High-risk population health record management system and method

Abstract

The application discloses a high-risk population health record management system and method, which performs semantic understanding and analysis on the health record of a user to be evaluated through advanced data processing technology and a deep learning algorithm, and performs multi-modal interactive coupling on the information in the health record to form a comprehensive user feature representation, so that a classifier is used to realize automatic archiving, intelligently judges whether the user belongs to a high-risk population, and provides strong technical support for public health management.

Description

Technical Field

[0001] This application relates to the field of intelligent management, and more specifically, to a health record management system and method for high-risk groups. Background Technology

[0002] With societal development and an aging population, the incidence of chronic and non-communicable diseases is rising, posing a significant challenge to public health systems. Early identification and intervention of high-risk groups are crucial for the effective prevention and management of these health problems. High-risk groups are individuals with a higher risk of disease due to factors such as genetics, environment, lifestyle, or medical history. Establishing health records for high-risk groups helps healthcare institutions effectively track individuals facing higher health risks.

[0003] However, traditional methods for identifying high-risk groups often rely on limited medical resources and the judgment of professionals, which is not only inefficient but also prone to subjectivity and error. Therefore, there is a need for an optimized health record management system and method for high-risk groups. Summary of the Invention

[0004] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a health record management system and method for high-risk populations. This system utilizes advanced data processing technology and deep learning algorithms to perform semantic understanding and analysis of the health records of users to be assessed. It also performs multimodal interactive coupling of information from the health records to form a comprehensive user feature representation. This allows for automated archiving using a classifier, enabling intelligent determination of whether a user belongs to a high-risk population, thus providing strong technical support for public health management.

[0005] According to one aspect of this application, a health record management system for high-risk groups is provided, comprising:

[0006] The health record acquisition module is used to acquire the health records of the users to be assessed.

[0007] The multimodal information extraction module is used to extract basic user information, user lifestyle information, and user medical history data from the health records of the user to be evaluated.

[0008] The semantic encoding module is used to perform semantic encoding on the user's basic information, the user's lifestyle information, and the user's medical history data to obtain semantic encoding vectors for user's basic information, user's lifestyle habits, and user's medical history data.

[0009] The multimodal information interaction fusion module is used to perform interactive fusion based on the user's basic information semantic encoding vector, user's lifestyle semantic encoding vector, and user's medical history data semantic encoding vector to obtain the user's multimodal information interaction coupled representation vector.

[0010] The archive result generation module is used to determine the archive result based on the user multimodal information interaction coupling representation vector, and the archive result is used to indicate whether the user to be evaluated is a high-risk group.

[0011] According to another aspect of this application, a method for managing health records of high-risk groups is provided, comprising:

[0012] Obtain the health records of the user to be evaluated;

[0013] Extract basic user information, lifestyle information, and medical history data from the health records of the users to be evaluated.

[0014] Semantic encoding is performed on the user's basic information, the user's lifestyle information, and the user's medical history data to obtain semantic encoding vectors for user's basic information, user's lifestyle, and user's medical history data.

[0015] The semantic encoding vectors of user basic information, user lifestyle habits, and user medical history data are interactively fused based on a balanced threshold and coupling interaction weights to obtain a user multimodal information interaction coupling representation vector.

[0016] The archiving result is determined based on the user's multimodal information interaction coupling representation vector, and the archiving result is used to indicate whether the user to be evaluated is a high-risk group.

[0017] Compared with existing technologies, the health record management system and method for high-risk groups provided in this application use advanced data processing technology and deep learning algorithms to perform semantic understanding and analysis of the health records of users to be evaluated, and couple the information in the health records in a multimodal manner to form a comprehensive user feature representation. This allows for automated archiving using a classifier, enabling intelligent determination of whether a user belongs to a high-risk group, and providing strong technical support for public health management. Attached Figure Description

[0018] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0019] Figure 1 This is a block diagram of a health record management system for high-risk groups according to an embodiment of this application;

[0020] Figure 2 This is a system architecture diagram of a health record management system for high-risk populations according to an embodiment of this application;

[0021] Figure 3 This is a block diagram of the multimodal information interaction and fusion module in the high-risk population health record management system according to an embodiment of this application;

[0022] Figure 4 This is a block diagram of an adaptive fusion unit in a high-risk population health record management system according to an embodiment of this application;

[0023] Figure 5 This is a flowchart of a method for managing health records of high-risk groups according to an embodiment of this application. Detailed Implementation

[0024] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0025] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not specifically singular and may include plural forms. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0026] While this application makes various references to certain modules of the systems according to embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The modules described are merely illustrative, and different aspects of the systems and methods may use different modules.

[0027] Flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0028] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0029] Traditional methods for identifying high-risk populations often rely on limited medical resources and the judgment of professionals, which is not only inefficient but also prone to subjectivity and error. Therefore, there is a need for an optimized health record management system and method for high-risk populations.

[0030] The technical solution of this application proposes a health record management system for high-risk groups. Figure 1 This is a block diagram of a health record management system for high-risk groups according to an embodiment of this application. Figure 2 This is a system architecture diagram of a health record management system for high-risk groups according to an embodiment of this application. Figure 1 and Figure 2 As shown, the high-risk population health record management system according to an embodiment of this application includes: a health record acquisition module 310, used to acquire the health record of a user to be assessed; a multimodal information extraction module 320, used to extract basic user information, user lifestyle information, and user medical history data from the health record of the user to be assessed; a semantic encoding module 330, used to perform semantic encoding on the basic user information, the user lifestyle information, and the user medical history data to obtain semantic encoding vectors for basic user information, lifestyle, and medical history data; a multimodal information interaction fusion module 340, used to perform interaction fusion on the semantic encoding vectors for basic user information, lifestyle, and medical history data based on a balance threshold and coupling interaction weight to obtain a multimodal information interaction coupling representation vector; and an archiving result generation module 350, used to determine an archiving result based on the multimodal information interaction coupling representation vector, wherein the archiving result indicates whether the user to be assessed is a high-risk population.

[0031] Specifically, the health record acquisition module 310 and the multimodal information extraction module 320 are used to acquire the health record of the user to be assessed; and to extract basic user information, lifestyle information, and medical history data from the health record of the user to be assessed. In practical applications, health records typically contain comprehensive medical information about the user to be assessed. For example, basic user information includes age, gender, occupation, etc., which are closely related to an individual's health status and disease risk; lifestyle information includes habits such as diet, exercise, sleep, smoking, and drinking, which directly affect an individual's health status, and unhealthy habits may increase the risk of certain diseases; and medical history data includes past medical history, family medical history, chronic disease status, surgical history, etc., which helps to understand an individual's health status and disease development trends. Although health records contain comprehensive medical information about the user to be assessed, the identification of high-risk groups still relies on professional medical personnel for judgment and analysis, which may lead to inefficiency. Therefore, the technical solution of this application aims to utilize intelligent algorithms to analyze the above data to automatically determine whether the user to be assessed belongs to a high-risk group, providing support for the management of health records for high-risk groups.

[0032] Specifically, the semantic encoding module 330 is used to perform semantic encoding on the user's basic information, user's lifestyle information, and user's medical history data to obtain semantic encoding vectors for user's basic information, user's lifestyle, and user's medical history data. In the technical solution of this application, semantic encoding is performed on the user's basic information, user's lifestyle information, and user's medical history data to obtain semantic encoding vectors for user's basic information, user's lifestyle, and user's medical history data. That is, key semantic meanings are extracted from the user's basic information, user's lifestyle information, and user's medical history data through semantic encoding, and text data that is difficult to process directly is converted into a structured vector representation, which helps with subsequent model reading and analysis. Specifically, in a specific example of this application, a semantic encoder containing a Transformer layer is used to perform semantic encoding on the user's basic information, user's lifestyle information, and user's medical history data to obtain semantic encoding vectors for user's basic information, user's lifestyle, and user's medical history data. It is worth mentioning that the main feature of Transformer is that it is based on the attention mechanism, which avoids some of the limitations of traditional recurrent neural networks (RNN) and convolutional neural networks (CNN), such as long-range dependencies and parallel computing capabilities.

[0033] Specifically, the multimodal information interaction fusion module 340 is used to perform interactive fusion based on a balance threshold and coupling interaction weights on the user's basic information semantic encoding vector, user's lifestyle habit semantic encoding vector, and user's medical history data semantic encoding vector to obtain a user multimodal information interaction coupled representation vector. Specifically, in a specific example of this application, such as... Figure 3 As shown, the multimodal information interaction fusion module 340 includes: an adaptive fusion unit 341, used to input the semantic encoding vector of the user's basic information and the semantic encoding vector of the user's medical history data into the balanced threshold feature vector adaptive fusion module to obtain the semantic fusion representation vector of the user's basic information and the user's medical history; and a nonlinear coupling unit 342, used to input the semantic fusion representation vector of the user's basic information and the user's medical history data and the semantic encoding vector of the user's lifestyle habits into the nonlinear coupling interaction module to obtain the coupling representation vector of the user's multimodal information interaction.

[0034] Specifically, the adaptive fusion unit 341 is used to input the semantic encoding vector of the user's basic information and the semantic encoding vector of the user's medical history data into the balanced threshold feature vector adaptive fusion module to obtain the semantic fusion representation vector of the user's basic information and medical history. It should be understood that since a user's basic information and medical history are key factors in assessing an individual's health status, and there may be complex relationships and complementary information between the two, it is important to understand that certain genetic predispositions may not be directly reflected in the medical history, but family medical history in the basic information can provide clues in this regard. However, traditional fusion methods, such as positional addition, may ignore the importance between basic information and medical history information, resulting in the loss of important information in the fused feature representation. Therefore, in the technical solution of this application, it is desirable to input the semantic encoding vector of the user's basic information and the semantic encoding vector of the user's medical history data into the balanced threshold feature vector adaptive fusion module to obtain the semantic fusion representation vector of the user's basic information and medical history. In particular, in a specific example of this application, such as... Figure 4As shown, the adaptive fusion unit 341 includes: a dynamic fusion weight determination subunit 3411, used to determine a first dynamic fusion weight, a second dynamic fusion weight, and a third dynamic fusion weight based on the user basic information semantic encoding vector and the user medical history data semantic encoding vector; an adaptive dynamic integration subunit 3412, used to adaptively and dynamically integrate the user basic information semantic encoding vector and the user medical history data semantic encoding vector with the first dynamic fusion weight, the second dynamic fusion weight, and the third dynamic fusion weight to obtain a user basic information semantic dynamic interaction fusion item and a user medical history semantic dynamic interaction fusion item; and an element-by-element interaction fusion subunit 3413, used to perform element-by-element fusion of the user basic information semantic dynamic interaction fusion item and the user medical history semantic dynamic interaction fusion item to obtain the user basic information-user medical history semantic fusion representation vector.

[0035] More specifically, the equilibrium threshold dynamic fusion weight determination subunit 3411 is used to determine a first equilibrium threshold dynamic fusion weight, a second equilibrium threshold dynamic fusion weight, and a third equilibrium threshold dynamic fusion weight based on the user's basic information semantic encoding vector and the user's medical history data semantic encoding vector. Here, the first, second, and third equilibrium threshold dynamic fusion weights are established based on the importance of the data itself and the predictive power of the features. In this way, the equilibrium threshold dynamic fusion weights can be dynamically adjusted, which means that the model can adaptively respond to changes in the data to adjust the feature contribution from the user's basic information semantic encoding vector and the user's medical history data semantic encoding vector.

[0036] In a specific example of this application, determining the first equilibrium threshold dynamic fusion weight, the second equilibrium threshold dynamic fusion weight, and the third equilibrium threshold dynamic fusion weight based on the user basic information semantic encoding vector and the user medical history data semantic encoding vector includes: concatenating the user basic information semantic encoding vector and the user medical history data semantic encoding vector to obtain a user basic information-user medical history semantic concatenated vector; calculating the element-wise addition of the user basic information semantic encoding vector and the user medical history data semantic encoding vector to obtain a user basic information-user medical history semantic element-wise addition vector; and calculating the element-wise multiplication of the user basic information semantic encoding vector and the user medical history data semantic encoding vector. The process involves obtaining an element-wise multiplication vector of user basic information and user medical history semantics; multiplying the first transformation vector by the concatenated vector of user basic information and user medical history semantics and adding it to a first bias parameter to obtain a concatenation coefficient; multiplying the second transformation vector by the element-wise addition vector of user basic information and user medical history semantics and adding it to a second bias parameter to obtain an addition coefficient; multiplying the third transformation vector by the element-wise multiplication vector of user basic information and user medical history semantics and adding it to a third bias parameter to obtain a multiplication coefficient; and then applying the concatenation coefficient, addition coefficient, and multiplication coefficient to a sigmoid activation function to obtain the first equilibrium threshold dynamic fusion weight, the second equilibrium threshold dynamic fusion weight, and the third equilibrium threshold dynamic fusion weight.

[0037] More specifically, the adaptive dynamic integration subunit 3412 is used to adaptively and dynamically integrate the user basic information semantic encoding vector and the user medical history data semantic encoding vector with the first equilibrium threshold dynamic fusion weight, the second equilibrium threshold dynamic fusion weight, and the third equilibrium threshold dynamic fusion weight to obtain the user basic information semantic dynamic interaction fusion term and the user medical history semantic dynamic interaction fusion term. In a specific example of this application, firstly, the average value of the first equilibrium threshold dynamic fusion weight, the second equilibrium threshold dynamic fusion weight, and the third equilibrium threshold dynamic fusion weight is used as the user medical history semantic dynamic interaction fusion coefficient; then, a value minus the user medical history semantic dynamic interaction fusion coefficient is used as the user basic information semantic dynamic interaction fusion coefficient; subsequently, the dot product of the user basic information semantic dynamic interaction fusion coefficient and the user basic information semantic encoding vector is calculated to obtain the user basic information semantic dynamic interaction fusion term; and the dot product of the user medical history semantic dynamic interaction fusion coefficient and the user medical history data semantic encoding vector is calculated to obtain the user medical history semantic dynamic interaction fusion term.

[0038] More specifically, the element-by-element interactive fusion subunit 3413 is used to perform element-by-element fusion of the user basic information semantic dynamic interactive fusion item and the user medical history semantic dynamic interactive fusion item to obtain the user basic information-user medical history semantic fusion representation vector. That is, by using the first equilibrium threshold dynamic fusion weight, the second equilibrium threshold dynamic fusion weight, and the third equilibrium threshold dynamic fusion weight, the basic information semantic feature distribution and the medical history semantic feature distribution can be adaptively and dynamically integrated, and the element-by-element fusion method ensures that fine-grained information is properly considered, rather than simply treating the vector as a whole.

[0039] In summary, in the above embodiments, inputting the semantic encoding vector of the user's basic information and the semantic encoding vector of the user's medical history data into the balanced threshold feature vector adaptive fusion module to obtain the semantic fusion representation vector of the user's basic information and the user's medical history data includes: using the balanced threshold feature vector adaptive fusion module to adaptively fuse the semantic encoding vector of the user's basic information and the semantic encoding vector of the user's medical history data with the following formula to obtain the semantic fusion representation vector of the user's basic information and the user's medical history data; wherein, the formula is: ;in, and These are the semantic encoding vectors of the user's basic information and the semantic encoding vectors of the user's medical history data, respectively. Indicates cascading. This represents vector multiplication. This indicates that elements are added one by one according to their positions. This indicates element-wise multiplication by position. , and These are the first transformation vector, the second transformation vector, and the third transformation vector, respectively. , and These are the first bias parameter, the second bias parameter, and the third bias parameter, respectively. This represents the sigmoid activation function. , and These are the first equilibrium threshold dynamic fusion weight, the second equilibrium threshold dynamic fusion weight, and the third equilibrium threshold dynamic fusion weight. , It is a semantic fusion representation vector of user basic information and user medical history.

[0040] It is worth mentioning that, in other specific examples of this application, the semantic encoding vector of the user's basic information and the semantic encoding vector of the user's medical history data can also be input into the adaptive fusion module of the equilibrium threshold feature vector in other ways to obtain the semantic fusion representation vector of the user's basic information and the user's medical history data. For example: input the semantic encoding vector of the user's basic information and the semantic encoding vector of the user's medical history data; define an equilibrium threshold feature vector, which is used to control the fusion weight between the basic information and the medical history data; use the adaptive fusion module to fuse the semantic encoding vector of the basic information and the semantic encoding vector of the medical history data; wherein, the adaptive fusion module may include a neural network layer, an attention mechanism or other model components to learn how to effectively fuse the two types of information; according to the setting in the equilibrium threshold feature vector, the module will automatically adjust the fusion ratio of the two types of information; in the adaptive fusion module, the semantic encoding vectors of the basic information and the medical history data are fused to obtain the semantic fusion representation vector of the user's basic information and the user's medical history data.

[0041] Specifically, the nonlinear coupling unit 342 is used to input the semantic fusion representation vector of user basic information-user medical history and the semantic encoding vector of user lifestyle habits into the nonlinear coupling interaction module to obtain the user multimodal information interaction coupling representation vector. In particular, during data processing, the nonlinear coupling interaction module introduces quadratic term operations to achieve nonlinear optimization of the data distribution, thereby better fitting the complex interaction between the semantic fusion features of user basic information-user medical history and the semantic features of user lifestyle habits. Simultaneously, coupling interaction weights are applied to adjust the relative importance of different features in the nonlinear transformation interaction, balancing the scale of different feature values ​​and achieving dynamic adjustment and adaptive change of feature distribution. In this way, the user multimodal information interaction coupling representation vector can comprehensively understand the complex semantic interaction relationships between user basic information, user medical history information, and lifestyle habit information, forming the user's comprehensive health characteristics.

[0042] In a specific example of this application, the semantic fusion representation vector of user basic information-user medical history and the semantic encoding vector of user lifestyle habits are input into a nonlinear coupling interaction module to obtain the user multimodal information interaction coupling representation vector. This includes: calculating the positional response of the semantic fusion representation vector of user basic information-user medical history relative to the semantic encoding vector of user lifestyle habits to obtain the user multimodal information association response interaction feature vector; and performing nonlinear coupling optimization on the user multimodal information association response interaction feature vector to obtain the user multimodal information interaction coupling representation vector.

[0043] More specifically, in the embodiments of this application, the specific implementation process of calculating the positional response of the user basic information-user medical history semantic fusion representation vector relative to the user lifestyle semantic encoding vector to obtain the user multimodal information association response interaction feature vector is as follows: dividing the i-th feature value of the user basic information-user medical history semantic fusion representation vector by the i-th feature value of the user lifestyle semantic encoding vector to obtain the i-th feature value of the user multimodal information association response interaction feature vector. The specific implementation process of performing nonlinear coupling optimization on the user multimodal information association response interaction feature vector to obtain the user multimodal information interaction coupled representation vector is as follows: determining the first coupling interaction weight, the second coupling interaction weight, the third coupling interaction weight, and the fourth coupling interaction weight; calculating the positional response of the user multimodal information association response interaction feature vector... The first coupled interaction quadratic term of the first coupled interaction quadratic term, the quadratic coefficient of the first coupled interaction quadratic term is the first coupled interaction weight, the first coefficient of the first coupled interaction quadratic term is the second coupled interaction weight, and the constant coefficient of the first coupled interaction quadratic term is zero; calculate the first eigenvalue in the user multimodal information associated response interaction feature vector. The second coupled interaction quadratic term of the eigenvalues, the quadratic coefficient of the second coupled interaction quadratic term is the third coupled interaction weight, the first coefficient of the second coupled interaction quadratic term is the fourth coupled interaction weight, and the constant coefficient of the second coupled interaction quadratic term is one; the third... The first coupled interaction quadratic term of the eigenvalue is divided by the i-th eigenvalue. The second coupling interaction quadratic term of the eigenvalue is used to obtain the th feature value of the user multimodal information interaction coupling representation vector. Each feature value.

[0044] In summary, in the above embodiments, inputting the user basic information-user medical history semantic fusion representation vector and the user lifestyle habit semantic encoding vector into the nonlinear coupling interaction module to obtain the user multimodal information interaction coupling representation vector includes: using the nonlinear coupling interaction module to perform nonlinear coupling interaction on the user basic information-user medical history semantic fusion representation vector and the user lifestyle habit semantic encoding vector using the following formula to obtain the user multimodal information interaction coupling representation vector; wherein, the formula is: ;in, and These are the semantic fusion representation vector of the user's basic information and medical history, and the semantic encoding vector of the user's lifestyle habits, respectively. 1 eigenvalue, , , and These are the first coupling interaction weight, the second coupling interaction weight, the third coupling interaction weight, and the fourth coupling interaction weight. It is the first in the user multimodal information interaction coupling representation vector. Each feature value.

[0045] It is worth mentioning that, in other specific examples of this application, the semantic encoding vectors of user basic information, user lifestyle habits, and user medical history data can also be interactively fused based on a balance threshold and coupling interaction weights to obtain a user multimodal information interaction coupling representation vector. For example: input the semantic encoding vectors of user basic information, user lifestyle habits, and user medical history data; define coupling interaction weights for interaction fusion, which can be learned parameters used to control the importance between different information sources; set a balance threshold, which can be a threshold value used to control the intensity of information interaction. Interaction fusion will only be performed when the similarity between two information sources exceeds the threshold; perform interaction fusion on the semantic encoding vectors of user basic information, user lifestyle habits, and user medical history data; for each pair of vectors, calculate their similarity or correlation, and perform interaction fusion according to the coupling interaction weights and the balance threshold; integrate the vectors after interaction fusion to obtain the user multimodal information interaction coupling representation vector.

[0046] Specifically, the archiving result generation module 350 is used to determine the archiving result based on the user multimodal information interaction coupling representation vector, wherein the archiving result is used to indicate whether the user to be evaluated is a high-risk group. That is, in the technical solution of this application, the user multimodal information interaction coupling representation vector is input into a classifier-based automatic archivist to obtain the archiving result, wherein the archiving result is used to indicate whether the user to be evaluated is a high-risk group.

[0047] Preferably, inputting the user multimodal information interaction coupling representation vector into a classifier-based automatic archiver to obtain the archive result includes:

[0048] The user multimodal information interaction coupling representation vector is multiplied by the length of the user multimodal information interaction coupling representation vector and the square root of the length of the user multimodal information interaction coupling representation vector to obtain the user multimodal information interaction coupling full-width representation vector and the user multimodal information interaction coupling half-width representation vector.

[0049] The full-scale semantic change vector of the user multimodal information interaction coupling is obtained by subtracting the norm of the user multimodal information interaction coupling vector by a dot, and the square root of the absolute value of each position of the subtraction result vector is calculated.

[0050] The user multimodal information interaction coupling half-amplitude representation vector is subtracted from the norm of the user multimodal information interaction coupling representation vector by a dot, and the square root of the absolute value of each position of the dot subtraction result vector is calculated to obtain the user multimodal information interaction coupling half-amplitude semantic change vector.

[0051] The base-2 logarithm of each feature value of the user multimodal information interaction coupling full-amplitude semantic change vector and the user multimodal information interaction coupling half-amplitude semantic change vector are calculated to obtain the user multimodal information interaction coupling full-amplitude semantic change information vector and the user multimodal information interaction coupling half-amplitude semantic change information vector, respectively.

[0052] The weighted sum of the full-amplitude semantic change information vector of the user multimodal information interaction coupling and the half-amplitude semantic change information vector of the user multimodal information interaction coupling is calculated using the balanced hyperparameter as weight to obtain the optimized user multimodal information interaction coupling representation vector;

[0053] The optimized user multimodal information interaction coupling representation vector is input into a classifier-based automatic archiver to obtain the archive results.

[0054] Here, due to the semantic differences in the source semantics of the user's basic information, user's lifestyle habits, and user's medical history data, the semantic features of the encoded vectors differ. After passing through the balanced threshold feature vector adaptive fusion module and the nonlinear coupling interaction module, the user's multimodal information interaction coupling representation vector will also suffer from insufficient aggregation of fusion-interaction semantic feature distribution due to the differences in the balanced threshold adaptive fusion weights and the nonlinear coupling interaction weights caused by the semantic differences in features. This affects the classification convergence efficiency of the classifier, that is, the efficiency of classification training and the accuracy of classification results.

[0055] Based on this, the applicant of this application uses the user multimodal information interaction coupling representation vector as a feature set. Under the condition of semantic change representation based on the feature values ​​of the user multimodal information interaction coupling representation vector, the applicant dynamically aggregates the semantic set composed of different semantic changes of the user multimodal information interaction coupling representation vector without ignoring individual semantic change information. Thus, the set expression of the individual features of the user multimodal information interaction coupling representation vector and the aggregated scale of the user multimodal information interaction coupling representation vector is used as the full amplitude and half amplitude of the features. Furthermore, the low-rank negative correlation of different dimensions of the overall semantics of the feature set of the user multimodal information interaction coupling representation vector is used as the phase and scaled to dynamically adjust the semantic content change relationship of the user multimodal information interaction coupling representation vector. This improves the overall semantic information expression aggregation of the feature set of the user multimodal information interaction coupling representation vector, thereby improving the classification convergence efficiency when the user multimodal information interaction coupling representation vector is classified by a classifier-based automatic archiver.

[0056] As described above, the high-risk population health record management system 300 according to the embodiments of this application can be implemented in various wireless terminals, such as servers with high-risk population health record management algorithms. In one possible implementation, the high-risk population health record management system 300 according to the embodiments of this application can be integrated into the wireless terminal as a software module and / or hardware module. For example, the high-risk population health record management system 300 can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the high-risk population health record management system 300 can also be one of many hardware modules of the wireless terminal.

[0057] Alternatively, in another example, the high-risk population health record management system 300 and the wireless terminal can also be separate devices, and the high-risk population health record management system 300 can be connected to the wireless terminal via wired and / or wireless networks, and transmit interactive information in accordance with an agreed data format.

[0058] Furthermore, a method for managing health records of high-risk groups is also provided.

[0059] Figure 5 This is a flowchart of a method for managing health records of high-risk groups according to an embodiment of this application. Figure 5As shown, the high-risk population health record management method according to an embodiment of this application includes the following steps: S1, obtaining the health record of the user to be assessed; S2, extracting basic user information, user lifestyle information, and user medical history data from the health record of the user to be assessed; S3, performing semantic encoding on the basic user information, the user lifestyle information, and the user medical history data to obtain semantic encoding vectors for basic user information, lifestyle, and medical history data; S4, performing interactive fusion on the semantic encoding vectors for basic user information, lifestyle, and medical history data based on a balance threshold and coupling interaction weight to obtain a multimodal information interaction coupling representation vector; S5, determining the archiving result based on the multimodal information interaction coupling representation vector, wherein the archiving result is used to indicate whether the user to be assessed is a high-risk population.

[0060] In summary, the high-risk population health record management method according to the embodiments of this application is explained. It uses advanced data processing technology and deep learning algorithms to perform semantic understanding and analysis on the health records of users to be evaluated, and performs multimodal interactive coupling of information in the health records to form a comprehensive user feature representation. In this way, a classifier is used to achieve automated archiving, so as to intelligently determine whether the user belongs to a high-risk population, providing strong technical support for public health management.

[0061] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A health record management system for high-risk groups, characterized in that, include: The health record acquisition module is used to acquire the health records of the users to be assessed. The multimodal information extraction module is used to extract basic user information, user lifestyle information, and user medical history data from the health records of the user to be evaluated. The semantic encoding module is used to perform semantic encoding on the user's basic information, the user's lifestyle information, and the user's medical history data to obtain semantic encoding vectors for user's basic information, user's lifestyle habits, and user's medical history data. The multimodal information interaction fusion module is used to perform interactive fusion based on the user's basic information semantic encoding vector, user's lifestyle semantic encoding vector, and user's medical history data semantic encoding vector to obtain the user's multimodal information interaction coupled representation vector. The archive result generation module is used to determine the archive result based on the user multimodal information interaction coupling representation vector, and the archive result is used to indicate whether the user to be evaluated is a high-risk group; The multimodal information interaction and fusion module includes: An adaptive fusion unit is used to input the semantic encoding vector of the user's basic information and the semantic encoding vector of the user's medical history data into the balanced threshold feature vector adaptive fusion module to obtain the semantic fusion representation vector of user's basic information and user's medical history. A nonlinear coupling unit is used to input the semantic fusion representation vector of the user's basic information-user's medical history and the semantic encoding vector of the user's lifestyle habits into the nonlinear coupling interaction module to obtain the user's multimodal information interaction coupling representation vector. In the data processing process, the nonlinear coupling interaction module introduces the operation of quadratic terms to realize the nonlinear optimization of data distribution, and applies coupling interaction weights to adjust the relative importance of different features in the nonlinear transformation interaction, so as to balance the scale of different feature values ​​and realize the dynamic adjustment and adaptive change of feature distribution. The adaptive fusion unit includes: a dynamic fusion weight determination subunit with equalization threshold and an adaptive dynamic integration subunit; The equilibrium threshold dynamic fusion weight determination subunit is used for: concatenating the user basic information semantic encoding vector and the user medical history data semantic encoding vector to obtain a user basic information-user medical history semantic concatenated vector; calculating the element-by-element addition of the user basic information semantic encoding vector and the user medical history data semantic encoding vector to obtain a user basic information-user medical history semantic element-by-element vector; calculating the element-by-element multiplication of the user basic information semantic encoding vector and the user medical history data semantic encoding vector to obtain a user basic information-user medical history semantic element-by-element multiplication vector; and performing the first transformation. The concatenated vector of user basic information and user medical history semantics is multiplied by the first bias parameter to obtain the concatenation coefficient. The element-wise summation vector of user basic information and user medical history semantics is multiplied by the second transformation vector and then added to the second bias parameter to obtain the summation coefficient. The element-wise multiplication vector of user basic information and user medical history semantics is multiplied by the third transformation vector and then added to the third bias parameter to obtain the multiplication coefficient. The concatenation coefficient, summation coefficient, and multiplication coefficient are then applied to the sigmoid activation function to obtain the first equilibrium threshold dynamic fusion weight, the second equilibrium threshold dynamic fusion weight, and the third equilibrium threshold dynamic fusion weight. An adaptive dynamic integration subunit is configured to: use the average of the first equilibrium threshold dynamic fusion weight, the second equilibrium threshold dynamic fusion weight, and the third equilibrium threshold dynamic fusion weight as the user's medical history semantic dynamic interaction fusion coefficient; use a value minus the user's medical history semantic dynamic interaction fusion coefficient as the user's basic information semantic dynamic interaction fusion coefficient; calculate the dot product of the user's basic information semantic dynamic interaction fusion coefficient and the user's basic information semantic encoding vector to obtain the user's basic information semantic dynamic interaction fusion term; and calculate the dot product of the user's medical history semantic dynamic interaction fusion coefficient and the user's medical history data semantic encoding vector to obtain the user's medical history semantic dynamic interaction fusion term.

2. The high-risk population health record management system according to claim 1, characterized in that, The semantic encoding module is used for: A semantic encoder containing a Transformer layer is used to semantically encode the user's basic information, the user's lifestyle information, and the user's medical history data to obtain semantic encoding vectors for the user's basic information, lifestyle, and medical history data.

3. The high-risk population health record management system according to claim 2, characterized in that, The adaptive fusion unit further includes: The element-by-element interactive fusion subunit is used to perform element-by-element fusion of the user basic information semantic dynamic interactive fusion item and the user medical history semantic dynamic interactive fusion item to obtain the user basic information-user medical history semantic fusion representation vector.

4. The high-risk population health record management system according to claim 3, characterized in that, The nonlinear coupling unit includes: The position-wise associated response interaction subunit is used to calculate the position-wise response of the user basic information-user medical history semantic fusion representation vector relative to the user lifestyle semantic encoding vector to obtain the user multimodal information associated response interaction feature vector. The multimodal information interaction coupling subunit is used to perform nonlinear coupling optimization on the user multimodal information association response interaction feature vector to obtain the user multimodal information interaction coupling representation vector.

5. The high-risk population health record management system according to claim 4, characterized in that, The multimodal information interaction coupling subunit is used for: Determine the first coupling interaction weight, the second coupling interaction weight, the third coupling interaction weight, and the fourth coupling interaction weight; Calculate the first feature vector in the user multimodal information association response interaction feature vector. The first coupled interaction quadratic term of the first eigenvalues, the quadratic coefficient of the first coupled interaction quadratic term is the first coupled interaction weight, the first coefficient of the first coupled interaction quadratic term is the second coupled interaction weight, and the constant coefficient of the first coupled interaction quadratic term is zero. Calculate the first feature vector in the user multimodal information association response interaction feature vector. The second coupled interaction quadratic term of the eigenvalues, the quadratic coefficient of the second coupled interaction quadratic term is the third coupled interaction weight, the first coefficient of the second coupled interaction quadratic term is the fourth coupled interaction weight, and the constant coefficient of the second coupled interaction quadratic term is one. The first The first coupled interaction quadratic term of the eigenvalue is divided by the i-th eigenvalue. The second coupling interaction quadratic term of the eigenvalue is used to obtain the th feature value of the user multimodal information interaction coupling representation vector. Each feature value.

6. The high-risk population health record management system according to claim 5, characterized in that, The archiving result generation module is used for: The user multimodal information interaction coupling representation vector is input into a classifier-based automatic archiver to obtain the archive result.

7. A method for managing health records of high-risk groups, using the high-risk group health record management system as described in claim 1, characterized in that, include: Obtain the health records of the user to be evaluated; Extract basic user information, lifestyle information, and medical history data from the health records of the users to be evaluated. Semantic encoding is performed on the user's basic information, the user's lifestyle information, and the user's medical history data to obtain semantic encoding vectors for user's basic information, user's lifestyle, and user's medical history data. The semantic encoding vectors of user basic information, user lifestyle habits, and user medical history data are interactively fused based on a balanced threshold and coupling interaction weights to obtain a user multimodal information interaction coupling representation vector. The archiving result is determined based on the user's multimodal information interaction coupling representation vector, and the archiving result is used to indicate whether the user to be evaluated is a high-risk group.

Images