An early warning analysis method and system based on big data of chronic disease medications

Abstract

The present invention discloses a warning analysis method and system based on big data of chronic disease medications, including the following steps: Step S1, constructing a historical user purchase portrait of chronic disease users based on the drug purchase logs of chronic disease users; Step S2, performing similarity classification on chronic disease users based on the historical user purchase portrait to obtain multiple user groups so as to achieve group management of chronic disease users; Step S3, in each user group, real-time monitoring the current behavior characteristics of chronic disease users when purchasing drugs, predicting chronic disease users with abnormal drug purchase behaviors in the user group, and then giving a warning feedback to the chronic disease users with abnormal drug purchase behaviors. The present invention establishes a user purchase portrait for chronic disease users, and synchronously extracts the common features and personalized features in the behavior characteristics of chronic disease users when establishing the user purchase portrait, and comprehensively grasps the behavior characteristics of chronic disease users.

Description

Technical Field

[0001] The present invention relates to the technical field of medical big data, and particularly to a warning analysis method and system based on big data of chronic disease medications. Background Art

[0002] At present, there are a large number of chronic disease patients in our country. More than 300 million people across the country are chronic disease patients, and the number of deaths caused by chronic diseases has accounted for 80% of the total number of deaths due to diseases in our country. Most chronic diseases require long-term medication, and the average age of people with chronic diseases is relatively high. Often, the medicine has been used up and they forget to buy it again. Medication interruption will seriously threaten the physical health of patients, and pharmacies will also lose users and reduce the sales volume of drugs as a result. Therefore, if a pharmacy can predict the medication time of users who have purchased chronic disease drugs in this pharmacy and remind them at the appropriate time, then this will not only be beneficial for patients to take medicine in time and ensure their lives, but also improve the drug sales volume of the pharmacy and maintain the users who buy medicine.

[0003] Currently, the general prediction of commodity purchases is made by analyzing all the commodities purchased by users and their browsing and clicking behaviors. However, this prediction method is not applicable to the prediction of chronic disease drugs, and it has the following deficiencies: Chronic disease users usually regularly purchase a fixed type of drug. Therefore, for the warning analysis of chronic disease medications, it is determined whether users take medicine regularly by observing whether they regularly purchase drugs. The purchasing behavior characteristics of chronic disease users are simple and can be monitored and managed through groups. However, the existing technology usually monitors each chronic disease user individually, which will lead to a large amount of monitoring data and a large number of data processing threads, reducing the warning efficiency. Summary of the Invention

[0004] The purpose of the present invention is to provide a warning analysis method and system based on big data of chronic disease medications to solve the technical problem in the prior art that each chronic disease user is monitored individually, which will lead to a large amount of monitoring data and a large number of data processing threads, reducing the warning efficiency.

[0005] To solve the above technical problem, the present invention specifically provides the following technical solutions:

[0006] A warning analysis method based on big data of chronic disease medications, comprising the following steps:

[0007] Step S1, constructing a historical user purchase portrait of chronic disease users based on the drug purchase logs of chronic disease users, where the historical user purchase portrait is used to visualize the historical behavior characteristics of chronic disease users in purchasing drugs;

[0008] Step S2: Based on the historical user purchase profiles, classify chronic disease users into multiple user groups for group management of chronic disease users. Among them, the historical user purchase profiles of all chronic disease users in the same user group have low differences, and the historical user purchase profiles of chronic disease users in different user groups have high differences;

[0009] Step S3: In each user group, monitor in real time the current behavior characteristics of chronic disease users when purchasing drugs, compare the differences in the current behavior characteristics of all chronic disease users in the user group, predict the chronic disease users with abnormal drug purchase behavior in the user group, and then give early warning feedback to the chronic disease users with abnormal drug purchase behavior to standardize the drug purchase behavior of chronic disease users.

[0010] As a preferred solution of the present invention, constructing the historical user purchase profile of chronic disease users based on the drug purchase logs of chronic disease users includes:

[0011] Extract the feature fields representing user behavior characteristics from the drug purchase logs to extract the common features of the historical user purchase profiles of chronic disease users, and convert the drug purchase logs into user behavior samples based on the feature fields. The feature fields include: user attributes, drug attributes, and purchase behavior attributes;

[0012] Add feature weights to the feature fields to extract the personalized features of the historical user purchase profiles of chronic disease users, and perform weight combination on the feature fields in the user behavior samples based on the feature weights to obtain the historical user purchase profile. The functional expression of the historical user purchase profile is:

[0013] F i ={[α i，1，t f i，1，t ,α i，2，t f i，2，t ,α i，3，t f i，3，t |t∈[1,n]};

[0014] In the formula, F i represents the historical user purchase profile of the i-th chronic disease user, [α i,1,t f i,1,t ,α i,2 , t f i,2,t ,α i,3,t f i,3,t represents the historical user purchase profile of the i-th chronic disease user at the historical moment t, f i,1,t 、f i,2,t 、f i,3,tData vectors respectively representing the user attributes, drug attributes, and purchase behavior attributes of the \(i\)-th chronic disease user at historical moment \(t\), \(\alpha\) i,1,t ,\(\alpha\) i,2,t ,\(\alpha\) i,3,t respectively represent the feature weights of the user attributes, drug attributes, and purchase behavior attributes of the \(i\)-th chronic disease user at historical moment \(t\). \(i\) and \(t\) are measurement constants without substantial meaning, and \(n\) represents the total number of historical moments.

[0015] As a preferred solution of the present invention, adding feature weights to the feature fields includes:

[0016] Calculating in sequence the degree of fluctuation of the data vectors representing the drug attributes and purchase behavior attributes of chronic disease users. The degree of fluctuation is a quantitative index for measuring the personalized characteristics of the historical user purchase portrait of chronic disease users. The calculation formula for the degree of fluctuation is:

[0017]

[0018]

[0019] In the formula, \(f\) i，2，t-1 , \(f\) i，3，t-1 respectively represent the data vectors representing the drug attributes and purchase behavior attributes of the \(i\)-th chronic disease user at historical moment \(t - 1\);

[0020] Based on the degree of fluctuation, setting the feature weights of the user attributes, drug attributes, and purchase behavior attributes of chronic disease users. The feature weights of the user attributes, drug attributes, and purchase behavior attributes are in sequence:

[0021] \(\alpha\) i，1，t = 1 - \(\alpha\) i，2，t - \(\alpha\) i，3，t ;

[0022]

[0023]

[0024] In the formula, \(w\) i，2，t , \(w\) i，3，t respectively represent the degrees of fluctuation of the data vectors of the drug attributes and purchase behavior attributes.

[0025] As a preferred solution of the present invention, the data vectors representing the user attributes, drug attributes, and purchase behavior attributes of the \(i\)-th chronic disease user at historical moment \(t\) are respectively vectors composed of the data representing the user attributes, drug attributes, and purchase behavior attributes of the \(i\)-th chronic disease user in the user behavior samples from historical moment 1 to historical moment \(t\), where

[0026] f i，1，t = {user i，τ | τ ∈ [1, t]};

[0027] f i，2，t = {object i，τ | τ ∈ [1, t]};

[0028] f i，3，t = {action i，τ | τ ∈ [1, t]};

[0029] In the formula, user i，t , object i，τ , action i，τ respectively represent the data of the user attribute, drug attribute, and purchase behavior attribute of the i-th chronic disease user at the historical moment τ; τ is a measurement constant without substantial meaning.

[0030] As a preferred solution of the present invention, the similarity classification of the historical user purchase portraits to obtain multiple user groups includes:

[0031] Step 1: Quantify each historical user purchase portrait into a single user group, and sequentially perform averaging processing on all historical user purchase portraits in each user group to obtain the group portrait of each user group;

[0032] Step 2: Calculate the Euclidean distance between the group portraits of any two user groups in sequence to measure the similarity of any two user groups, and fuse the two user groups corresponding to the minimum similarity, and then perform averaging processing on all historical user purchase portraits in each user group to complete the update of the group portrait. The similarity is used to measure the probability that two user groups represent the same type of historical behavior characteristics;

[0033] Step 3: Repeat Step 2 until the similarity of any two user groups is greater than the set threshold, and output the current user group as the classification result of the historical user purchase portrait.

[0034] As a preferred solution of the present invention, the calculation formula of the group portrait is:

[0035]

[0036] In the formula, G j represents the group portrait of the j-th user group, m j represents the total number of chronic disease users in the j-th user group, and j is a measurement constant without substantial meaning;

[0037] The calculation formula of the similarity is:

[0038]

[0039] Wherein, I j,k represents the similarity between the j-th user group and the k-th user group, and G j , G k respectively represent the group portraits of the j-th and k-th user groups, T is the transpose symbol, and k is a measurement constant without substantial meaning.

[0040] As a preferred embodiment of the present invention, predicting the chronic disease users with abnormal drug purchase behavior in the user group by comparing the differences in the existing behavior characteristics of all chronic disease users in the user group includes:

[0041] Quantifying the purchase data representing the existing behavior characteristics of the chronic disease users into the existing behavior samples in the form of the characteristic fields;

[0042] Combining the characteristic fields in the existing behavior samples to obtain the real-time user portrait, and the function expression of the real-time user portrait is:

[0043] R i，r，ε ={f r，1，ε , f r，2，ε , f r，3，ε};

[0044] Wherein, R j,r,ε represents the real-time user portrait of the r-th chronic disease user in the j-th user group at the current moment ε, and f r,1,ε , f r,2,ε , f r,3,ε respectively represent the data vectors of the user attributes, drug attributes, and purchase behavior attributes of the r-th chronic disease user at the current moment ε;

[0045] Calculate in sequence the similarity between the real-time user portrait R j,r,ε of the r-th chronic disease user in the user group and the real-time user portraits {R j,l,ε |l≠r∩l∈[1,m j} of all the remaining chronic disease users in the user group, and count the number of similarities less than or equal to the similarity threshold;

[0046] Take the ratio of the number of similarities less than or equal to the similarity threshold to the total number m j of the chronic disease users in the j-th user group as the probability that the r-th chronic disease user has abnormal drug purchase behavior.

[0047] As a preferred embodiment of the present invention, predicting the chronic disease users with abnormal drug purchase behavior in the user group by comparing the differences in the existing behavior characteristics of all chronic disease users in the user group further includes:

[0048] Set a probability threshold, and compare the probability that the r-th chronic disease user has abnormal drug purchase behavior with the probability threshold, where

[0049] When the probability that the r-th chronic disease user has abnormal drug purchase behavior exceeds the probability threshold, it is determined that the r-th chronic disease user has abnormal drug purchase behavior;

[0050] When the probability that the r-th chronic disease user has abnormal drug purchase behavior does not exceed the probability threshold, it is determined that the r-th chronic disease user does not have abnormal drug purchase behavior.

[0051] As a preferred embodiment of the present invention, the present invention provides an early warning system according to an early warning analysis method based on big data of chronic disease medication, including:

[0052] A portrait building unit, which constructs a historical user purchase portrait of chronic disease users based on the drug purchase logs of chronic disease users, and the historical user purchase portrait is used to visualize the historical behavior characteristics of chronic disease users' drug purchases;

[0053] A group classification unit, which is used to perform similarity classification on chronic disease users based on the historical user purchase portrait to obtain multiple user groups to realize group management of chronic disease users. Among them, the historical user purchase portraits of all chronic disease users in the same user group have low differences, and the historical user purchase portraits of chronic disease users in different user groups have high differences;

[0054] A group monitoring unit, which is used to monitor the current behavior characteristics of chronic disease users' drug purchases in real time in each user group, compare the differences in the current behavior characteristics of all chronic disease users in the user group, predict the chronic disease users with abnormal drug purchase behavior in the user group, and then give an early warning feedback to the chronic disease users with abnormal drug purchase behavior to standardize the drug purchase behavior of chronic disease users;

[0055] A user terminal, which is used to receive the early warning feedback from the group monitoring unit.

[0056] As a preferred embodiment of the present invention, the group monitoring unit and the user terminal interact feedback information through a network communication protocol.

[0057] The present invention has the following beneficial effects compared with the prior art:

[0058] The present invention creates a user purchase profile for chronic disease users, and synchronously extracts the common features and personalized features in the behavior characteristics of chronic disease users when creating the user purchase profile, so as to more comprehensively grasp the behavior characteristics of chronic disease users. Moreover, based on the historical user purchase profiles, the chronic disease users are classified into multiple user groups for group management, reducing the monitoring threads and improving the management efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only exemplary, and for those of ordinary skill in the art, without creative efforts, other implementation drawings can be obtained according to the provided drawings.

[0060] Figure 1 It is a flowchart of the early warning analysis method provided by the embodiment of the present invention;

[0061] Figure 2 It is a structural block diagram of the early warning system provided by the embodiment of the present invention;

[0062] Figure 3 It is a schematic diagram of the user group structure provided by the embodiment of the present invention.

[0063] The reference numerals in the drawings are respectively represented as follows:

[0064] 1 - Portrait establishment unit; 2 - Group classification unit; 3 - Group monitoring unit; 4 - User terminal; 5 - User group; 6 - Chronic disease user. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0065] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0066] Such as Figure 1As shown, chronic diseases usually only require regular intake of the same drug for treatment. Therefore, the behavioral characteristics of chronic disease users are relatively simple. The purchase population, purchase frequency, and purchased drugs are relatively fixed. Only the purchase behavior data of the same drug needs to be analyzed to identify chronic disease users with irregular drug purchases, and then early warnings can be issued to chronic disease users with irregular drug purchases. It does not involve analyzing all the goods purchased by users and the user browsing and clicking behaviors for user purchase behavior analysis. Therefore, the present invention provides a warning analysis method based on big data of chronic disease medication, constructs user portraits for chronic disease users, and conducts group management for chronic disease users, reducing the data processing volume to improve the warning efficiency.

[0067] A warning analysis method based on big data of chronic disease medication, comprising the following steps:

[0068] Step S1, constructing a historical user purchase portrait of chronic disease users based on the drug purchase logs of chronic disease users, where the historical user purchase portrait is used to visualize the historical behavioral characteristics of chronic disease users in purchasing drugs;

[0069] Constructing a historical user purchase portrait of chronic disease users based on the drug purchase logs of chronic disease users includes:

[0070] Extracting feature fields representing user behavioral characteristics from the drug purchase logs to extract the common features of the historical user purchase portrait of chronic disease users, and converting the drug purchase logs into user behavior samples based on the feature fields. The feature fields include: user attributes, drug attributes, and purchase behavior attributes;

[0071] The full name of chronic diseases is chronic non-communicable diseases. It does not refer to a specific disease, but is a general term for a group of diseases with insidious onset, long course, and protracted illness, lacking conclusive evidence of infectious biological causes, having complex etiologies, and some of which have not been fully confirmed. Common chronic diseases mainly include cardiovascular and cerebrovascular diseases, cancer, diabetes, and chronic respiratory diseases. Among them, cardiovascular and cerebrovascular diseases include hypertension, stroke, and coronary heart disease. Chronic disease users suffering from a certain chronic disease usually only need to regularly purchase the designated drugs for treating the chronic disease in the hospital or pharmacy. For example, hypertension patients only need to regularly purchase antihypertensive drugs. Therefore, in constructing the user portrait of chronic disease users, it only involves user attributes (age, gender, etc.), drug attributes (drug types, etc.), and purchase behavior attributes (purchase frequency, etc.), without involving complex multi-source data. User attributes (age, gender, etc.), drug attributes (drug types, etc.), and purchase behavior attributes (purchase frequency, etc.) can already completely express the common characteristics of the historical behavior characteristics of chronic disease users in purchasing drugs. The common characteristics are the attribute characteristics that each chronic disease user has. Based on the common characteristics of the historical behavior characteristics, by adding weight attributes to user attributes (age, gender, etc.), drug attributes (drug types, etc.), and purchase behavior attributes (purchase frequency, etc.), the personalized attributes of chronic disease users are highlighted. Both drug attributes and purchase behavior attributes are the main features of concern in the historical behavior characteristics of chronic disease users. The weight will highlight the personalized characteristics of each chronic disease user. For example, if the purchase frequency of chronic disease user A is higher than that of B, then in constructing the historical user purchase portraits of chronic disease users A and B, the weights of the purchase behavior attributes are different, further highlighting the differences in the historical user purchase portraits of chronic disease users A and B, and realizing personalized representation.

[0072] Add feature weights to the feature fields to extract the personalized characteristics of the historical user purchase portraits of chronic disease users, and based on the feature weights, perform weight combination on the feature fields in the user behavior samples to obtain the historical user purchase portraits. The function expression of the historical user purchase portraits is:

[0073] F i ={[α i，1，t f i，1，t ,α i，2，t f i，2，t ,α i，3，t f i，3，t |t∈[1,n]};

[0074] In the formula, F i represents the historical user purchase portrait of the i-th chronic disease user. The user portrait is a vector composed of data covering user attributes, drug attributes, and purchase behavior attributes at all historical moments. [α i，1，t f i，1，t ,α i，2， tf i，2，t ,α i，3，t f i，3，t represents the historical user purchase profile of the \(i\)-th chronic disease user at the historical moment \(t\), \(f i，1，t 、f i，2，t 、f i，3，t respectively represent the data vectors of the user attributes, drug attributes, and purchase behavior attributes of the \(i\)-th chronic disease user at the historical moment \(t\), and \(α i，1，t 、α i，2，t 、α i，3，t respectively represent the feature weights of the user attributes, drug attributes, and purchase behavior attributes of the \(i\)-th chronic disease user at the historical moment \(t\). \(i\) and \(t\) are measurement constants without actual meaning, and \(n\) represents the total number of historical moments.

[0075] The data vectors representing the user attributes, drug attributes, and purchase behavior attributes of the \(i\)-th chronic disease user at the historical moment \(t\) are respectively composed of the data vectors of the user attributes, drug attributes, and purchase behavior attributes of the \(i\)-th chronic disease user in the user behavior samples from historical moment 1 to historical moment \(t\). Among them,

[0076] f i，1，t = {user i，τ |τ ∈ [1, t]};

[0077] f i，2，t = {object i，τ |τ ∈ [1, t]};

[0078] f i，3，t = {action i，τ |τ ∈ [1, t]};

[0079] In the formula, user i,τ 、object i,τ 、action i,τ respectively represent the data of the user attributes, drug attributes, and purchase behavior attributes of the \(i\)-th chronic disease user at the historical moment τ; τ is a measurement constant without actual meaning.

[0080] Add feature weights to the feature fields, including:

[0081] Calculate the fluctuation degree of the data vectors representing the drug attributes and purchase behavior attributes of the chronic disease user in sequence. The fluctuation degree is a quantitative index used to measure the personalized characteristics of the historical user purchase profile of the chronic disease user. The calculation formula of the fluctuation degree is:

[0082]

[0083]

[0084] where, f i,2,t-1 and f i,3,t-1 respectively represent data vectors characterizing the drug attribute and the purchase behavior attribute of the i-th chronic disease user at the historical moment t - 1;

[0085] The drug attribute and the purchase behavior attribute are personalized feature characterizations for constructing the historical user purchase portrait. Therefore, when focusing on personalized features, it is only necessary to focus on constructing the weights of the drug attribute and the purchase behavior attribute, that is, to focus on the degree of fluctuation of the drug attribute and the purchase behavior attribute. The greater the degree of fluctuation of the drug attribute and the purchase behavior attribute, the more worthy of attention the personalized features of the drug attribute and the purchase behavior attribute are, that is, they need to be further highlighted when constructing the user portrait. Therefore, it is necessary to dynamically adjust the feature weights of the drug attribute and the purchase behavior attribute according to the degree of fluctuation to achieve the highlighting of the personalized features of the drug attribute and the purchase behavior attribute using the weights. The specific method is as follows:

[0086] Based on the degree of fluctuation, set the feature weights of the user attribute, drug attribute, and purchase behavior attribute of the chronic disease user. The feature weights of the user attribute, drug attribute, and purchase behavior attribute are in turn:

[0087] α i，1，t = 1 - α i，2，t -α i，3，t ;

[0088]

[0089]

[0090] where, w i,2,t and w i,3,t respectively represent the degrees of fluctuation of the data vectors of the drug attribute and the purchase behavior attribute.

[0091] The personalization priority of purchase behavior attributes is higher than that of drug attributes. Therefore, when the fluctuation degree of purchase behavior attributes is large, it is necessary to increase the weight of purchase behavior attributes on the basis of being greater than 0.5. Synchronously, when the drug attributes fluctuate greatly with the purchase behavior attributes and the fluctuation degree of drug attributes is large, the weight will be slightly reduced on the basis of being lower than 0.5, and the personalized highlighting effect of drug attributes will be given to the personalized highlighting effect of purchase behavior attributes. Then, the personalized highlighting effect of the obtained user attributes will be basically maintained in a stable state, that is, the weights of user attributes are stable, which conforms to the actual situation of the historical behavior characteristics of chronic disease users. The actual fluctuation possibilities of purchase behavior attributes, drug attributes and user attributes are from high to low, and the attention degrees to purchase behavior attributes, drug attributes and user attributes are also from high to low when studying the historical behavior characteristics of users. Therefore, the representativeness degrees of purchase behavior attributes, drug attributes and user attributes for the personalized characteristics of the user purchase portrait are from high to low. Therefore, setting the characteristic weights of user attributes, drug attributes and purchase behavior attributes as in this embodiment can realize the extraction of the personalized characteristics of the user purchase portrait from purchase behavior attributes, drug attributes and user attributes.

[0092] Constructing a historical user purchase portrait with coexisting common characteristics and personalized characteristics can more effectively perform subsequent user classification tasks and improve the classification accuracy.

[0093] As Figure 3 shown, step S2: Based on the historical user purchase portrait, similar classification is performed on chronic disease users to obtain multiple user groups to realize group management of chronic disease users. Among them, the historical user purchase portraits of all chronic disease users in the same user group have low differences, and the historical user purchase portraits of chronic disease users in different user groups have high differences;

[0094] Performing similar classification on the historical user purchase portraits to obtain multiple user groups, including:

[0095] Step 1: Quantify each historical user purchase portrait into a single user group, and sequentially perform averaging processing on all historical user purchase portraits in each user group to obtain the group portrait of each user group;

[0096] Step 2: Sequentially calculate the Euclidean distance between the group portraits in any two user groups to measure the similarity between any two user groups, fuse the two user groups corresponding to the minimum similarity, and then perform averaging processing on all historical user purchase portraits in each user group to complete the update of the group portrait. The similarity is used to measure the probability that two user groups represent the same type of historical behavior characteristics;

[0097] Step 3: Repeat Step 2 until the similarity between any two user groups is greater than the set threshold, and output the current user groups as the classification result of the historical user purchase portrait.

[0098] The calculation formula for the group portrait is:

[0099]

[0100] In the formula, G j represents the group portrait of the j-th user group, and m j represents the total number of chronic disease users in the j-th user group. j is a measurement constant without substantial meaning;

[0101] The calculation formula for the similarity is:

[0102]

[0103] In the formula, I j,k represents the similarity between the j-th user group and the k-th user group, and G j , G k represent the group portraits of the j-th and k-th user groups respectively. T is the transpose symbol, and k is a measurement constant without substantial meaning.

[0104] Based on the historical user purchase portrait, similar classification of chronic disease users is performed to obtain multiple user groups to achieve group management of chronic disease users, effectively changing the monitoring thread from setting a single monitoring thread for all chronic disease users to only setting monitoring threads for multiple user groups, greatly reducing the number of monitoring threads and releasing hardware resources.

[0105] Step S3: In each user group, monitor the current behavior characteristics of chronic disease users when purchasing drugs in real time, compare the differences in the current behavior characteristics of all chronic disease users in the user group to predict chronic disease users with abnormal drug purchase behavior in the user group, and then give an early warning feedback to the chronic disease users with abnormal drug purchase behavior to standardize the drug purchase behavior of chronic disease users.

[0106] Comparing the differences in the current behavior characteristics of all chronic disease users in the user group to predict chronic disease users with abnormal drug purchase behavior in the user group includes:

[0107] Quantify the purchase data representing the current behavior characteristics of chronic disease users into the current behavior samples in the form of characteristic fields;

[0108] Combine the characteristic fields in the current behavior samples to obtain the user real-time portrait. The function expression of the user real-time portrait is:

[0109] R i，r，ε ={fr，1，ε , f r，2，ε , f r，3，ε};

[0110] Wherein, R j,r,ε represents the real-time user profile of the r-th chronic disease user in the j-th user group at the current moment ε, and f r,1,ε , f r,2,ε , f r,3,ε respectively represent data vectors characterizing the user attributes, drug attributes, and purchase behavior attributes of the r-th chronic disease user at the current moment ε; f r , 1,ε = user r,ε , f r,2,ε = object r,ε , f r,3,ε = action r,ε .

[0111] Calculate the similarity between the real-time user profile R j,r,ε of the r-th chronic disease user in the user group and the real-time user profiles {R j,l,ε |l ≠ r ∩ l ∈ [1, m j} of all the remaining chronic disease users in the user group in turn, and count the number of similarities less than or equal to the similarity threshold;

[0112] Take the ratio of the number of similarities less than or equal to the similarity threshold to the total number m j of the chronic disease users in the j-th user group as the probability of the r-th chronic disease user having abnormal drug purchase behavior.

[0113] Comparing the differences in the existing behavior characteristics of all chronic disease users in the user group to predict the chronic disease users with abnormal drug purchase behavior in the user group further includes:

[0114] Set a probability threshold, and compare the probability of the r-th chronic disease user having abnormal drug purchase behavior with the probability threshold, wherein,

[0115] When the probability of the r-th chronic disease user having abnormal drug purchase behavior exceeds the probability threshold, it is determined that the r-th chronic disease user has abnormal drug purchase behavior;

[0116] When the probability of the r-th chronic disease user having abnormal drug purchase behavior does not exceed the probability threshold, it is determined that the r-th chronic disease user does not have abnormal drug purchase behavior.

[0117] The chronic disease anomaly detection method based on group management takes the chronic disease users in the same group as the benchmark, and judges the status of drug purchase behavior through mutual comparison, which can identify the abnormal behavior of chronic diseases at the early stage of the occurrence of abnormal drug purchase behavior of chronic disease users, and is of great significance for the drug use warning of chronic disease users to ensure the life safety of chronic disease users.

[0118] The method for detecting abnormal drug purchase behavior based on group management introduces the concept of overall management into the detection of abnormal drug purchase behavior, groups similar chronic disease users together to form each user group, which is beneficial to the long-term supervision of chronic disease users.

[0119] As Figure 2 shown, based on the above warning analysis method for big data of chronic disease medication, the present invention provides a warning system, including:

[0120] A portrait establishment unit 1 constructs a historical user purchase portrait of chronic disease users based on the drug purchase logs of chronic disease users. The historical user purchase portrait is used to embody the historical behavior characteristics of chronic disease users in purchasing drugs;

[0121] A group classification unit 2 is used to perform similarity classification on chronic disease users based on the historical user purchase portrait to obtain multiple user groups so as to realize group management of chronic disease users. Among them, the historical user purchase portraits of all chronic disease users in the same user group have low differences, and the historical user purchase portraits of chronic disease users in different user groups have high differences;

[0122] A group monitoring unit 3 is used to monitor in real time the current behavior characteristics of chronic disease users in purchasing drugs in each user group, compare the differences in the current behavior characteristics of all chronic disease users in the user group to predict the chronic disease users with abnormal drug purchase behavior in the user group, and then give a warning feedback to the chronic disease users with abnormal drug purchase behavior to standardize the drug purchase behavior of chronic disease users;

[0123] A user terminal 4 is used to receive the warning feedback from the group monitoring unit.

[0124] The group monitoring unit and the user terminal interact feedback information through a network communication protocol.

[0125] The present invention establishes a user purchase portrait for chronic disease users, and synchronously extracts the common features and personalized features in the behavior characteristics of chronic disease users when establishing the user purchase portrait, comprehensively grasps the behavior characteristics of chronic disease users, and realizes group management of chronic disease users by performing similarity classification on chronic disease users based on the historical user purchase portrait to obtain multiple user groups, reduces the monitoring threads, and improves the management efficiency.

[0126] The above embodiments are only exemplary embodiments of the present application and are not used to limit the present application. The protection scope of the present application is defined by the claims. Those skilled in the art can make various modifications or equivalent replacements within the essence and protection scope of the present application, and such modifications or equivalent replacements should also be regarded as falling within the protection scope of the present application.

Claims

1. A warning analysis method based on big data of chronic disease medications, characterized in that, it includes the following steps: Step S1: Construct a historical user purchase portrait of chronic disease users based on the drug purchase logs of chronic disease users, and the historical user purchase portrait is used to visualize the historical behavior characteristics of chronic disease users' drug purchases; Step S2: Conduct similar classification on chronic disease users based on the historical user purchase portrait to obtain multiple user groups to achieve group management of chronic disease users. Among them, the historical user purchase portraits of all chronic disease users in the same user group have low differences, and the historical user purchase portraits of chronic disease users in different user groups have high differences; Step S3: In each user group, real-time monitor the current behavior characteristics of chronic disease users when purchasing drugs, and compare the differences in the current behavior characteristics of all chronic disease users in the user group to predict chronic disease users with abnormal drug purchase behaviors in the user group, and then give a warning feedback to the chronic disease users with abnormal drug purchase behaviors to standardize the drug purchase behaviors of chronic disease users; The construction of the historical user purchase portrait of chronic disease users based on the drug purchase logs of chronic disease users includes: Extract feature fields representing user behavior characteristics from the drug purchase logs to extract the common features of the historical user purchase portrait of chronic disease users, and convert the drug purchase logs into user behavior samples based on the feature fields. The feature fields include: user attributes, drug attributes, and purchase behavior attributes; Add feature weights to the feature fields to extract the personalized features of the historical user purchase portrait of chronic disease users, and perform weight combination on the feature fields in the user behavior samples based on the feature weights to obtain the historical user purchase portrait. The function expression of the historical user purchase portrait is: F i = {[α i,1,t f i,1,t , α i,2,t f i,2,t , α i,3,t f i,3,t | t ∈ [1, n]}; Where, F i represents the historical user purchase portrait of the i-th chronic disease user, [α i,1,t f i,1,t , α i,2,t f i,2,t , α i,3, t f i,3,t represents the historical user purchase portrait of the i-th chronic disease user at the historical moment t, and f i,1,t , f i,2,t , f i,3,t respectively represent the data vectors of the user attributes, drug attributes, and purchase behavior attributes that characterize the i-th chronic disease user at the historical moment t, and α i,1,t , α i,2,t , α i,3,t respectively represent the feature weights of the user attributes, drug attributes, and purchase behavior attributes of the i-th chronic disease user at the historical moment t. i and t are measurement constants without actual meaning, and n represents the total number of historical moments; The addition of feature weights to the feature fields includes: Calculate the fluctuation degree of the data vectors representing the drug attributes and purchase behavior attributes of chronic disease users in sequence. The fluctuation degree is a quantitative index for measuring the personalized features of the historical user purchase portrait of chronic disease users. The calculation formula of the fluctuation degree is: where f i,2,t-1 and f i,3,t-1 respectively represent the data vectors characterizing the drug attributes and purchase behavior attributes of the i-th chronic disease user at the historical moment t-1; Set the feature weights of the user attributes, drug attributes, and purchase behavior attributes of chronic disease users based on the fluctuation degree. The feature weights of the user attributes, drug attributes, and purchase behavior attributes are in sequence: α i,1,t ＝1-α i,2,t -α i,3,t ； where w i,2,t and w i,3,t respectively represent the fluctuation degrees of the data vectors of the drug attribute and the purchase behavior attribute.

2. The warning analysis method based on big data of chronic disease medications according to claim 1, characterized in that: The data vectors representing the user attributes, drug attributes, and purchase behavior attributes of the i-th chronic disease user at the historical moment t are respectively composed of vectors of the user attributes, drug attributes, and purchase behavior attributes of the i-th chronic disease user in the user behavior samples from historical moment 1 to historical moment t, where, f i,1,t = {user i,τ | τ ∈ [1, t]}; f i,2,t = {object i,τ | τ ∈ [1, t]}; f i,3,t = {action i,τ | τ ∈ [1, t]}; where user i,τ 、object i,τ 、action i,τ respectively represent the data of the user attribute, drug attribute, and purchase behavior attribute of the i-th chronic disease user at the historical moment τ; τ is a measurement constant and has no substantial meaning.

3. The warning analysis method based on big data of chronic disease medications according to claim 2, characterized in that: The obtaining of multiple user groups through similar classification based on the historical user purchase portrait includes: Step 1: Quantify each historical user purchase portrait into a single user group, and sequentially perform averaging processing on all historical user purchase portraits in each user group to obtain the group portrait of each user group; Step 2: Sequentially calculate the Euclidean distance between the group portraits of any two user groups to measure the similarity between any two user groups, fuse the two user groups corresponding to the minimum similarity, and then perform averaging processing on all historical user purchase portraits in each user group to complete the update of the group portrait. The similarity is used to measure the probability that two user groups represent the same type of historical behavior characteristics; Step 3: Repeat Step 2 until the similarity between any two user groups is greater than the set threshold, and output the current user groups as the classification result of the historical user purchase portraits.

4. A warning analysis method based on big data of chronic disease medications according to claim 3, characterized in that: The calculation formula of the group portrait is: where G j represents the group portrait of the j-th user group, and m j represents the total number of chronic disease users in the j-th user group. j is a measurement constant without substantial meaning; The calculation formula of the similarity is: where I j,k represents the similarity between the j-th user group and the k-th user group, G j , G k respectively represent the group portraits of the j-th and k-th user groups, T is the transpose symbol, and k is a measurement constant with no substantial meaning.

5. A warning analysis method based on big data of chronic disease medications according to claim 4, characterized in that, Judging the chronic disease users with abnormal drug purchase behaviors in the comparison user group by comparing the differences in the existing behavior characteristics of all chronic disease users in the user group, including: Quantify the purchase data representing the existing behavior characteristics of the chronic disease users into the existing behavior samples in the form of the characteristic field representation; Combine the characteristic fields in the existing behavior samples to obtain the real-time user portrait. The function expression of the real-time user portrait is: R j,r,ε = {f r,1,ε , f r,2,ε , f r,3,ε}; where R j,r,ε represents the real-time user profile of the r-th chronic disease user in the j-th user group at the current moment ε, and f r,1,ε , f r,2,ε , f r,3,ε respectively represent data vectors characterizing the user attributes, drug attributes, and purchase behavior attributes of the r-th chronic disease user at the current moment ε; Calculate the user real-time portrait R of the r-th chronic disease user in the user group in sequence j,r,ε with the user real-time portraits {R j,l,ε |l≠r∩l∈[1,m j} of all the remaining chronic disease users in the user group, and count the number of similarities less than or equal to the similarity threshold; The ratio of the number of similarities less than or equal to the similarity threshold to the total number m of chronic disease users in the j-th user group is used as the probability of abnormal drug purchase behavior of the r-th chronic disease user. j ​ 6. A warning analysis method based on big data of chronic disease medications according to claim 5, characterized in that, Judging the chronic disease users with abnormal drug purchase behaviors in the comparison user group by comparing the differences in the existing behavior characteristics of all chronic disease users in the user group, further includes: Set a probability threshold, and compare the probability that the r-th chronic disease user has an abnormal drug purchase behavior with the probability threshold, where, When the probability that the r-th chronic disease user has an abnormal drug purchase behavior exceeds the probability threshold, it is determined that the r-th chronic disease user has an abnormal drug purchase behavior; When the probability that the r-th chronic disease user has an abnormal drug purchase behavior does not exceed the probability threshold, it is determined that the r-th chronic disease user does not have an abnormal drug purchase behavior.

7. A warning system for a warning analysis method based on big data of chronic disease medications according to any one of claims 1-6, characterized in that, including: A portrait establishment unit, which constructs a historical user purchase portrait of chronic disease users based on the drug purchase logs of chronic disease users. The historical user purchase portrait is used to visualize the historical behavior characteristics of chronic disease users' drug purchases; A group classification unit, which is used to perform similarity classification on chronic disease users based on the historical user purchase portrait to obtain multiple user groups to realize group management of chronic disease users. Among them, the historical user purchase portraits of all chronic disease users in the same user group have low differences, and the historical user purchase portraits of chronic disease users in different user groups have high differences; A population monitoring unit is used to monitor in real time the existing behavioral characteristics of chronic disease users when purchasing drugs at the current moment in each user population, compare the differences in the existing behavioral characteristics of all chronic disease users in the user population to predict chronic disease users with abnormal drug purchase behaviors in the user population, and then give early warning feedback to the chronic disease users with abnormal drug purchase behaviors to standardize the drug purchase behaviors of chronic disease users; A user terminal is used to receive the early warning feedback from the population monitoring unit.

8. An early warning system according to claim 7, characterized in that, the population monitoring unit and the user terminal interact feedback information through a network communication protocol.

Images