Personalized health intervention opinion recommendation method and system based on double attention mechanism
By generating personalized health intervention opinions through a dual attention mechanism and deep learning algorithms, this approach addresses the issue that vertical industry models in the healthcare field fail to consider individual differences, thereby achieving greater operability and accuracy of health intervention opinions.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- JIANGSU FULIXI INTELLIGENT TECH CO LTD
- Filing Date
- 2024-12-05
- Publication Date
- 2026-06-05
AI Technical Summary
Large-scale vertical industry models in the healthcare field have failed to adequately consider individual differences, resulting in low operability of health intervention recommendations.
A personalized health intervention recommendation method based on a dual attention mechanism is adopted. By focusing on the user profile and initial personalized health suggestions through the global and local attention modules of the dual attention mechanism, and combining deep learning algorithms and classifier models, personalized health intervention opinions are generated.
It provides health intervention recommendations that are highly actionable, better suited to individual needs, and improve the accuracy and operability of health intervention opinions.
Smart Images
Figure CN122158153A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of intelligent health intervention technology, specifically relating to a personalized health intervention recommendation method and system based on a dual attention mechanism. Background Technology
[0002] With the development of artificial intelligence technology, various large-scale models have experienced explosive growth. While general-purpose large-scale models, due to their broad knowledge base, are applicable to multiple fields, their performance in certain specialized areas, particularly healthcare, is not as good as that of vertical industry models. Therefore, various industries have launched vertical industry-specific large-scale models, incorporating expert experience and industry knowledge, resulting in superior performance in specific domains, especially demonstrating higher accuracy in specific tasks.
[0003] However, the healthcare industry deals with a population with vastly different individual characteristics, and large-scale industry models often fail to adequately consider these individual differences. Therefore, although the health intervention recommendations they provide may align with common medical knowledge, their practical operability is weak, resulting in low individual implementation rates. Summary of the Invention
[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a personalized health intervention recommendation method and system based on a dual attention mechanism. This method not only inherits the knowledge accuracy of vertical industry models but also pays full attention to individual differences, providing health intervention recommendations with strong executable capabilities.
[0005] To achieve the above objectives, the present invention is implemented using the following technical solution:
[0006] On the one hand, this invention provides a personalized health intervention recommendation method based on a dual attention mechanism, comprising the following steps:
[0007] The system utilizes preprocessed full user data to generate corresponding user profiles and obtains initial personalized health recommendations based on a large-scale health intervention industry model.
[0008] Two global and local attention modules using a dual attention mechanism are used to extract features by separately focusing on the user profile and the initial personalized health suggestions through adaptive weight parameters.
[0009] The initial personalized health suggestions and user profiles, which have undergone weighting and feature extraction, are transformed and combined using a classifier model. Based on the similarity of feature vectors, the probability distribution of a series of personalized health intervention recommendations is determined, and the health intervention recommendation results are determined based on the probability distribution.
[0010] Furthermore, the method for generating corresponding user profiles includes the following steps:
[0011] Acquire basic user indicator data, including blood glucose, uric acid, and blood pressure, and complete the standardization processing of the basic user indicator data, including data cleaning, extraction, and fusion, to form basic user standard data;
[0012] By combining users' health history data, including behavioral habits, physical defects, historical medical conditions, current diseases, drug allergies, types of Chinese and Western medicines, and medical payment records, the standardized user basic data is tagged to obtain a user profile that reflects an individual's health status.
[0013] Furthermore, the method for obtaining initial personalized health recommendations based on a large-scale industry model for health interventions includes the following steps:
[0014] Based on the general basic large language model, a medical basic large language model is formed by pre-training and fine-tuning authoritative knowledge databases, medical report databases, and medical prediction knowledge bases.
[0015] By employing single-round medical dialogue, multi-round medical dialogue, and multi-task medical methods, a large-scale medical reasoning model is formed through supervised fine-tuning of the data model based on medical data.
[0016] By using deep learning algorithms to analyze user feedback information to obtain user intent, and inputting the user intent into a large medical reasoning model, a series of personalized health recommendations based on medical industry knowledge and a large health intervention industry model are obtained.
[0017] Furthermore, the dual attention mechanism module learns to dynamically adjust and weight data and feature channels in different directions through adaptive weight parameters, including the following steps:
[0018] The input sequence, which includes user profile data and initial personalized health suggestions, is preprocessed and converted into a format that the model can process, including word vectors and feature maps.
[0019] In the first-layer attention mechanism, the weight of each element in the input sequence is calculated. The weights are normalized by calculating the similarity or correlation between elements and using functions such as softmax.
[0020] The input sequence is weighted and summed or weighted averaged based on the weights calculated by the first-layer attention mechanism to obtain a feature representation containing key information, which will be used as the input of the second-layer attention mechanism.
[0021] The feature representations obtained from the first attention mechanism are processed in the second attention mechanism. By calculating the correlation or importance between different features, the features are weighted again to extract and integrate key information.
[0022] The feature representation processed by the second-layer attention mechanism is used as the final output of the dual-attention mechanism model.
[0023] Furthermore, the classifier includes fully connected layers, Softmax classifiers, support vector machines, and decision trees or ensemble learning methods;
[0024] The fully connected layer is connected after the output of the feature extraction network, and is responsible for transforming and combining the extracted user intent features and the initial personalized health interventions obtained after processing by the dual attention mechanism, so as to determine the probability distribution of whether a series of health intervention opinions should be recommended.
[0025] The Softmax classifier transforms the output of the fully connected layer into a probability distribution of whether or not a recommendation should be made, thereby determining whether a health intervention recommendation should be made.
[0026] The support vector machine is used as a classifier to classify the extracted features;
[0027] The decision tree or ensemble learning method uses decision trees, random forests, gradient boosting trees, and ensemble learning methods as classifiers to determine whether to recommend health interventions.
[0028] On the other hand, the present invention provides a personalized health intervention recommendation system based on a dual attention mechanism, comprising the following modules:
[0029] The data input and processing module is used to generate corresponding user profiles using preprocessed full user data, and to obtain initial personalized health recommendations based on a large-scale health intervention industry model.
[0030] The dual attention mechanism module is used to extract features by utilizing two global and local attention modules of the dual attention mechanism, which separately focus on the user profile and the initial personalized health suggestions through adaptive weight parameters.
[0031] The classifier recommendation module is used to transform and combine the initial personalized health suggestions and user profiles after weighting and feature extraction using a classifier model. It determines the probability distribution of a series of personalized health intervention recommendations based on feature vector similarity and determines the health intervention recommendation results based on the probability distribution.
[0032] Furthermore, the data input and processing module includes a profile generation unit, which is used to acquire user basic indicator data, including blood glucose, uric acid, and blood pressure, and to complete the standardization processing of user basic indicator data, including data cleaning, extraction, and fusion, to form user basic standard data.
[0033] In addition, by combining users' health history data, including behavioral habits, physical defects, historical medical conditions, current diseases, drug allergies, types of Chinese and Western medicines, and medical payment records, the standardized user basic data is tagged to obtain a user profile that reflects an individual's health status.
[0034] Furthermore, the data input and processing module includes an initial suggestion generation unit, which is used to pre-train and fine-tune the authoritative knowledge database, medical report database, and medical prediction knowledge base based on the general basic large language model to form a medical basic large language model;
[0035] And by employing single-round medical dialogue, multi-round medical dialogue, and multi-task medical methods, we supervise and fine-tune the data model based on medical data to form a large medical reasoning model;
[0036] Furthermore, it utilizes deep learning algorithms to analyze user feedback to obtain user intent, and inputs this intent into a large-scale medical reasoning model to obtain a series of personalized health recommendations based on medical industry knowledge and a large-scale health intervention model.
[0037] Furthermore, the dual attention mechanism module includes a preprocessing unit, a weight normalization unit, a feature representation unit, an extraction and integration unit, and an output unit;
[0038] The preprocessing unit is used to preprocess the input sequence, which includes user profile data and initial personalized health suggestions, and convert it into a format that the model can process, including word vectors and feature maps.
[0039] The weight normalization unit is used to calculate the weight of each element in the input sequence in the first-layer attention mechanism. By calculating the similarity or correlation between elements, the weights are normalized using functions such as softmax.
[0040] The feature representation unit is used to perform a weighted summation or weighted average on the input sequence based on the weights calculated by the first-layer attention mechanism to obtain a feature representation containing key information, which will be used as the input of the second-layer attention mechanism.
[0041] The extraction and integration unit is used to process the feature representation obtained by the first attention mechanism in the second attention mechanism. By calculating the correlation or importance between different features, it performs weighted processing again to extract and integrate key information.
[0042] The output unit is used as the final output of the dual attention mechanism model based on the feature representation processed by the second-layer attention mechanism.
[0043] Furthermore, the classifier includes fully connected layers, Softmax classifiers, support vector machines, and decision trees or ensemble learning methods;
[0044] The fully connected layer is connected after the output of the feature extraction network, and is responsible for transforming and combining the extracted user intent features and the initial personalized health interventions obtained after processing by the dual attention mechanism, so as to determine the probability distribution of whether a series of health intervention opinions should be recommended.
[0045] The Softmax classifier transforms the output of the fully connected layer into a probability distribution of whether or not a recommendation should be made, thereby determining whether a health intervention recommendation should be made.
[0046] The support vector machine is used as a classifier to classify the extracted features;
[0047] The decision tree or ensemble learning method uses decision trees, random forests, gradient boosting trees, and ensemble learning methods as classifiers to determine whether to recommend health interventions.
[0048] Compared with existing technologies, the beneficial effects achieved by this invention are as follows: The personalized health intervention recommendation method and system based on the dual attention mechanism provided by this invention not only inherits the knowledge accuracy of vertical industry models, but also pays full attention to individual differences, providing health intervention suggestions with strong executable capabilities; by constructing a personalized health intervention recommendation method based on the dual attention mechanism based on user personal data, the model used by each user can better understand and adapt to their personal needs, thereby providing more accurate and personalized health intervention opinions. Attached Figure Description
[0049] Figure 1 This is a flowchart illustrating a personalized health intervention recommendation method based on a dual attention mechanism provided according to an embodiment of the present invention.
[0050] Figure 2 A flowchart of a large-scale medical reasoning model provided according to an embodiment of the present invention.
[0051] Figure 3 A block diagram of a classifier provided according to an embodiment of the present invention.
[0052] Figure 4 This is a block diagram of a personalized health intervention recommendation system based on a dual attention mechanism, provided according to an embodiment of the present invention. Detailed Implementation
[0053] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.
[0054] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.
[0055] like Figure 1 As shown, this embodiment of the invention provides a personalized health intervention recommendation method based on a dual attention mechanism, including the following steps:
[0056] The system utilizes preprocessed full user data to generate corresponding user profiles and obtains initial personalized health recommendations based on a large-scale health intervention industry model.
[0057] Two global and local attention modules using a dual attention mechanism are used to extract features by separately focusing on user profiles and initial personalized health suggestions through adaptive weight parameters.
[0058] The initial personalized health suggestions and user profiles, which have undergone weighting and feature extraction, are transformed and combined using a classifier model. Based on the similarity of feature vectors, the probability distribution of a series of personalized health intervention recommendations is determined, and the health intervention recommendation results are determined based on the probability distribution.
[0059] In this embodiment, the method for generating the corresponding user profile includes the following steps:
[0060] Acquire basic user indicator data, including blood glucose, uric acid, and blood pressure, and complete the standardization processing of the basic user indicator data, including data cleaning, extraction, and fusion, to form basic user standard data;
[0061] By combining users' health history data, including behavioral habits, physical defects, medical history, current illnesses, drug allergies, types of Chinese and Western medicines, and medical payment records, the standardized user baseline data is tagged to obtain user profiles reflecting individual health conditions. Specifically, user health history data mainly considers factors such as whether the user has hypertension, physical defects, medical history, drug allergies, preference for Chinese or Western medicine, preference for imported or domestic medicines, spending power, and disability status, among other personal profile data.
[0062] A large-scale health intervention industry model is trained by aggregating authoritative multimodal medical knowledge, constructing a knowledge graph based on the original data in the knowledge base, forming medical knowledge information triplets, and completing the construction of entity relationships and attributes. In this embodiment, the method for obtaining initial personalized health recommendations based on the large-scale health intervention industry model includes the following steps:
[0063] Based on a general-purpose basic large language model, a medical basic large language model is formed by pre-training and fine-tuning authoritative knowledge databases, medical report databases, and medical prediction knowledge bases. Figure 2 As shown.
[0064] By employing single-round medical dialogues, multi-round medical dialogues, and multi-task medical approaches, a large-scale medical reasoning model is formed through supervised fine-tuning of the data model based on medical data.
[0065] By using deep learning algorithms to analyze user feedback information to obtain user intent, and inputting the user intent into a large medical reasoning model, a series of personalized health recommendations based on medical industry knowledge and a large health intervention industry model are obtained.
[0066] The method of analyzing user question feedback information using deep learning algorithm models to obtain user intent includes the following steps:
[0067] The problem text needs to be cleaned by removing irrelevant characters and punctuation marks, and then segmented into individual lexical units. Simultaneously, grammatical analysis is required to determine the grammatical roles and relationships of the words within the sentence.
[0068] Extract features that represent user intent from the preprocessed text. These features may include keywords, phrases, grammatical structures, etc.
[0069] Deep learning algorithms are used to analyze extracted features and predict the intent behind user questions. This typically involves machine learning or deep learning algorithms, such as support vector machines and neural networks.
[0070] The system outputs one or more possible intent categories, each corresponding to an action or information the user might want to perform. Based on the determination, the system selects the appropriate processing flow or component to respond to the user's query, and also provides a user feedback module to indicate the accuracy of the intent determination.
[0071] Dual attention is an attention mechanism in deep learning models that uses two different attention mechanisms to process input data. This helps the model better capture key information in the input data and improves the model's performance when handling complex tasks.
[0072] In dual-attention mechanisms, two attention mechanisms are typically used to process different parts of the input data separately. For example, one attention mechanism can focus on local information in the input data, while the other attention mechanism can focus on global information in the input data. This allows the model to understand the input data more comprehensively and handle complex tasks better.
[0073] The dual attention mechanism consists of one attention mechanism that focuses on the initial health intervention opinions and suggestions generated by the large model of the health intervention industry, and another attention mechanism that focuses on the user's health history data, including behavioral habits, physical defects, historical medical conditions, current diseases, drug allergies, types of Chinese and Western medicines, medical payment records, etc., as well as user profiles.
[0074] The dual-attention mechanism module learns through adaptive weight parameters to dynamically adjust and weight data and feature channels in different directions. This results in a series of health intervention recommendations, such as considering whether the user has a disability or is suitable for strenuous exercise when recommending exercise; considering the user's consumption habits and preferences for Chinese and Western medicine when recommending medication; and considering the differences in diet between the North and South when recommending diet.
[0075] In this embodiment, the dual attention mechanism module learns to dynamically adjust and weight data and feature channels in different directions through adaptive weight parameters, including the following steps:
[0076] The input sequence, which includes user profile data and initial personalized health suggestions, is preprocessed and converted into a format that the model can process, including word vectors and feature maps.
[0077] In the first-layer attention mechanism, the weight of each element in the input sequence is calculated. The weights are normalized by calculating the similarity or correlation between elements and using functions such as softmax.
[0078] The input sequence is weighted and summed or weighted averaged based on the weights calculated by the first-layer attention mechanism to obtain a feature representation containing key information, which will be used as the input of the second-layer attention mechanism.
[0079] The feature representations obtained from the first attention mechanism are processed in the second attention mechanism. By calculating the correlation or importance between different features, the features are weighted again to extract and integrate key information.
[0080] The feature representation processed by the second-layer attention mechanism is used as the final output of the dual-attention mechanism model.
[0081] In this embodiment, as Figure 3 As shown, the classifier includes fully connected layers, Softmax classifiers, support vector machines, and decision trees or ensemble learning methods;
[0082] The fully connected layer is connected after the output of the feature extraction network. It is responsible for transforming and combining the extracted user intent features and the initial personalized health interventions obtained after processing by the dual attention mechanism to determine the probability distribution of whether a series of health intervention opinions should be recommended.
[0083] The Softmax classifier transforms the output of the fully connected layer into a probability distribution of whether or not a recommendation should be made, thus determining whether a health intervention recommendation should be made.
[0084] Support Vector Machines (SVMs) are used as classifiers to classify extracted features, helping to effectively handle high-dimensional feature spaces and possessing strong generalization capabilities.
[0085] Decision trees or ensemble learning methods refer to the use of decision trees, random forests, gradient boosting trees, and ensemble learning methods as classifiers to determine whether to recommend health interventions.
[0086] On the other hand, such as Figure 4 As shown, this embodiment of the invention also provides a personalized health intervention recommendation system based on a dual attention mechanism, including the following modules:
[0087] The data input and processing module is used to generate corresponding user profiles using preprocessed full user data, and to obtain initial personalized health recommendations based on a large-scale health intervention industry model.
[0088] The dual attention mechanism module is used to extract features by utilizing two global and local attention modules of the dual attention mechanism, which separately focus on the user profile and the initial personalized health suggestions through adaptive weight parameters.
[0089] The classifier recommendation module is used to transform and combine the initial personalized health suggestions and user profiles after weighting and feature extraction using a classifier model. It determines the probability distribution of a series of personalized health intervention recommendations based on feature vector similarity and determines the health intervention recommendation results based on the probability distribution.
[0090] The data input and processing module includes a profile generation unit, which is used to acquire basic user indicator data, including blood sugar, uric acid, and blood pressure, and to complete the standardization processing of the basic user indicator data, including data cleaning, extraction, and fusion, to form basic user standard data.
[0091] In addition, by combining users' health history data, including behavioral habits, physical defects, historical medical conditions, current diseases, drug allergies, types of Chinese and Western medicines, and medical payment records, the standardized user basic data is tagged to obtain a user profile that reflects an individual's health status.
[0092] The data input and processing module includes an initial suggestion generation unit, which is used to pre-train and fine-tune the authoritative knowledge database, medical report database, and medical prediction knowledge base based on the general basic large language model to form a medical basic large language model;
[0093] And by employing single-round medical dialogue, multi-round medical dialogue, and multi-task medical methods, we supervise and fine-tune the data model based on medical data to form a large medical reasoning model;
[0094] Furthermore, it utilizes deep learning algorithms to analyze user feedback to obtain user intent, and inputs this intent into a large-scale medical reasoning model to obtain a series of personalized health recommendations based on medical industry knowledge and a large-scale health intervention model.
[0095] The dual attention mechanism module includes a preprocessing unit, a weight normalization unit, a feature representation unit, an extraction and integration unit, and an output unit;
[0096] The preprocessing unit is used to preprocess the input sequence, which includes user profile data and initial personalized health suggestions, and convert it into a format that the model can process, including word vectors and feature maps.
[0097] The weight normalization unit is used to calculate the weight of each element in the input sequence in the first-layer attention mechanism. It normalizes the weights by calculating the similarity or correlation between elements and using functions such as softmax.
[0098] The feature representation unit is used to perform weighted summation or weighted average on the input sequence based on the weights calculated by the first-layer attention mechanism to obtain a feature representation containing key information, which will be used as the input of the second-layer attention mechanism.
[0099] The extraction and integration unit is used to process the feature representations obtained from the first attention mechanism in the second attention mechanism. By calculating the correlation or importance between different features, it performs weighted processing again to extract and integrate key information.
[0100] The output unit is used as the final output of the dual attention mechanism model, based on the feature representation processed by the second-layer attention mechanism.
[0101] Classifiers include fully connected layers, Softmax classifiers, support vector machines, and decision trees or ensemble learning methods.
[0102] The fully connected layer is connected after the output of the feature extraction network. It is responsible for transforming and combining the extracted user intent features and the initial personalized health interventions obtained after processing by the dual attention mechanism to determine the probability distribution of whether a series of health intervention opinions should be recommended.
[0103] The Softmax classifier transforms the output of the fully connected layer into a probability distribution of whether or not a recommendation should be made, thus determining whether a health intervention should be recommended.
[0104] Support vector machines are used as classifiers to classify extracted features;
[0105] Decision tree or ensemble learning methods use decision trees, random forests, gradient boosting trees, and ensemble learning methods as classifiers to determine whether to recommend health interventions.
[0106] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A personalized health intervention recommendation method based on a dual attention mechanism, characterized in that, Includes the following steps: The system utilizes preprocessed full user data to generate corresponding user profiles and obtains initial personalized health recommendations based on a large-scale health intervention industry model. Two global and local attention modules using a dual attention mechanism are used to extract features by separately focusing on the user profile and the initial personalized health suggestions through adaptive weight parameters. The initial personalized health suggestions and user profiles, which have undergone weighting and feature extraction, are transformed and combined using a classifier model. Based on the similarity of feature vectors, the probability distribution of a series of personalized health intervention recommendations is determined, and the health intervention recommendation results are determined based on the probability distribution.
2. The personalized health intervention recommendation method based on dual attention mechanism according to claim 1, characterized in that, The method for generating corresponding user profiles includes the following steps: Acquire basic user indicator data, including blood glucose, uric acid, and blood pressure, and complete the standardization processing of the basic user indicator data, including data cleaning, extraction, and fusion, to form basic user standard data; By combining users' health history data, including behavioral habits, physical defects, historical medical conditions, current diseases, drug allergies, types of Chinese and Western medicines, and medical payment records, the standardized user basic data is tagged to obtain a user profile that reflects an individual's health status.
3. The personalized health intervention recommendation method based on dual attention mechanism according to claim 2, characterized in that, The method for obtaining initial personalized health recommendations based on a large-scale industry model of health interventions includes the following steps: Based on the general basic large language model, a medical basic large language model is formed by pre-training and fine-tuning authoritative knowledge databases, medical report databases, and medical prediction knowledge bases. By employing single-round medical dialogue, multi-round medical dialogue, and multi-task medical methods, a large-scale medical reasoning model is formed through supervised fine-tuning of the data model based on medical data. By using deep learning algorithms to analyze user feedback information to obtain user intent, and inputting the user intent into a large medical reasoning model, a series of personalized health recommendations based on medical industry knowledge and a large health intervention industry model are obtained.
4. The personalized health intervention recommendation method based on dual attention mechanism according to claim 3, characterized in that, The dual attention mechanism module learns to dynamically adjust and weight data and feature channels in different directions through adaptive weight parameters, including the following steps: The input sequence, which includes user profile data and initial personalized health suggestions, is preprocessed and converted into a format that the model can process, including word vectors and feature maps. In the first-layer attention mechanism, the weight of each element in the input sequence is calculated. The weights are normalized by calculating the similarity or correlation between elements and using functions such as softmax. The input sequence is weighted and summed or weighted averaged based on the weights calculated by the first-layer attention mechanism to obtain a feature representation containing key information, which will be used as the input of the second-layer attention mechanism. The feature representations obtained from the first attention mechanism are processed in the second attention mechanism. By calculating the correlation or importance between different features, the features are weighted again to extract and integrate key information. The feature representation processed by the second-layer attention mechanism is used as the final output of the dual-attention mechanism model.
5. The personalized health intervention recommendation method based on dual attention mechanism according to claim 4, characterized in that, The classifier includes fully connected layers, Softmax classifiers, support vector machines, and decision trees or ensemble learning methods; The fully connected layer is connected after the output of the feature extraction network, and is responsible for transforming and combining the extracted user intent features and the initial personalized health interventions obtained after processing by the dual attention mechanism, so as to determine the probability distribution of whether a series of health intervention opinions should be recommended. The Softmax classifier transforms the output of the fully connected layer into a probability distribution of whether or not a recommendation should be made, thereby determining whether a health intervention recommendation should be made. The support vector machine is used as a classifier to classify the extracted features; The decision tree or ensemble learning method uses decision trees, random forests, gradient boosting trees, and ensemble learning methods as classifiers to determine whether to recommend health interventions.
6. A personalized health intervention recommendation system based on a dual attention mechanism, characterized in that, Includes the following modules: The data input and processing module is used to generate corresponding user profiles using preprocessed full user data, and to obtain initial personalized health recommendations based on a large-scale health intervention industry model. The dual attention mechanism module is used to extract features by utilizing two global and local attention modules of the dual attention mechanism, which separately focus on the user profile and the initial personalized health suggestions through adaptive weight parameters. The classifier recommendation module is used to transform and combine the initial personalized health suggestions and user profiles after weighting and feature extraction using a classifier model. It determines the probability distribution of a series of personalized health intervention recommendations based on feature vector similarity and determines the health intervention recommendation results based on the probability distribution.
7. The personalized health intervention recommendation system based on dual attention mechanism according to claim 6, characterized in that, The data input and processing module includes a profile generation unit, which is used to acquire user basic indicator data, including blood glucose, uric acid, and blood pressure, and to complete the standardization processing of user basic indicator data, including data cleaning, extraction, and fusion, to form user basic standard data. In addition, by combining users' health history data, including behavioral habits, physical defects, historical medical conditions, current diseases, drug allergies, types of Chinese and Western medicines, and medical payment records, the standardized user basic data is tagged to obtain a user profile that reflects an individual's health status.
8. The personalized health intervention recommendation system based on dual attention mechanism according to claim 7, characterized in that, The data input and processing module includes an initial suggestion generation unit, which is used to pre-train and fine-tune an authoritative knowledge database, a medical report database, and a medical prediction knowledge base based on a general basic large language model to form a medical basic large language model. And by employing single-round medical dialogue, multi-round medical dialogue, and multi-task medical methods, we supervise and fine-tune the data model based on medical data to form a large medical reasoning model; Furthermore, it utilizes deep learning algorithms to analyze user feedback to obtain user intent, and inputs this intent into a large-scale medical reasoning model to obtain a series of personalized health recommendations based on medical industry knowledge and a large-scale health intervention model.
9. The personalized health intervention recommendation system based on dual attention mechanism according to claim 8, characterized in that, The dual attention mechanism module includes a preprocessing unit, a weight normalization unit, a feature representation unit, an extraction and integration unit, and an output unit; The preprocessing unit is used to preprocess the input sequence, which includes user profile data and initial personalized health suggestions, and convert it into a format that the model can process, including word vectors and feature maps. The weight normalization unit is used to calculate the weight of each element in the input sequence in the first-layer attention mechanism. By calculating the similarity or correlation between elements, the weights are normalized using functions such as softmax. The feature representation unit is used to perform a weighted summation or weighted average on the input sequence based on the weights calculated by the first-layer attention mechanism to obtain a feature representation containing key information, which will be used as the input of the second-layer attention mechanism. The extraction and integration unit is used to process the feature representation obtained by the first attention mechanism in the second attention mechanism. By calculating the correlation or importance between different features, it performs weighted processing again to extract and integrate key information. The output unit is used as the final output of the dual attention mechanism model based on the feature representation processed by the second-layer attention mechanism.
10. The personalized health intervention recommendation system based on dual attention mechanism according to claim 9, characterized in that, The classifier includes fully connected layers, Softmax classifiers, support vector machines, and decision trees or ensemble learning methods; The fully connected layer is connected after the output of the feature extraction network, and is responsible for transforming and combining the extracted user intent features and the initial personalized health interventions obtained after processing by the dual attention mechanism, so as to determine the probability distribution of whether a series of health intervention opinions should be recommended. The Softmax classifier transforms the output of the fully connected layer into a probability distribution of whether or not a recommendation should be made, thereby determining whether a health intervention recommendation should be made. The support vector machine is used as a classifier to classify the extracted features; The decision tree or ensemble learning method uses decision trees, random forests, gradient boosting trees, and ensemble learning methods as classifiers to determine whether to recommend health interventions.