Evaluation method based on five-transportation-six-gas and big data collaborative filtering analysis

Abstract

The invention belongs to the technical field of collaborative filtering algorithms, and discloses a method which utilizes multi-modal data and a deep learning technology and integrates a standardized traditional Chinese medicine diagnosis index system and objective medical data to keep the essence of the traditional Chinese medicine theory and construct a clear scientific verification mechanism. Carrying out multi-modal data acquisition; step 2, data preprocessing; step 3, constructing a multi-modal deep fusion model; 4, performing model training and scientific verification; and step 5, outputting personalized health-care guidance. According to the method, the traditional Chinese medicine theory and the modern data science and technology are integrated, and the accuracy and robustness of health risk assessment are effectively improved through standardized preprocessing and multi-modal deep fusion of multi-source data, so that the assessment result not only reflects the traditional Chinese medicine physique characteristics, but also has objective modern medical data support; therefore, a more personalized and refined health management scheme is provided for the user.

Description

Technical Field

[0001] The present invention belongs to the technical field of collaborative filtering algorithms, and specifically relates to an evaluation method based on the combination of the five - element theory of the five - element theory of traditional Chinese medicine and big data collaborative filtering analysis. Background Technique

[0002] The five - element theory of the five - element theory of traditional Chinese medicine is the core content of the theory of five - element theory of traditional Chinese medicine. In the application in the field of traditional Chinese medicine, it expands the framework of the traditional Chinese medicine theory system and has the function of inspiring and promoting the development of traditional Chinese medicine theory. At present, the evaluation of people's health status and corresponding health care measures are mainly given by attending doctors in major hospitals. In order to understand the health status of the body intelligently, an evaluation method has been proposed. For example, Chinese Patent Publication (Announcement) No.: CN115662628A "An Evaluation Method Based on the Combination of the Five - Element Theory of the Five - Element Theory of Traditional Chinese Medicine and Big Data Collaborative Filtering Analysis". This method combines the evaluation system of traditional Chinese medicine's five - element theory of the five - element theory of traditional Chinese medicine and its evaluation method based on big data collaborative filtering analysis to intelligently and accurately give people's health risk assessment and corresponding health care guidance plans, which has important scientific significance and application value for realizing personal health management.

[0003] However, in this solution, the five - element theory of the five - element theory of traditional Chinese medicine relies on the calculation of physical constitution information based on the date of birth, gender, etc. However, such qualitative descriptions are difficult to standardize and verify, resulting in strong data subjectivity and poor consistency. Moreover, the five - element theory of the five - element theory of traditional Chinese medicine is based on traditional philosophy, while collaborative filtering relies on statistics and mathematical modeling, and the combination of the two lacks a clear scientific verification mechanism. Therefore, it needs to be improved and optimized. Summary of the Invention

[0004] The purpose of the present invention is to provide an evaluation method based on the combination of the five - element theory of the five - element theory of traditional Chinese medicine and big data collaborative filtering analysis to solve the problems raised in the above - mentioned background technique.

[0005] To achieve the above purpose, the present invention provides the following technical solution: An evaluation method based on the combination of the five - element theory of the five - element theory of traditional Chinese medicine and big data collaborative filtering analysis. This method utilizes multi - modal data and deep learning technology, and through the integration of a standardized traditional Chinese medicine diagnosis index system and objective medical data, enables the method to not only retain the essence of traditional Chinese medicine theory but also build a clear scientific verification mechanism. The specific steps are as follows:

[0006] Step 1, multi - modal data collection: Collect the user's physical attribute data and objective medical data, and apply a traditional Chinese medicine qualitative questionnaire to convert the qualitative descriptions of traditional Chinese medicine constitution and symptoms into scale data;

[0007] Step 2, data pre - processing: Pre - process the data collected in Step 1. The pre - processing stage includes data cleaning and missing value processing, as well as data standardization and discretization;

[0008] Step 3, construct a multi-modal deep fusion model: Extract different pre-processed data sources, perform feature extraction respectively, then fuse the information of each modality, and finally construct a user health portrait and a similarity analysis model;

[0009] Step 4, model training and scientific verification: Ensure that the model fully learns the features of multi-modal data during the training process through self-supervised training, and evaluate the output results of the model through scientific cross-validation and expert review;

[0010] Step 5, output personalized health care guidance: Based on the fused user features, the model outputs a comprehensive health risk score, gives a standardized classification of traditional Chinese medicine constitutions, and further generates personalized health care suggestions.

[0011] Preferably, the physical attribute data of the user includes gender, age, birth date, height, physical constitution information, bad habits and sleep status, the objective medical data includes wearable device data, electronic medical record records and gene test results, and the traditional Chinese medicine qualitative questionnaire uses a scoring system of 1-5 to quantify the data.

[0012] Preferably, after collecting the multi-modal data provided in Step 1, the step 2 detects outliers through the Z-score method, and for the missing categorical data, uses similar samples provided by big data for predictive filling. Secondly, the data from different sources are converted into a unified data format to ensure that the field names, data types and units are consistent, providing a unified data basis for the construction of the subsequent multi-modal deep fusion model.

[0013] Preferably, the specific steps for constructing the multi-modal deep fusion model are as follows:

[0014] A1, perform feature extraction on the processed multi-modal data provided in Step 2 respectively to obtain the extracted feature vectors;

[0015] A2, map the medical data feature vectors and questionnaire data feature vectors to the same dimension through fully connected layers respectively, then splice the mapped feature vectors to obtain preliminary fusion features, and finally adaptively learn the correlation between different modalities through a multi-head attention module and output the final fusion features;

[0016] A3, construct a node for each user through the features fused previously, and then calculate the edge weights between users according to the cosine similarity method to form an adjacency matrix;

[0017] A4, then perform convolution operations on the user image information through a graph convolutional network to update the representation of each node. At this time, through multiple layers of GCN, more robust user embedding vectors can be obtained.

[0018] Preferably, the specific implementation steps of the model training and scientific verification stage are as follows:

[0019] B1. First, in the data partitioning and self-supervised pre-training stage, the preprocessed dataset is strictly partitioned into a training set, a validation set, and a test set. In particular, for time series data, the time continuity is maintained to avoid data leakage.

[0020] B2. Through the occlusion task and the reconstruction task, the model performs self-supervised learning on the internal structure of the data, thereby obtaining more robust pre-training parameters.

[0021] B3. Then, multi-task joint training is carried out. During the training stage, the model synchronously completes three tasks: health risk score regression, traditional Chinese medicine constitution classification, and data reconstruction on the basis of pre-training. It is optimized through a joint loss function, and the parameters are continuously adjusted on the validation set until the model performance reaches the best.

[0022] B4. Finally, a multi-level verification system of cross-validation and expert review is used to verify the scientific nature and practical application effect of the model output.

[0023] Preferably, the fifth step is implemented based on the Transformer-based Seq2Seq model, and the specific steps are as follows:

[0024] C1. Using the user features generated by the multi-modal deep fusion model and the graph neural network, combined with the regression model to obtain the health risk score.

[0025] C2. Combining the Seq2Seq model with the pre-constructed traditional Chinese medicine clinical guidelines and health care experience, and using conditional generation technology to ensure that the output content conforms to traditional Chinese medicine theory and at the same time takes into account the requirements of modern health management.

[0026] C3. Finally, based on the Seq2Seq model, taking the user's health risk score, constitution classification, and other key indicators as context inputs, and automatically generating targeted health care guidance text.

[0027] Preferably, in A1, it includes the extraction of objective medical data features and the extraction of traditional Chinese medicine questionnaire data and index features.

[0028] Among them, for the extraction of objective medical data features, after preprocessing the medical data, a multi-layer CNN is constructed, and then a fully connected layer is added at the end of the network to output a feature vector of a fixed dimension. For the extraction of traditional Chinese medicine questionnaire data and index features, the questionnaire is numerically encoded and then input into the Transformer Encoder module. The relationship between each index is captured through the multi-head self-attention mechanism, and finally the output of the last layer of the encoder is taken to output a feature vector of a fixed dimension.

[0029] Preferably, after automatically generating targeted health care guidance texts, feedback data needs to be regularly collected, which is used for model retraining and knowledge base updating to continuously improve the accuracy and personalization of the suggestions.

[0030] The beneficial effects of the present invention are as follows:

[0031] 1. The present invention integrates traditional Chinese medicine theory and modern data science technology. Through the standardized preprocessing of multi-source data and multi-modal deep fusion, it effectively improves the accuracy and robustness of health risk assessment, making the assessment results not only reflect the characteristics of traditional Chinese medicine constitution but also have objective modern medical data support. Thus, it provides a more personalized and refined health management plan for users. Moreover, through multiple verification strategies such as cross-validation and expert review, it ensures that the output results have high scientific credibility and practical application value, reducing the errors caused by data subjectivity and noise.

[0032] 2. In terms of outputting and personalized health care guidance, by combining the generation model with knowledge base rule reasoning, the present invention can generate targeted, rich in content and easy-to-understand health care suggestions according to the comprehensive health portrait of each user, forming a closed-loop feedback system and achieving dynamic optimization and continuous improvement.

[0033] 3. During the model training process, the present invention uses self-supervised pre-training and multi-task learning to improve the learning effect of data features, enabling the model to share knowledge between different tasks, enhancing the overall prediction ability, and ensuring that the assessment results are more stable and consistent. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a flowchart for evaluating health care suggestions of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0035] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0036] As Figure 1 shown, the embodiments of the present invention provide an evaluation method based on the collaborative filtering analysis of the five-element theory of the five evolutive phases and six climatic factors and big data. This method uses multi-modal data and deep learning technology, and through the integration of a standardized traditional Chinese medicine diagnosis index system and objective medical data, it not only retains the essence of traditional Chinese medicine theory but also constructs a clear scientific verification mechanism. The specific steps are as follows:

[0037] Step 1, Multimodal data collection: Collect the user's physical attribute data and objective medical data, and apply a traditional Chinese medicine qualitative questionnaire to convert the qualitative descriptions of traditional Chinese medicine constitutions (such as qi deficiency, blood stasis, damp-heat, yin deficiency, etc.) and symptoms into scale data;

[0038] Step 2, Data preprocessing: Preprocess the data collected in Step 1. The preprocessing stage includes data cleaning and missing value handling, as well as data standardization and discretization;

[0039] Step 3, Construct a multimodal deep fusion model: Extract different data sources after preprocessing, perform feature extraction separately, and then fuse the information of each modality. Finally, construct a user health portrait and a similarity analysis model;

[0040] Step 4, Model training and scientific verification: Ensure that the model fully learns the features of multimodal data during the training process through self-supervised training, and evaluate the output results of the model through scientific cross-validation and expert review;

[0041] Step 5, Output personalized health care guidance: Based on the fused user features, the model outputs a comprehensive health risk score, and gives a standardized traditional Chinese medicine constitution classification (such as qi deficiency, blood stasis, etc.). Each category corresponds to a clear numerical range and index explanation, and further generates personalized health care suggestions.

[0042] By integrating traditional Chinese medicine theory and modern data science technology, through the standardized preprocessing of multi-source data and multimodal deep fusion, the accuracy and robustness of health risk assessment are effectively improved. The assessment results not only reflect the characteristics of traditional Chinese medicine constitutions but also have objective modern medical data support, thus providing a more personalized and refined health management plan for users. Additionally, through multiple verification strategies such as cross-validation and expert review, it is ensured that the output results have high scientific credibility and practical application value, reducing the errors caused by data subjectivity and noise.

[0043] Among them, the user's physical attribute data includes gender, age, birth date, height, constitution information, bad habits, and sleep status. The objective medical data includes wearable device data (electronic bracelets), electronic medical record records, and gene test results. The traditional Chinese medicine qualitative questionnaire uses a 1-5 scoring system to quantify the data.

[0044] The original qualitative data descriptions (such as "qi deficiency", "blood stasis") are converted into digital codes. For example, "qi deficiency" is coded as 1, "blood stasis" is coded as 2, etc.

[0045] Among them, after collecting the multi-modal data provided in Step 1, the Z-score method (standard score) is used to detect outliers in Step 2. If a data point exceeds the mean ± 3 standard deviations, it is regarded as an outlier, and the data is then removed. For missing categorical data, prediction filling is performed using similar samples provided by big data. Secondly, data from different sources are converted into a unified data format to ensure that field names, data types, and units are consistent, providing a unified data basis for the construction of the subsequent multi-modal deep fusion model.

[0046] Among them, the specific steps for constructing the multi-modal deep fusion model are as follows:

[0047] A1. Respectively extract features from the processed multi-modal data provided in Step 2 to obtain the extracted feature vectors;

[0048] Use CNN to extract local and temporal features of medical data, and use Transformer to extract global semantic information of traditional Chinese medicine questionnaire data;

[0049] A2. Map the medical data feature vectors and questionnaire data feature vectors to the same dimension through fully connected layers respectively, then splice the mapped feature vectors to obtain the preliminary fusion features, and finally adaptively learn the correlation between different modalities through the multi-head attention module and output the final fusion features;

[0050] Through fully connected layer mapping, splicing, and the multi-head attention module, effective information interaction between different data modalities is achieved, improving the robustness and representativeness of the overall feature expression;

[0051] A3. Based on the features fused previously, construct a node for each user, and then calculate the edge weights between users according to the cosine similarity method to form an adjacency matrix;

[0052] A4. Then perform convolution operations on the user image information through the graph convolutional network (GCN) to update the representation of each node. At this time, through multiple layers of GCN, more robust user embedding vectors can be obtained, facilitating subsequent similarity calculation and personalized health risk assessment.

[0053] Construct the fused features into a user graph, and update the node features through GCN / GAT, so that the relationship between users no longer depends only on the statistical similarity of traditional collaborative filtering, but comprehensively combines deep semantic information and multi-modal features, enhancing the accuracy and interpretability of user similarity analysis.

[0054] Through the above steps, the constructed multi-modal deep fusion model can not only extract and fuse the key features in traditional Chinese medicine data and modern medical data, but also model the complex relationships between users through the graph neural network.

[0055] Among them, the specific implementation steps of the model training and scientific verification stage are as follows:

[0056] B1. First, in the data partitioning and self-supervised pre-training stage, the preprocessed dataset is strictly partitioned into a training set, a validation set, and a test set. In particular, the time continuity of time series data is maintained to avoid data leakage;

[0057] The preprocessed dataset is partitioned into a training set (e.g., 70%), a validation set (15%), and a test set (15%).

[0058] B2. Through occlusion tasks and reconstruction tasks, the model performs self-supervised learning on the internal structure of the data, thereby obtaining more robust pre-training parameters;

[0059] For text data, randomly occlude some inputs (e.g., randomly occlude some indicators in the questionnaire), and let the model predict the occluded parts, thereby learning the internal structure of the data. The reconstruction task is to reconstruct the input data through an autoencoder, forcing the model to learn more efficient representations.

[0060] B3. Then, multi-task joint training is carried out. During the training stage, the model synchronously completes three tasks: health risk score regression, traditional Chinese medicine constitution classification, and data reconstruction on the basis of pre-training. It is optimized through a joint loss function (weighted sum of the losses of each task), and the parameters are continuously adjusted on the validation set until the model performance reaches the best;

[0061] B4. Finally, a multi-level verification system of cross-validation and expert review is used to verify the scientific nature and practical application effect of the model output.

[0062] Cross-validation means that the training set is divided into K folds, and each subset is separately trained and verified. The mean and standard deviation of the model performance on each fold are calculated to ensure the generalization ability of the model. Expert review is to conduct blind reviews on the model prediction results by traditional Chinese medicine and modern medicine experts, and compare the matching degree between the risk scores given by the model and the actual health assessments.

[0063] The model parameters are continuously optimized through real-time monitoring and regular re-training to ensure that the overall solution is both accurate and robust in the health risk assessment of the integration of traditional Chinese medicine and modern data.

[0064] Among them, the fifth step is implemented based on the Transformer-based Seq2Seq model, and the specific steps are as follows:

[0065] C1. Utilize the user features generated by the multi-modal deep fusion model and the graph neural network, and combine with the regression model to obtain the health risk score;

[0066] C2. Incorporate the pre - constructed traditional Chinese medicine (TCM) clinical guidelines and health preservation experiences into the Seq2Seq model. Through conditional generation technology, ensure that the output content conforms to TCM theory while taking into account the requirements of modern health management.

[0067] In this step, big data technology can be used to sort out TCM classics, clinical guidelines and expert consensus, and construct a structured knowledge base containing association rules such as constitution categories, risk indicators, and recommended solutions.

[0068] C3. Finally, based on the Seq2Seq model, take the user's health risk score, constitution classification and other key indicators as context inputs, and automatically generate targeted health preservation guidance text.

[0069] The output suggestions include multiple aspects such as diet, exercise, emotion regulation, and medicinal diet conditioning.

[0070] Among them, A1 includes the extraction of objective medical data features and the extraction of TCM questionnaire data and index features.

[0071] Among them, for the extraction of objective medical data features, after pre - processing the medical data, a multi - layer CNN is constructed, and then a fully - connected layer is added at the end of the network to output a feature vector of a fixed dimension. For the extraction of TCM questionnaire data and index features, the questionnaire is numerically encoded and then input into the Transformer Encoder module. The relationship between each index is captured through the multi - head self - attention mechanism, and finally the output of the last layer of the encoder is taken to output a feature vector of a fixed dimension.

[0072] Among them, after automatically generating the targeted health preservation guidance text, feedback data needs to be collected regularly. This data is used for model retraining and knowledge base update to continuously improve the accuracy and personalization of the suggestions.

[0073] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "including", "comprising" or any other variant thereof is intended to cover non - exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0074] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An evaluation method based on the collaborative filtering analysis of the five - element theory of the five - element cycles of qi and six - qi and big data, characterized in that: This method utilizes multi-modal data and deep learning techniques, and through the integration of a standardized traditional Chinese medicine (TCM) diagnosis index system and objective medical data, it not only retains the essence of TCM theory but also constructs a clear scientific verification mechanism. The specific steps are as follows: Step 1, Multi-modal data collection: Collect the user's physical attribute data and objective medical data, and apply a TCM qualitative questionnaire to convert the qualitative descriptions of TCM constitution and symptoms into scale data; Step 2, Data preprocessing: Preprocess the data collected in Step 1. The preprocessing stage includes data cleaning and missing value handling, as well as data standardization and discretization; Step 3, Construct a multi-modal deep fusion model: Extract different data sources after preprocessing, perform feature extraction separately, and then fuse the information of each modality. Finally, construct a user health portrait and a similarity analysis model; Step 4, Model training and scientific verification: Ensure that the model fully learns the features of multi-modal data during training through self-supervised training, and evaluate the output results of the model through scientific cross-validation and expert review; Step 5, Output personalized health care guidance: Based on the fused user features, the model outputs a comprehensive health risk score, gives a standardized TCM constitution classification, and further generates personalized health care suggestions.

2. The evaluation method based on the collaborative filtering analysis of the five - element theory of the five - element theory and big data according to claim 1, wherein: The user's physical attribute data includes gender, age, birth date, height, constitution information, bad habits, and sleep status. The objective medical data includes wearable device data, electronic medical record records, and gene test results. The TCM qualitative questionnaire uses a 1-5 point scoring system to quantify the data.

3. The evaluation method based on the collaborative filtering analysis of the five - element theory of the five - element theory of traditional Chinese medicine and big data according to claim 1, wherein: After collecting the multi-modal data provided in Step 1 in Step 2, detect outliers through the Z-score method. For missing categorical data, use similar samples provided by big data for predictive filling. Secondly, convert data from different sources into a unified data format to ensure that the field names, data types, and units are consistent, providing a unified data basis for the construction of the subsequent multi-modal deep fusion model.

4. The evaluation method based on the collaborative filtering analysis of the five - element theory of the five - element cycles of qi and big data according to claim 1, characterized in that: The specific steps for constructing the multi-modal deep fusion model are as follows: A1, Respectively perform feature extraction on the processed multi-modal data provided in Step 2 to obtain the extracted feature vectors; A2, Map the medical data feature vectors and questionnaire data feature vectors to the same dimension through fully connected layers respectively, then splice the mapped feature vectors to obtain preliminary fusion features. Finally, adaptively learn the correlation between different modalities through a multi-head attention module and output the final fusion features; A3, Based on the features fused previously, construct a node for each user, and then calculate the edge weights between users according to the cosine similarity method to form an adjacency matrix; A4, Then perform convolution operations on the user image information through a graph convolutional network to update the representation of each node. At this time, through multiple layers of GCN, more robust user embedding vectors can be obtained.

5. The evaluation method based on the collaborative filtering analysis of the five - element theory of the five - element cycles of nature and six - qi and big data according to claim 1, characterized in that: The specific implementation steps in the model training and scientific verification stage are as follows: B1, First, in the data partitioning and self-supervised pre-training stage, strictly partition the preprocessed data set into a training set, a validation set, and a test set. Especially for time series data, maintain time continuity to avoid data leakage; B2. Through the occlusion task and the reconstruction task, the model performs self-supervised learning on the internal structure of the data, thereby obtaining more robust pre-training parameters; B3. Then, multi-task joint training is carried out. During the training phase, the model synchronously completes three tasks of health risk score regression, traditional Chinese medicine constitution classification, and data reconstruction on the basis of pre-training. It is optimized through a joint loss function and continuously tuned on the validation set until the model performance reaches the best; B4. Finally, a multi-level verification system of cross-validation and expert review is used to verify the scientificity and practical application effect of the model output.

6. The evaluation method based on the collaborative filtering analysis of the five - element theory of the five - element theory and big data according to claim 1, wherein: The fifth step is implemented based on the Transformer-based Seq2Seq model. The specific steps are as follows: C1. Utilize the user features generated by the multi-modal deep fusion model and the graph neural network, and combine with the regression model to obtain the health risk score; C2. Incorporate the Seq2Seq model with the pre-constructed traditional Chinese medicine clinical guidelines and health preservation experience, and through conditional generation technology, ensure that the output content conforms to traditional Chinese medicine theory while taking into account the requirements of modern health management; C3. Finally, based on the Seq2Seq model, use the user's health risk score, constitution classification, and other key indicators as context inputs to automatically generate targeted health preservation guidance text.

7. An evaluation method based on the collaborative filtering analysis of the five - element theory of the five - element cycles of qi and six - qi and big data according to claim 1, characterized in that: The A1 includes the extraction of objective medical data features and the extraction of traditional Chinese medicine questionnaire data and index features; Among them, for the extraction of objective medical data features, after preprocessing the medical data, a multi-layer CNN is constructed, and then a fully connected layer is added at the end of the network to output a feature vector of a fixed dimension. For the extraction of traditional Chinese medicine questionnaire data and index features, the questionnaire is numerically encoded and then input into the Transformer Encoder module. The relationship between each index is captured through the multi-head self-attention mechanism, and finally the output of the last layer of the encoder is taken to output a feature vector of a fixed dimension.

8. An evaluation method based on the collaborative filtering analysis of the five-element theory of the five evolutive phases and six climatic factors and big data according to claim 6, characterized in that: After automatically generating the targeted health preservation guidance text, feedback data needs to be collected regularly. This data is used for model retraining and knowledge base update to continuously improve the accuracy and personalization of the suggestions.

Images