An Internet-based online health service method

Abstract

The present invention discloses an online health service method based on the Internet, which acquires the historical data of patients, including physical sign data, health data and medical history data, acquires the health standard range, and preprocesses the physical sign data; compares the physical sign data with the health standard range to obtain deviation data, and calculates the influence degree of the deviation data on the medical history data to obtain influence data; performs risk identification on the health data according to the medical history data to obtain the first weight of the influence data, and obtains the second weight of the influence data based on the medical history data; establishes a patient model according to the influence data, the first weight and the second weight, and inputs the information to be detected into the patient model to output a detection result. The present invention can improve the efficiency and accuracy of health services, and contribute to the realization of early disease prevention and intervention.

Description

Technical Field

[0001] The present invention relates to the field of medical services, and particularly to an Internet-based online health service method. Background Art

[0002] In modern society, with the acceleration of the pace of life and the increase of work pressure, people are paying more and more attention to personal health problems. Traditional health management models mostly rely on regular medical examinations and face-to-face consultations with doctors, which have certain limitations in terms of time and space. With the popularization of smart wearable devices and mobile Internet technologies, it has become possible to monitor health conditions in real time and obtain instant feedback. However, how to effectively utilize the large amount of health data collected by these devices to provide users with accurate health assessments and personalized health suggestions is still an urgent problem to be solved. Therefore, developing an Internet-based online health service method that can realize intelligent analysis and processing of health data is of great significance for improving the efficiency and quality of health management. Summary of the Invention

[0003] The object of the present invention is to provide an Internet-based online health service method.

[0004] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0005] The present invention includes the following steps:

[0006] A Obtain the historical data of the patient, including physical sign data, health data, and medical history data, obtain the health standard range, and preprocess the physical sign data;

[0007] B Compare the physical sign data with the health standard range to obtain deviation data, and calculate the influence degree of the deviation data on the medical history data to obtain influence data;

[0008] C Identify the risk of the health data according to the medical history data to obtain the first weight of the influence data, and obtain the second weight of the influence data based on the medical history data;

[0009] D Establish a patient model according to the influence data, the first weight, and the second weight, and input the information to be detected into the patient model to output a detection result.

[0010] Further, the method for obtaining the physical sign data, the health data, and the medical history data includes:

[0011] Use a smart wearable device to obtain the physical sign data, the health data and the medical history data are filled in and submitted by the patient himself / herself, and the health data is the patient's living habits.

[0012] Further, the method for comparing the physical sign data with the health standard range to obtain deviation data includes performing standardized preprocessing on the health standard range and the physical sign data, extracting abnormal data points of the physical sign data relative to the health standard range through the Isolation Forest algorithm, calculating the deviation amount of the abnormal data points relative to the health standard range, and using the deviation amount as the deviation data.

[0013] Further, the method for calculating the influence degree of the deviation data on the medical history data to obtain influence data includes:

[0014] Preprocessing the deviation data, pairing the deviation data with the medical history data, constructing a feature matrix by extracting feature vectors, and performing trend analysis on the feature matrix using the input layer and the hidden layer. Each layer consists of a group of nodes, and the nodes are connected by weights:

[0015] The number of nodes in the input layer is the same as the dimension of the feature matrix, and the number of nodes in the hidden layer is less than that of the input layer. Activation function:

[0016]

[0017] where k and u are the states of the hidden layer nodes and the input layer nodes respectively, u a is the input feature value, σ is the Sigmoid function, a and b represent the a-th input layer node and the b-th hidden layer node, w is the weight, c is the bias, and the formula of the energy function:

[0018]

[0019] Obtaining the probability of the feature vector through the energy function, and compressing the feature matrix by adjusting the weights and biases of the hidden layer.

[0020] The update formula of the weight is as follows:

[0021] Δw ab = η(<V x H y > data - <V x H y > model )

[0022] where η is the learning rate, controlling the step size of weight update, <V x H y > data represents the average value of the simultaneous activation of the x-th input layer node and the y-th hidden layer node under the given training data distribution, <V x H y > modelRepresents the average value of the simultaneous activation of two nodes under the distribution defined by the activation function at the current weight.

[0023] Calculate the contrastive divergence:

[0024]

[0025] Where ΔF(w,c a ,c b ) is the said contrastive divergence, and the weights and biases are optimized by maximizing the said contrastive divergence. and are respectively the expected activation values of the input layer nodes and the hidden layer nodes under the current weight and current bias. The potential pattern of the feature matrix is learned by iteratively adjusting the weights and biases. The activation state of the feature matrix in the hidden layer is used as the potential feature. A logistic regression model is trained using the potential feature. The influence degree of the deviation data on the medical history data is calculated using the logistic regression model, and the influence degree is used as the influence data:

[0026]

[0027] Where P(Y = 1|X) is the said influence degree, representing the probability that the response variable takes the value of 1 under the condition that the covariates are Z1, Z2,..., Z m . β0 is the intercept term, representing the probability that Y = 1 predicted by the logistic regression model when all covariates are 0. β1, β2,..., β m are the correlation coefficients of the covariates, and β0, β1, β2,..., β m are determined by maximum likelihood estimation during the training process of the logistic regression model.

[0028] Furthermore, the method for obtaining the first weight by performing risk identification on the health data according to the medical history data includes:

[0029] Encode and normalize the medical history data and the health data. The medical history data is the target vector, and the health data is the independent variable. A data set is established using the independent variable and the target vector. The data set is divided into a training set and a test set. The neural network algorithm is trained using the training set, and a receiving layer, a function layer, and a response layer are designed:

[0030] The number of nodes in the receiving layer is the same as the dimension of the training set. The function layer has n layers. The function layer learns the training set. The loss function of the function layer:

[0031] H(y,a) = -(y·log(a)+(1 - y)log(1 - a))

[0032] Among them, H(y, a) represents the loss function, y is the target vector, which is 0 or 1, a is the predicted probability, representing the probability of predicting the target vector as 1. An L2 regularization term is added to the loss function:

[0033]

[0034] Among them, L r is the regularized loss function, W is the weight matrix, and W ij is an element in W, λ is the regularization coefficient, and the function layer updates the weights by the gradient descent method:

[0035]

[0036] Among them, β is the learning rate, is the gradient of the loss function L r for the weight matrix W, X is the training set, λW is the gradient of the regularization term, the response layer has the same number of nodes as the receiving layer, the activation function of the function layer is the Sigmoid function, the response layer outputs the predicted probability of the independent variable, and the test set is used to verify the accuracy of the predicted probability, and the predicted probability is used as the first weight.

[0037] Furthermore, the method for obtaining the second weight of the influence data based on the medical history data includes:

[0038] Define a set of disease states, including the healthy state without disease and the disease state. Calculate the frequency of state transition as the state transition probability according to the medical history data, and construct a state transition matrix:

[0039]

[0040] Among them, the element P ij represents the probability of transitioning from state i to state j, and the state transition probability of the previous time step for the disease state is used as the second weight.

[0041] Furthermore, the method for establishing a patient model according to the influence data, the first weight, and the second weight includes:

[0042] Use the first weight and the second weight to perform weighted averaging on the influence data to obtain an influence factor. Use the current state of the second weight as the label of the influence factor. Use the influence factor and the label to establish a data matrix. The label is +1 or -1. For the diseases in the second weight, +1 indicates having the disease, and -1 indicates not having the disease. Use the radial basis kernel function to map the influence factor to a high-dimensional space, and find the optimal hyperplane to predict the prevalence of the disease. The objective function is:

[0043]

[0044] wherein, is the objective function, w is the weight vector, and b is the bias term. is the regularization term, and γ i and are slack variables. λ is the penalty factor, 0.1 ≤ λ ≤ 10, N is the total number of data points in the data matrix. The Lagrangian objective function introducing the Lagrange multipliers is:

[0045]

[0046] where μ i and μ j are the Lagrange multipliers. x i and x j are the data points, K is the kernel function used to calculate the similarity between two data points, ε is the regularization parameter, y i and y j are the labels of the data points x i and x j . The formula for converting the output value to the predicted value is:

[0047]

[0048] where is the predicted value. and are the maximum and minimum values of the output value, a is the curve growth coefficient, t is the time step, and T s is the theoretical time point. Create an initial population using the penalty factor and the parameters of the kernel function, and calculate the fitness of the initial population based on the medical history data:

[0049]

[0050] where g is the fitness function, σ(y, F(θ)) is the Pearson correlation coefficient between the label y and the output value F(θ), and P(D, w) is the number of outliers. is and the median absolute deviation of y. Set the initial temperature and perform random perturbation on the initial population to generate a new population:

[0051]

[0052] Where α is the current temperature, sign represents the sign function, η is a random number in [0, 1], ΔY is the perturbation range, z and z' are the initial population and the new population respectively. Compare the fitness of the initial population and the new population, adjust the acceptance probability for the new population according to the temperature, reduce the temperature and generate the new population until the fitness no longer improves. Use the penalty factor corresponding to the optimal population and the parameters of the kernel function to predict the disease, use cross-validation testing to predict the accuracy rate, and take the predicted value as the prediction result of the patient model.

[0053] Further, the information to be detected includes real-time physical sign data, real-time living habit data, and historical disease data.

[0054] The beneficial effects of the present invention are as follows:

[0055] By comprehensively analyzing the patient's physical sign data, health data, and medical history data, the present invention realizes personalized health assessment and disease risk prediction, improves the efficiency of medical services, promotes telemedicine and long-term health management, and helps with timely intervention and optimized resource allocation. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 is a flowchart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0057] The present invention will be further described below through specific embodiments. The illustrative embodiments and explanations of the present invention are used to explain the present invention, but do not limit the present invention.

[0058] An online health service method based on the Internet according to the present invention includes the following steps:

[0059] As Figure 1 shown, in this embodiment, it includes the following steps:

[0060] The purpose of the present invention is to provide an online health service method based on the Internet.

[0061] To achieve the above object, the present invention is implemented according to the following technical solution:

[0062] The present invention includes the following steps:

[0063] A Obtain the historical data of the patient, including physical sign data, health data, and medical history data, obtain the health standard range, and preprocess the physical sign data;

[0064] B Compare the physical sign data with the health standard range to obtain deviation data, and calculate the influence degree of the deviation data on the medical history data to obtain influence data;

[0065] C Identify risks for the health data based on the medical history data to obtain the first weight of the impact data, and obtain the second weight of the impact data based on the medical history data;

[0066] D Establish a patient model according to the impact data, the first weight, and the second weight, and input the information to be detected into the patient model to output the detection result.

[0067] In this embodiment, the obtained physical sign data (fasting blood glucose and postprandial blood glucose), health data, and medical history data are shown in Table 1:

[0068] Table 1

[0069] Patient Time step Fasting blood glucose (mmol / L) Postprandial blood glucose (mmol / L) Medical history data Health data 1 T1 7.2 11.5 Diabetes High sugar and high fat 1 T2 8.0 12.5 Diabetes High sugar and high fat 1 T3 6.5 9.0 Slightly elevated blood glucose Healthy diet 2 T1 5.8 7.0 Normal blood glucose Healthy diet 2 T2 6.2 8.5 Slightly elevated blood glucose High sugar diet 2 T3 5.5 6.8 Normal blood glucose Healthy diet 3 T1 9.5 14.0 Diabetes High sugar and high fat 3 T2 8.5 12.5 Diabetes High polysaccharide and high fat 3 T3 7.0 10.5 Diabetes Healthy diet

[0070] The healthy standard range: fasting blood glucose 3.9 to 5.6 mmol / L, postprandial blood glucose < 7.8 mmol / L.

[0071] In this embodiment, the methods for obtaining the physical sign data, health data, and medical history data include:

[0072] Use a smart wearable device to obtain the physical sign data, and the health data and medical history data are filled in and submitted by the patient himself / herself. The health data is the patient's living habits.

[0073] In this embodiment, the method for comparing the physical sign data with the healthy standard range to obtain deviation data includes performing standardized preprocessing on the healthy standard range and the physical sign data, extracting the abnormal data points of the physical sign data relative to the healthy standard range through the Isolation Forest algorithm, calculating the deviation amount of the abnormal data points relative to the healthy standard range, and using the deviation amount as the deviation data.

[0074] In this embodiment, the deviation data is:

[0075] Patient 1: At time point T1: fasting blood glucose is high (deviation 1.6 mmol / L), postprandial blood glucose is high (deviation 3.7 mmol / L); at time point T2: fasting blood glucose is high (deviation 2.4 mmol / L), postprandial blood glucose is high (deviation 4.7 mmol / L); at time point T3: fasting blood glucose is normal, postprandial blood glucose is high (deviation 1.2 mmol / L);

[0076] Patient 2: At time point T1: fasting blood glucose is normal, postprandial blood glucose is normal; at time point T2: fasting blood glucose is normal, postprandial blood glucose is high (deviation 0.7 mmol / L); at time point T3: fasting blood glucose is normal, postprandial blood glucose is normal;

[0077] Patient 3: At time point T1: Fasting blood glucose is on the high side (deviation 3.9 mmol / L), and postprandial blood glucose is on the high side (deviation 6.2 mmol / L); At time point T2: Fasting blood glucose is on the high side (deviation 2.9 mmol / L), and postprandial blood glucose is on the high side (deviation 4.7 mmol / L); At time point T3: Fasting blood glucose is on the high side (deviation 1.4 mmol / L), and postprandial blood glucose is on the high side (deviation 2.7 mmol / L).

[0078] In this embodiment, the method for calculating the influence degree of the deviation data on the medical history data to obtain the influence data includes:

[0079] Preprocess the deviation data, pair the deviation data with the medical history data, construct a feature matrix by extracting feature vectors, and perform trend analysis on the feature matrix using the input layer and the hidden layer. Each layer consists of a group of nodes, and the nodes are connected by weights:

[0080] The number of nodes in the input layer is the same as the dimension of the feature matrix, and the number of nodes in the hidden layer is less than that of the input layer. Activation function:

[0081]

[0082] where k and u are the states of the hidden layer nodes and the input layer nodes respectively, u a is the input feature value, σ is the Sigmoid function, a and b represent the a-th input layer node and the b-th hidden layer node, w is the weight, c is the bias, and the formula of the energy function:

[0083]

[0084] Obtain the probability of the feature vector through the energy function, and compress the feature matrix by adjusting the weights and biases of the hidden layer,

[0085] The update formula of the weight is as follows:

[0086] Δw ab =η(<V x H y > data -<V x H y > model )

[0087] where η is the learning rate, controlling the step size of weight update, <V x H y > data represents the average value of the simultaneous activation of the x-th input layer node and the y-th hidden layer node under the given training data distribution, <V x H y >model It represents the average value of the simultaneous activation of two nodes under the distribution defined by the activation function at the current weight.

[0088] Calculate the contrastive divergence:

[0089]

[0090] where ΔF(w, c a , c b ) is the said contrastive divergence, and the weights and biases are optimized by maximizing the said contrastive divergence. and are respectively the expected activation values of the input layer nodes and the hidden layer nodes under the current weight and current bias. The potential pattern of the feature matrix is learned by iteratively adjusting the weights and biases. The activation state of the feature matrix in the hidden layer is used as the potential feature, and a logistic regression model is trained using the potential feature. The influence degree of the deviation data on the medical history data is calculated using the logistic regression model, and the influence degree is used as the influence data:

[0091]

[0092] where P(Y = 1|X) is the said influence degree, representing the probability that the response variable takes the value of 1 under the condition that the covariates are Z1, Z2, …, Z m . β0 is the intercept term, representing the probability that Y = 1 predicted by the logistic regression model when all the covariates are 0. β1, β2, …, β m are the correlation coefficients of the covariates, and β0, β1, β2, …, β m are determined by maximum likelihood estimation during the training process of the logistic regression model.

[0093] In this embodiment, the influence data is:

[0094] Patient 1: T1, 0.7; T2, 0.8; T3, 0.6;

[0095] Patient 2: T1, 0.3; T2, 0.4; T3, 0.3;

[0096] Patient 3: T1, 0.9; T2, 0.85; T3, 0.8.

[0097] In this embodiment, the method for obtaining the first weight by risk identifying the health data according to the medical history data includes:

[0098] Encode and normalize the medical history data and the health data. The medical history data is the target vector, and the health data is the independent variable. Use the independent variable and the target vector to establish a data set, divide the data set into a training set and a test set, and use the training set to train the neural network algorithm. Design a receiving layer, a function layer, and a response layer:

[0099] The number of nodes in the receiving layer is the same as the dimension of the training set. The function layer has n layers. The function layer learns the training set. The loss function of the function layer:

[0100] H(y,a) = -(y·log(a)+(1-(y·log(1-a))

[0101] Where H(y,a) represents the loss function, y is the target vector, which is 0 or 1, a is the predicted probability, representing the probability of predicting the target vector as 1. Add an L2 regularization term to the loss function:

[0102]

[0103] Where L r Is the regularized loss function, W is the weight matrix, and W ij Is an element in W, λ is the regularization coefficient. The function layer updates the weights by the gradient descent method:

[0104]

[0105] Where β is the learning rate, Is the gradient of the loss function L r For the gradient of the weight matrix W, X is the training set, and λW is the gradient of the regularization term. The response layer has the same number of nodes as the receiving layer. The activation function of the function layer is the Sigmoid function. The response layer outputs the predicted probability of the independent variable. Use the test set to verify the accuracy of the predicted probability, and use the predicted probability as the first weight.

[0106] In this embodiment, the first weight is:

[0107] Patient 1: T1, 0.65; T2, 0.7; T3, 0.68;

[0108] Patient 2: T1, 0.45; T2, 0.5; T3, 0.48;

[0109] Patient 3: T1, 0.85; T2, 0.80; T3, 0.78.

[0110] In this embodiment, the method for obtaining the second weight of the impact data based on the medical history data includes:

[0111] Define a set of disease states, including the healthy state of not having the disease and the disease state. Calculate the frequency of state transition as the state transition probability according to the medical history data, and construct a state transition matrix:

[0112]

[0113] where the element P ij represents the probability of transitioning from state i to state j, and use the state transition probability of the disease state at the previous time step as the second weight.

[0114] In this embodiment, state 1 represents the non - diabetic state, and state 2 represents the diabetic state. The state transition matrix:

[0115]

[0116] The state transition probability of the patient for the disease state:

[0117] Patient 1: T1, T2, P 22 = 0.75; T3, P 12 = 0.33;

[0118] Patient 2: T1, T2, T3, P 12 = 0.33;

[0119] Patient 3: T1, T2, P 22 = 0.75; T3, P 12 = 0.33;

[0120] The second weight:

[0121] Patient 1: For T1, the state at the previous time step is unknown, take P i2 = 0.5; T2, T3, P 22 = 0.75; T4, P 12 = 0.33;

[0122] Patient 2: T1, P i2 = 0.5; T2, T3, T4, P 12 = 0.33;

[0123] Patient 3: T1, P i2 = 0.5; T2, T3, P 22 = 0.75; T4, P 12 = 0.33.

[0124] In this embodiment, the method for establishing a patient model according to the influence data, the first weight, and the second weight includes:

[0125] The influence data is weighted and averaged using the first weight and the second weight to obtain an influence factor. The current state of the second weight is used as the label of the influence factor, and a data matrix is established using the influence factor and the label. The label is +1 or -1. For the diseases in the second weight, +1 indicates having the disease, and -1 indicates not having the disease. The influence factor is mapped to a high-dimensional space using a radial basis kernel function, and an optimal hyperplane is found to predict the prevalence of the disease. The objective function is as follows:

[0126]

[0127] Where, is the objective function, w is the weight vector, b is the bias term, is the regularization term, γ i and are slack variables, λ is the penalty factor, 0.1 ≤ λ ≤ 10, take λ = 5, N is the total number of data points in the data matrix. The Lagrangian objective function introducing Lagrange multipliers is:

[0128]

[0129] Where μ i and μ j are the Lagrange multipliers, x i and x j are the data points, K is the kernel function used to calculate the similarity between two data points, ε is the regularization parameter, y i and y j are the labels of the data points x i and x j The formula for converting the output value to the predicted value is:

[0130]

[0131] Where is the predicted value, and are the maximum and minimum values of the output value, a is the curve growth coefficient, t is the time step, T s is the theoretical time point. An initial population is created using the penalty factor and the parameters of the kernel function, and the fitness of the initial population is calculated based on the medical history data:

[0132]

[0133] where \(g\) is the fitness function, \(\sigma(y, F(\theta))\) is the Pearson correlation coefficient between the label \(y\) and the output value \(F(\theta)\), and \(P(D, w)\) is the number of outliers. is and the median absolute deviation of \(y\). Set the initial temperature and perform random perturbations on the initial population to generate a new population:

[0134]

[0135] where \(\alpha\) is the current temperature, sign represents the sign function, \(\eta\) is a random number in \([0, 1]\), \(\Delta Y\) is the perturbation range, \(z\) and \(z'\) are the initial population and the new population respectively. Compare the fitness of the initial population and the new population, and adjust the acceptance probability for the new population according to the temperature:

[0136]

[0137] where \(A\) is the acceptance probability for the new population, \(g(z)\) and \(g(z')\) are the fitnesses of the initial population and the new population respectively, \(\Delta H\) is the difference in fitness. Decrease the temperature and generate the new population until the fitness no longer improves. Use the penalty factor corresponding to the optimal population and the parameters of the kernel function to predict the disease, and use cross-validation testing to predict the accuracy rate. Take the predicted value as the prediction result of the patient model.

[0138] In this embodiment, the information to be detected is: Patient 4, fasting blood glucose 7.0 mmol / L, postprandial blood glucose 10.0 mmol / L, high sugar and high fat, and had diabetes at the previous time step.

[0139] In this embodiment, the information to be detected includes real-time physical sign data, real-time living habit data, and historical disease data.

[0140] In this embodiment, the detection result is: The probability that Patient 4 has diabetes is 0.85.

[0141] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. An online health service method based on the Internet, characterized in that, It includes the following steps: A. Obtain the historical data of the patient, including physical sign data, health data, and medical history data, obtain the health standard range, and preprocess the physical sign data; B. Compare the physical sign data with the health standard range to obtain deviation data, and calculate the influence degree of the deviation data on the medical history data to obtain influence data; C. Identify the risk of the health data based on the medical history data to obtain the first weight of the influence data, and obtain the second weight of the influence data based on the medical history data; D. Establish a patient model according to the influence data, the first weight, and the second weight, and input the information to be detected into the patient model to output the detection result.

2. The online health service method based on the Internet according to claim 1, wherein The methods for obtaining the physical sign data, the health data, and the medical history data in step A include: Obtain the physical sign data using a smart wearable device, and the health data and the medical history data are filled in and submitted by the patient himself / herself, and the health data is the patient's living habits.

3. The online health service method based on the Internet according to claim 1, wherein The method for comparing the physical sign data with the health standard range to obtain deviation data in step B includes performing standardized preprocessing on the health standard range and the physical sign data, extracting abnormal data points of the physical sign data relative to the health standard range through the Isolation Forest algorithm, calculating the deviation amount of the abnormal data points relative to the health standard range, and using the deviation amount as the deviation data.

4. The online health service method based on the Internet according to claim 1, characterized in that The method for calculating the influence degree of the deviation data on the medical history data to obtain influence data in step B includes: Preprocess the deviation data, pair the deviation data with the medical history data, construct a feature matrix by extracting feature vectors, and perform trend analysis on the feature matrix using the input layer and the hidden layer. Each layer consists of a group of nodes, and the nodes are connected by weights: The number of nodes in the input layer is the same as the dimension of the feature matrix, the number of nodes in the hidden layer is less than that of the input layer, and the activation function: where k and u are the states of the hidden layer node and the input layer node respectively, and u a is the input feature value, σ is the Sigmoid function, a and b represent the a-th input layer node and the b-th hidden layer node, w is the weight, c is the bias, and the formula of the energy function is: Obtain the probability of the feature vector through the energy function, and compress the feature matrix by adjusting the weights and biases of the hidden layer. The update formula of the weights is as follows: Δw ab = η(<V x H y > data - <V x H y > model ) where η is the learning rate, controlling the step size of weight update, <V x H y > data represents the average value of the simultaneous activation of the x-th input layer node and the y-th hidden layer node under the given training data distribution, <V x H y > model represents the average value of the simultaneous activation of the two nodes under the distribution defined by the activation function with the current weights, Calculate the contrast divergence: where ΔF(w, c a , c b ) is the contrast divergence, and the weights and biases are optimized by maximizing the contrast divergence, and are the expected activation values of the input layer nodes and the hidden layer nodes under the current weights and current biases respectively. The latent patterns of the feature matrix are learned by iteratively adjusting the weights and biases. The activation state of the feature matrix in the hidden layer is used as the latent feature. The latent feature of the deviation data and the latent feature of the medical history data are used as the covariate and the response variable respectively to train a logistic regression model. The influence degree of the deviation data on the medical history data is calculated using the logistic regression model, and the influence degree is used as the influence data: Among them, P(Y = 1|X) is the influence degree, indicating that under the condition that the covariates are Z1, Z2, …, Z m , the probability that the response variable takes the value of 1, β0 is the intercept term, indicating that when all the covariates are 0, the probability that the logistic regression model predicts Y = 1, β1, β2, …, β m are the correlation coefficients of the covariates, and β0, β1, β2, …, β m are determined by maximum likelihood estimation during the training process of the logistic regression model.

5. The online health service method based on the Internet according to claim 1, wherein The method for identifying the risk of the health data based on the medical history data to obtain the first weight in step C includes: Encode and normalize the medical history data and the health data. The medical history data is the target vector, and the health data is the independent variable. Establish a data set using the independent variable and the target vector, divide the data set into a training set and a test set, train the neural network algorithm using the training set, and design a receiving layer, a function layer, and a response layer: The number of nodes in the receiving layer is the same as the dimension of the training set, the function layer has n layers, the function layer learns the training set, and the loss function of the function layer: H(y,a) = -(y·log(a) + (1 - y)·log(1 - a)) where H(y,a) represents the loss function, y is the target vector, which is 0 or 1, a is the predicted probability, representing the probability of predicting the target vector as 1. Add an L2 regularization term to the loss function: where L r is the regularized loss function, W is the weight matrix, and W ij is an element in W, λ is the regularization coefficient, and the function layer updates the weights by the gradient descent method: where β is the learning rate, is the loss function L r is the gradient of the weight matrix W, X is the training set, λW is the gradient of the regularization term, the nodes of the response layer are the same as those of the receiving layer, the activation function of the function layer is the Sigmoid function, the response layer outputs the predicted probability of the independent variable, the accuracy of the predicted probability is verified using the test set, and the predicted probability is used as the first weight.

6. The online health service method based on the Internet according to claim 1, wherein The method for obtaining the second weight of the influence data based on the medical history data in step C includes: Defining a set of disease states, including the healthy state without disease and the disease state, calculating the frequency of state transition as the state transition probability according to the medical history data, and constructing a state transition matrix: wherein the element P ij represents the probability of transitioning from state i to state j, and the state transition probability for the disease state in the previous time step is used as the second weight.

7. The online health service method based on the Internet according to claim 1, wherein The information to be detected in step D includes real-time physical sign data, real-time living habit data, and historical disease data.

Images