardized physiological characteristic value corresponding to each physiological characteristic of the user; Retrieve the user's corresponding time series risk score; wherein, the time series risk score is obtained by accumulating the time series relationship chain of triggering factors in the knowledge graph; Retrieve the number of risk triggers for the user and the average number of triggers for all users; performing a ratio process on the number of risk triggering factors of the user and the average number of triggering factors of all users to obtain the ratio between the number of risk triggering factors of the user and the average number of triggering factors of all users; The ratio between the number of risk trigger factors of the user and the average number of trigger factors of all users is identified as the time series chain confidence weight corresponding to the user; The correlation coefficient between the user's physiological characteristics and the constipation risk is obtained by using the standardized physiological characteristic value, the temporal risk score and the temporal chain confidence weight corresponding to each user's physiological characteristic.

8. The constipation prediction system based on artificial intelligence large model according to claim 7, characterized in that: Intelligent prediction module, including: a model determination unit configured to match corresponding constipation prediction models for user groups of different risk levels based on the dynamic classification results of user groups, and to adjust the matched constipation prediction model according to the latest user group classification level when the comprehensive risk index of the user group exceeds a preset fluctuation threshold; A constipation prediction unit is configured to input the feature extraction results of the valid data into the corresponding constipation prediction model to perform constipation prediction and output the probability distribution data of constipation occurring in the user in the next few days; The risk assessment unit is configured to adjust the risk assessment threshold based on the constipation field knowledge graph, and perform risk assessment based on the user's constipation prediction results. At the same time, a constipation risk causal path diagram is generated based on a subset of the constipation field knowledge graph.

9. The constipation prediction system based on the artificial intelligence large model according to claim 8, characterized in that: Dynamic decision-making module, including: a strategy matching unit configured to match a strategy set corresponding to the risk level of the user's constipation prediction result and the physiological feature label according to the constructed constipation domain knowledge graph, and to prioritize the strategies in the strategy set according to the risk level of the user's constipation prediction result; The doctor-patient collaborative decision-making unit is configured to obtain and parse the doctor-patient interaction text, identify the doctor's adjustment needs and recommended parameters, obtain the patient's feedback needs, and adjust the strategy set to generate the user's constipation improvement plan; The intervention effect feedback unit is configured to continuously collect physiological data of the user after the user implements the constipation improvement plan, evaluate the intervention effect of the constipation improvement plan, and adjust the unimplemented strategies in the constipation improvement plan according to the intervention effect.

10. The constipation prediction system based on artificial intelligence large model according to claim 9, characterized in that: The policy matching unit also includes: Based on the results of dynamic segmentation of user groups, the typical feature-strategy association rules of each user group are extracted from the constipation knowledge graph to generate corresponding strategy templates. Establish a group strategy effect comparison matrix, analyze the intervention effect response differences of different user groups based on the feedback data obtained by the intervention effect feedback unit, and adjust the weights of each strategy combination in the strategy set; Based on the physiological characteristic labels of the user group and the risk level of the user's constipation prediction results, a prevention strategy plan is generated, and timely prevention recommendations are pushed to the corresponding user group in combination with environmental characteristics.

Images