Traditional Chinese medicine remote diagnosis and treatment and online education integrated interaction system

Abstract

The invention discloses a traditional Chinese medicine remote diagnosis and treatment and online education integrated interaction system, which relates to the field of medical education and comprises a remote diagnosis and treatment module, a personalized online education module, a cloud resource management module, a training generation module and a safety maintenance module. A cloud resource management module is connected with an electronic medical record platform and a remote diagnosis and treatment platform to realize real-time updating and sharing of traditional Chinese medicine diagnosis and treatment case resources, so that timeliness and richness of learning resources are ensured, and a self-attention deep learning model is adopted to perform deep analysis on the traditional Chinese medicine diagnosis and treatment case resources. Diversified knowledge point training questions are generated based on the diagnostic logic; the problem that the learning effect is low due to the fact that learned theoretical knowledge cannot be combined with practical application in the traditional Chinese medicine learning process is solved, the remote diagnosis and treatment process and the online education requirement of a learning user can be closely combined, and the learning efficiency and the learning effect are improved.

Description

Technical Field

[0001] The present invention relates to the field of medical education, and in particular to an integrated interactive system of traditional Chinese medicine remote diagnosis and treatment and online education. Background Art

[0002] Traditional Chinese medicine is an important part of Chinese culture. In the traditional Chinese medicine education model, theoretical lectures in school classrooms and practical operations in clinical internships are the main learning methods. With the rapid development of information technology, the field of education is undergoing a profound digital transformation. Online education, with its advantages of breaking through time and space limitations, rich and diverse resources, and flexible and convenient learning methods, makes the learning and dissemination of traditional Chinese medicine knowledge more diverse and efficient.

[0003] However, the existing TCM education model has certain limitations. During the theoretical learning stage, students acquire the relevant knowledge system of TCM through classroom lectures and online learning. Then, they come into contact with TCM patients during clinical internships and perform actual operations to improve their diagnosis and treatment capabilities. Theoretical learning and practice are separated, and there are limited opportunities for contact with clinical practice. Especially in the early stages of learning, students often lack practical operation experience and are unable to combine the theoretical knowledge they have learned with practical application, resulting in poor learning effects. In addition, the geographical distribution of students also limits their exposure to different cases, and they cannot fully grasp the essence of TCM theory and diagnosis and treatment techniques.

[0004] Therefore, an integrated interactive system of TCM telediagnosis and treatment and online education is needed to solve the above problems. Summary of the invention

[0005] In view of the shortcomings of the existing technology, the present invention discloses an integrated interactive system of traditional Chinese medicine remote diagnosis and treatment and online education, which closely combines the remote diagnosis and treatment process with the online education needs of learning users, improves learning efficiency and learning effects, and has a high degree of automation and intelligence.

[0006] The present invention adopts the following technical solutions: An integrated interactive system of TCM remote diagnosis and treatment and online education, comprising: A remote diagnosis and treatment module, which is used for learning users to remotely watch the diagnosis and treatment process in real time or regularly, and record the diagnosis and treatment experience and questions. The remote diagnosis and treatment module obtains the video, audio and data information of the remote diagnosis and treatment process in real time through wireless communication, and uses an asynchronous interactive method to realize the discussion of the diagnosis and treatment process of the learning users. The remote diagnosis and treatment module records the behavior data, diagnosis and treatment experience and questions of the learning users in the remote viewing of the diagnosis and treatment process in real time by building a front-end user interface, and transmits it to the cloud server for storage and processing; A personalized online education module, which performs intelligent Q&A and personalized course recommendations based on the learning user's behavior data, diagnosis and treatment experience, and questions. The personalized online education module includes an intelligent Q&A unit, a behavior analysis unit, a personalized course recommendation unit, and a learning progress management unit; A cloud resource management module, which connects the electronic medical record platform and the remote diagnosis and treatment platform through the application programming interface API and the data synchronization engine to achieve real-time update and sharing of TCM diagnosis and treatment case resources, and adopts the content management architecture CMS to achieve classified management and retrieval of medical resources and knowledge resources; A training generation module, wherein the training generation module uses a self-attention deep learning model to analyze the diagnostic logic of TCM diagnosis and treatment case resources, and generates knowledge point training questions based on the diagnostic logic of TCM diagnosis and treatment case resources, and then classifies and stores the generated knowledge point training questions into a training question bank, wherein the knowledge point training questions include multiple-choice questions, fill-in-the-blank questions, short-answer questions, and simulated practice training questions; Security maintenance module, which uses the role-based access control (RBAC) model to manage user access rights and uses a three-layer firewall to protect data security, kernel platform security, and service security; The output end of the remote diagnosis and treatment module is connected to the input end of the personalized online education module, the output end of the remote diagnosis and treatment module is connected to the input end of the cloud resource management module, the output end of the cloud resource management module is connected to the input end of the training generation module, the output end of the training generation module is bidirectionally connected to the personalized online education module, and the security maintenance module works throughout the whole process.

[0007] Furthermore, the intelligent question-answering unit adopts a large natural language model to perform semantic analysis on the questions raised by the learning users during the remote viewing of the diagnosis and treatment process, and adopts a multi-way recall mechanism to perform knowledge retrieval in the cloud medical resources, knowledge resources and training question bank, match and infer the answers. The behavior analysis unit adopts a deep hierarchical clustering model to construct a learning user portrait. The deep hierarchical clustering model identifies the learning user's learning pattern, knowledge level, learning difficulties, questions and interests based on the learning user's behavior data and diagnosis and treatment experience. The personalized course recommendation unit adopts an improved COSMO search recommendation algorithm to recommend personalized course resources to the learning user. The learning progress management unit captures the learning user's completion status of the course or exercises in real time through the learning tracking engine, and dynamically adjusts the learning plan according to the learning user's learning progress and goals.

[0008] Furthermore, the working method of the deep hierarchical clustering model comprises the following steps: Step 1: Collect the behavioral data and treatment experience of learning users, including at least viewing time, learning records and interactive feedback, and clean, organize and standardize the collected data information; Step 2: Use multi-layer nonlinear transformation to extract semantic features, spatiotemporal features, and attention features from the collected data information, and perform hash coding on the extracted features to convert high-dimensional features into low-dimensional data; Step 3: Input the hash-coded low-dimensional data into a hierarchical clustering algorithm for cluster analysis, and calculate the similarity between the input data and the central clusters of learning mode, knowledge level, learning difficulty, question point and interest point. The hierarchical clustering algorithm hierarchically clusters the data according to the learning mode, knowledge level, learning difficulty and question point to form a cluster structure. If the similarity between the input data and the central clusters of learning mode, knowledge level, learning difficulty, question point and interest point is greater than the similarity low threshold, the input data is classified; Step 4: Generate learning user portraits based on clustering results.

[0009] Furthermore, the improved COSMO search recommendation algorithm searches for course resources that match the learning mode, knowledge level, learning difficulties, questions and interests of the learning user based on the learning user portrait and historical learning data. The learning user set is , Indicates the system Learning users, represents the sequence number of the learning user, , the set of course resources to be recommended is , Indicates the system Course resources to be recommended, Indicates the sequence number of the course resource to be recommended. , No. The user portrait of a learning user is , No. The attributes of the course resources to be recommended are , and Respectively represent The user portrait of the first learning user and the The kth factor affecting the matching degree among the attributes of the recommended course resources, , n represents the The user portrait of the first learning user and the The number of factors that affect the matching degree in the attributes of the recommended course resources, the number of recommended course resources With learning usersWith learning users The calculation formula for personalized matching is: In formula (1), represents the matching degree between the i-th learning user and the j-th recommended course resource, k represents the factors affecting the matching degree, represents the weight of factor k in the overall matching degree, It represents the matching function between the i-th learning user and the j-th recommended course resource affected by factor k. The matching function calculation formula between the i-th learning user and the j-th recommended course resource affected by factor k is: In formula (2), represents the kth factor feature vector that affects the matching degree in the user profile of the i-th learning user, represents the modulus of the feature vector of the kth factor affecting the matching degree in the user profile of the i-th learning user, represents the kth factor feature vector that affects the matching degree in the attributes of the jth recommended course resource, Represents the modulus of the kth eigenvector of the factor that affects the degree of matching in the attributes of the jth course resource to be recommended.

[0010] Furthermore, the content management architecture CMS creates categories of medical resources and knowledge resources through a content editor, labels the medical resources and knowledge resources based on the created categories, and learns users to search and view medical resources and knowledge resources matching keywords through search engines.

[0011] Furthermore, the self-attention deep learning model includes an input layer, a multi-head attention layer, a linear mapping layer, a residual connection layer, a position encoding layer, a feedforward neural network layer and an output layer, and the self-attention deep learning model includes the following steps: S1, receiving TCM diagnosis and treatment case resources through the input layer, and converting the received text data into an embedding vector; S2, using the position encoding layer to encode the position information of the input sequence, wherein the position encoding layer adds the order information of the input vocabulary through position embedding learning or fixed position encoding; S3, obtaining the attention weights between the words in the input sequence through the multi-head attention layer to obtain a weighted vector, wherein the multi-head attention layer captures the sequence attention weights between different positions in the input sequence by performing multiple self-attention operations in parallel, and generates a weighted sum representation of the input sequence embedding vector; S4, adjusting the dimension and size of the attention weighted vector through the linear mapping layer, wherein the linear mapping layer maps the output of the multi-head attention layer to a dimension and size suitable for the residual connection layer and the feedforward neural network layer through a linear transformation; S5, using the residual connection layer to perform feature extraction, the residual connection layer adds the vector with adjusted dimension and size to the original input sequence, and performs nonlinear transformation through an activation function; S6, extracting and transforming features through the feedforward neural network layer, wherein the feedforward neural network layer further performs nonlinear transformation and extraction on the features extracted in the previous layer by stacking multiple fully connected layers; S7. Mapping the output of the feedforward neural network layer to the task space through the output layer to obtain an output result.

[0012] Furthermore, the training generation module adopts an incremental calculation method to update the training questions. The incremental calculation method adds the newly added TCM diagnosis and treatment case resources to the data set within the time window by periodically sliding the time window forward, thereby integrating the newly added data with the data within the current time window, and forming a new training data set based on the integrated data set. The new training data set is then used to retrain the self-attention deep learning model to absorb the diagnostic logic in the newly added cases, and new knowledge point training questions are generated based on the updated self-attention deep learning model and added to the training question bank.

[0013] Furthermore, the role-based access control (RBAC) model implements control and management of user access rights by authorizing user roles, permissions and user groups. The three-layer firewall includes a network boundary firewall, a server front-end firewall and a server firewall. The network boundary firewall filters and controls malicious traffic and illegal access requests from the external network based on network layer information of IP addresses and port numbers. The server front-end firewall filters and controls illegal access to internal resources by internal users or applications based on application layer information. The server firewall prevents malware and attackers from attacking host vulnerabilities by monitoring the network traffic and process behavior of the host.

[0014] The beneficial effects of the present invention are: 1. The present invention closely combines the remote diagnosis and treatment process with the online education needs of learning users, providing TCM learners with the opportunity to watch, learn and discuss the remote diagnosis and treatment process in real time, and provides personalized intelligent Q&A and course recommendations based on learners' behavioral data and diagnosis and treatment experience, so that learning users can integrate theoretical knowledge learning and practical operations. This fusion model greatly improves the participation and effectiveness of learning.

[0015] 2. The present invention collects and analyzes the behavioral data, treatment experience and questions of learning users to achieve intelligent answering and personalized course recommendations. This personalized teaching based on big data and AI technology can accurately match learners' learning needs and interests, improve learning efficiency and learning effects, and achieve optimal allocation of educational resources.

[0016] 3. The present invention adopts a cloud resource management module to connect the electronic medical record platform and the remote diagnosis and treatment platform to realize the real-time update and sharing of TCM diagnosis and treatment case resources, ensuring the timeliness and richness of learning resources, so that learning users can be exposed to the latest TCM diagnosis and treatment technologies and cases, and adopts the content management architecture CMS to classify and manage and retrieve medical resources and knowledge resources, thereby improving the utilization efficiency and management level of resources. Learning users can quickly find the required learning resources and save a lot of time and energy.

[0017] 4. The present invention uses a self-attention deep learning model to conduct in-depth analysis of TCM diagnosis and treatment case resources, and generates a variety of knowledge point training questions based on diagnostic logic. These training questions cover multiple-choice questions, fill-in-the-blank questions, short-answer questions, and simulated practice training questions, which improves the quality and pertinence of training questions, helps learners to fully master TCM diagnosis and treatment knowledge and skills, and helps learners transform theoretical knowledge into practical operation capabilities.

[0018] 5. The present invention adopts the role-based access control (RBAC) model to manage user access rights, ensuring that only users with corresponding permissions can access and use system resources. This mechanism effectively prevents the risk of data leakage and illegal access, and adopts a three-layer firewall to protect data security, kernel platform security, and service security, building a multi-level protection system. This all-round protection mechanism ensures the stability of system operation and the security of data. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a schematic diagram of the overall system architecture of the present invention; Figure 2 This is an architecture diagram of the personalized online education module in the present invention; Figure 3 Schematic diagram of the workflow of the deep hierarchical clustering model in the present invention; Figure 4 Schematic diagram of the workflow of the self-attention deep learning model in the present invention; Figure 5 It is a schematic diagram of the architecture of the three-layer firewall in the present invention. DETAILED DESCRIPTION

[0020] The following will be combined with the attached embodiment of the present invention Figure 1 To Attachment Figure 5 , the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] The embodiment of the present invention discloses an integrated interactive system of TCM remote diagnosis and treatment and online education, the system comprising: Remote diagnosis and treatment module: Learning users can watch the remote diagnosis and treatment process through mobile phones, computers and other terminal devices, and can also record diagnosis and treatment experience and questions to facilitate subsequent learning and discussion. The remote diagnosis and treatment module can be connected to a variety of remote diagnosis and treatment platforms to achieve the application of a variety of remote diagnosis and treatment scenarios.

[0022] Personalized online education module: By analyzing the behavioral data, treatment experience and questions of learning users, the intelligent question-answering unit can provide personalized question-answering services to learning users. The behavioral analysis unit can analyze the learning behavior and learning progress of learning users, and then recommend course content suitable for learning users. The personalized course recommendation unit and learning progress management unit can tailor course progress and learning plans for learning users.

[0023] Cloud resource management module: By connecting the electronic medical record platform and the remote diagnosis and treatment platform, real-time updating and sharing of TCM diagnosis and treatment case resources can be achieved. At the same time, by adopting the CMS content management architecture, medical resources and knowledge resources can be classified, managed, retrieved and viewed, making it easier for learning users to study and research.

[0024] Training generation module: The training generation module uses the self-attention deep learning model to analyze the diagnostic logic of TCM diagnosis and treatment case resources, and generates knowledge point training questions based on the diagnostic logic of TCM diagnosis and treatment case resources. The generated knowledge point training questions are then classified and stored in the training question bank, providing learning users with a variety of types of training questions, which helps to deepen their understanding and mastery of TCM knowledge.

[0025] Security maintenance module: By adopting the role-based access control (RBAC) model to manage user access rights and using a three-layer firewall to protect data security, kernel platform security, and service security, the system's security and reliability are improved, ensuring the learning and communication security of learning users.

[0026] The system integrates TCM remote diagnosis and treatment with online education. By watching and learning the remote diagnosis and treatment process in real time, students can deepen their understanding and mastery of TCM diagnosis and treatment techniques and methods. At the same time, it provides personalized online education services, tailors course content and learning plans for learning users, and helps improve the diagnosis and treatment level of TCM learners.

[0027] The output end of the remote diagnosis and treatment module is connected to the input end of the personalized online education module, the output end of the remote diagnosis and treatment module is connected to the input end of the cloud resource management module, the output end of the cloud resource management module is connected to the input end of the training generation module, the output end of the training generation module is bidirectionally connected to the personalized online education module, and the security maintenance module works throughout the process.

[0028] The intelligent question-answering unit uses natural language processing technology, including modules such as semantic analysis, knowledge base retrieval, and answer generation, to respond to questions raised by learning users and provide accurate and reliable answers. Specifically, the intelligent question-answering unit first performs semantic analysis on the questions raised by learning users, and uses natural language large model technology to perform operations such as word segmentation, part-of-speech tagging, and syntactic analysis on the questions, and accurately extracts and understands the semantics of the questions. Then, a multi-way recall mechanism is used to retrieve knowledge from cloud medical resources, knowledge resources, and training question banks, and the search results are matched, inferred, and other operations are performed to finally generate accurate answers.

[0029] The behavior analysis unit uses a deep hierarchical clustering model to identify key information such as the learning mode, knowledge level, learning difficulties, questions, and interests of the learning users through the analysis of the learning users' behavior data and treatment experience, and then build a portrait of the learning users. Specifically, this module uses big data analysis technology to mine and analyze the multi-dimensional data of learning users, including treatment experience, treatment records, course learning records, etc. Through cluster analysis, correlation analysis, classification analysis and other methods, it mines the knowledge points and problem points that the learning users are concerned about, builds a portrait of the learning users, and provides support for subsequent personalized recommendations and learning plans.

[0030] The personalized course recommendation unit uses an improved COSMO search recommendation algorithm to recommend personalized course resources for learning users. Specifically, the algorithm combines the two dimensions of content similarity and user interest, uses an integrated multi-feature model to recommend courses, integrates multi-dimensional features such as user behavior data, points of interest, and diagnosis and treatment experience, calculates the degree of match between courses and users, and recommends course resources that best meet user needs.

[0031] The learning progress management unit uses a learning tracking engine to capture the completion of courses or exercises by the learning user in real time, and dynamically adjusts the learning plan according to the learning progress and goals of the learning user. Specifically, the learning tracking engine can monitor the learning record of the learning user.

[0032] The deep hierarchical clustering model combines the advantages of deep learning and hierarchical clustering, aiming to extract useful information from user behavior data and generate user portraits. The following are the detailed steps of the model's working method and its implementation details: Step 1: Data collection and preprocessing Collect learning users' behavioral data and diagnosis and treatment experience, including but not limited to viewing time, learning records (such as course completion, learning path, etc.) and interactive feedback (such as likes, comments, questions, etc.). Clean the collected data to remove noise data, duplicate data and incomplete data. Organize and format the data to ensure data consistency and accuracy. Perform standardization so that data from different sources and scales can be compared and analyzed under the same framework. To ensure the reliability and diversity of data sources, data can be collected through user behavior logs, questionnaires, user feedback and other methods. Introduce automated tools or algorithms for data cleaning and standardization to improve the efficiency and accuracy of data processing.

[0033] Step 2: Feature extraction and dimensionality reduction Use multi-layer nonlinear transformations (such as convolutional layers, loop layers, and fully connected layers in deep learning models) to extract semantic features, spatiotemporal features, and attention features from the cleaned data. These features can reflect the user's learning habits, interest preferences, knowledge level, etc. Hash encoding is performed on the extracted high-dimensional features to convert high-dimensional features into low-dimensional data. Hash encoding can reduce the storage space and computational complexity of the data while retaining the main information of the data. Select a suitable deep learning model for feature extraction based on the specific application scenario to ensure that the extracted features can fully reflect the user's learning situation.

[0034] Step 3: Hierarchical cluster analysis Input the hash-encoded low-dimensional data into the hierarchical clustering algorithm for cluster analysis. The hierarchical clustering algorithm can hierarchically cluster the data according to the dimensions of learning mode, knowledge level, learning difficulty, question point, etc., to form a clustering tree with a hierarchical structure. Calculate the similarity between the input data and the central clusters of learning mode, knowledge level, learning difficulty, question point and interest point. The similarity can be calculated by distance metrics (such as Euclidean distance, cosine similarity, etc.). If the similarity between the input data and a central cluster is greater than the similarity low threshold, the input data is classified into the central cluster. Select a suitable hierarchical clustering algorithm (such as the bottom-up merging algorithm or the top-down splitting algorithm), and set the similarity threshold and clustering level according to the actual situation.

[0035] Step 4: Generate learning user portraits Based on the clustering results and the behavioral data of learning users, build a learning user portrait. The portrait should include key information such as the user's learning mode, knowledge level, learning difficulties, questions, and interests. Apply the learning user portrait to scenarios such as personalized learning recommendations, learning effect evaluation, and learning resource optimization to provide users with more accurate and effective learning support. Generate user portraits based on clustering results and user needs, and display portrait information through visualization tools so that users and managers can intuitively understand the user's learning situation.

[0036] Through the above steps and implementation details, the deep hierarchical clustering model can effectively extract useful information from user behavior data and generate learning user portraits, providing strong support for personalized learning recommendations and teaching resource optimization.

[0037] The improved COSMO search recommendation algorithm searches for course resources that match the learning mode, knowledge level, learning difficulties, questions and interests of learning users based on learning user portraits and historical learning data. The learning user set is , Indicates the system Learning users, represents the sequence number of the learning user, , the set of course resources to be recommended is , Indicates the system Course resources to be recommended, Indicates the sequence number of the course resource to be recommended. , No. The user portrait of a learning user is , No. The attributes of the course resources to be recommended are , and Respectively represent The user portrait of the first learning user and the The attributes of the recommended course resources Factors that affect the degree of matching, , n represents the The user portrait of the first learning user and the The number of factors that affect the matching degree in the attributes of the recommended course resources, the number of recommended course resources With learning users The calculation formula for personalized matching is: In formula (1), represents the matching degree between the i-th learning user and the j-th recommended course resource, k represents the factors affecting the matching degree, represents the weight of factor k in the overall matching degree, It represents the matching function between the i-th learning user and the j-th recommended course resource affected by factor k. The matching function calculation formula between the i-th learning user and the j-th recommended course resource affected by factor k is: In formula (2), represents the kth factor feature vector that affects the matching degree in the user profile of the i-th learning user, represents the modulus of the feature vector of the kth factor affecting the matching degree in the user profile of the i-th learning user, represents the kth factor feature vector that affects the matching degree in the attributes of the jth recommended course resource, Represents the modulus of the kth eigenvector of the factor that affects the degree of matching in the attributes of the jth course resource to be recommended.

[0038] In a specific embodiment, the attributes of the course resources to be recommended are extracted, including the knowledge points, difficulty, duration, type, content, etc. of the course, so as to provide basic data for the recommendation algorithm. According to the learning user portrait and the attributes of the course resources to be recommended, the factors affecting the matching degree are determined, including the learning user's interest points, knowledge level, learning difficulties and the knowledge points and difficulty of the course resources to be recommended. Different weights are set for the factors affecting the matching degree. According to the learning user portrait and the attributes of the course resources to be recommended and the actual needs, the different contribution values ​​of different factors to the matching degree are determined, which are used as the weights for calculating the personalized matching degree. The personalized matching degree of each learning user and the course resources to be recommended is calculated according to formula (2). The personalized matching degree of the learning user and the attribute information of the course resources to be recommended are used as the input of the recommendation algorithm, and the personalized matching degree of each learning user to each course resource to be recommended is calculated, and the results are sorted from high to low according to the matching degree, and the course resources that best meet the learning user's interests, knowledge level and learning needs are recommended.

[0039] The content management architecture CMS creates two categories, "medical resources" and "knowledge resources", which are further divided into different subcategories, such as departments, diseases, symptoms, etc., to facilitate learning users to find and use corresponding resources. For each created medical resource and knowledge resource, it is labeled through the content editor, and marked as the category and related keywords for subsequent search engine search and matching. Learning users can enter relevant keywords for search through the search engine of the CMS system, and the search engine will automatically match and return the corresponding medical resources and knowledge resources. In the search results returned by the search engine, learning users can directly view the detailed information of the relevant medical resources and knowledge resources, and understand their classification, description, content, tags and other related information. Through the above, learning users can easily find and use medical resources and knowledge resources through the search engine of the CMS system. At the same time, the classification and labeling architecture also provides the system with a better way to manage and maintain resources.

[0040] The hardware working environment of the improved COSMO search recommendation algorithm may include the following aspects: High-performance server: A high-performance CPU or GPU is required to support large-scale data processing and complex computing tasks.

[0041] Memory and storage: Large-capacity memory and high-speed storage devices (such as SSD) are used to cache and process large amounts of user behavior data and product information.

[0042] Distributed system: Due to the huge amount of data to be processed, the COSMO algorithm may need to run on a distributed system to achieve parallel processing and load balancing of data. This may include multiple server nodes, interconnected by high-speed networks, to jointly undertake the tasks of data processing and recommendation services.

[0043] Data warehouse and database: A dedicated data warehouse is used to store and manage user behavior data, search results, and other data. The database system needs to support high concurrent access and fast query to ensure the real-time and accuracy of the data.

[0044] Network infrastructure: A high-speed and stable network environment is key to ensuring the normal operation of algorithms and providing high-quality services. This includes network connections within data centers and between users and platforms.

[0045] Security equipment: Considering the importance of data security and privacy protection, the hardware working environment should also include necessary security equipment, such as firewalls, intrusion detection systems, etc.

[0046] Backup and disaster recovery facilities: In order to prevent data loss and system failure, corresponding backup and disaster recovery facilities need to be configured to ensure data integrity and system availability.

[0047] The laboratory configuration uses a Core i9 series 64G memory + 128G external storage computer, and the simulation environment is established using simulation software. The on-site experimental environment is set up, the simulation data accuracy is 95%, and the algorithm operation error does not exceed 2.5%. The improved COSMO search recommendation algorithm (Group A) and the COSMO search recommendation algorithm (Group B) are used for comparative experiments. Five learning user data are manually input for course resource matching, that is, i∈[1, 5], and 500 course resources to be recommended are manually input, that is, j∈[1, 500]. Five factors that affect the degree of matching are manually defined, that is, k∈[1, 5]. The matching accuracy, recall rate, F1 value and time consumption are used to evaluate the recommendation effect of the algorithm, which is recorded in Table 1.

[0048] According to the experimental results in Table 1, the improved COSMO algorithm outperforms the ordinary COSMO algorithm in terms of accuracy, recall and F1 value, and takes less time. It can be seen that the improved COSMO algorithm can obtain more accurate, comprehensive and user-friendly learning resource recommendation results faster, and has higher practical value.

[0049] The specific implementation of the self-attention deep learning model is as follows, which enhances understanding and analysis capabilities by capturing key information and contextual relationships in the text.

[0050] S1: Input data and embedding vector conversion The input layer receives text data from TCM diagnosis and treatment case resources, which may include descriptions of the condition, diagnosis results, treatment plans, etc., and converts the input text data into embedding vectors. This step is usually completed using a pre-trained word embedding model (such as Word2Vec, GloVe, or BERT's Token Embeddings) or a randomly initialized embedding matrix. The dimension of the embedding vector is fixed, and each word or character is mapped to a vector in a high-dimensional space.

[0051] S2: Position information encoding Position encoding layer: Since the self-attention mechanism itself does not contain the position information of the sequence, it needs to be added through the position encoding layer. This can be achieved through position embedding learning (such as Position Embeddings in BERT) or fixed position encoding (such as the position encoding generated by sine and cosine functions in Transformer). The position encoding vector is added to the embedding vector so that the model can perceive the position of the word in the sequence.

[0052] S3: Multi-head attention weighting Multi-Head Attention Layer: This layer performs multiple self-attention operations (called "heads") in parallel, with each head capturing the attention weights between different positions in the input sequence. Each head first computes the query (Q), key (K), and value (V) matrices, and then computes the weight vector via a scaled dot-product attention mechanism. The outputs of all heads are concatenated and conditioned via a linear layer to produce the final weighted sum representation.

[0053] S4: Dimensions and Sizing Linear mapping layer: The output of the multi-head attention layer is mapped to a dimension that matches the residual connection layer and the feedforward neural network layer through a linear transformation. This step ensures that data can flow smoothly between different layers of the model.

[0054] S5: Feature extraction and fusion Residual connection layer: The output of the linear mapping layer is added to the original input sequence (after position encoding). This step helps alleviate the gradient vanishing problem in deep networks and allows the model to learn more complex feature representations. After that, it is usually transformed nonlinearly through an activation function (such as ReLU).

[0055] S6: Feature extraction and transformation Feedforward neural network layer: usually composed of multiple fully connected layers stacked together, each of which is followed by an activation function. This layer performs further nonlinear transformation and feature extraction on the output of the residual connection layer to capture higher-level abstract features.

[0056] S7: Task Space Mapping Output layer: The output layer maps the output of the feedforward neural network layer into the task space. For the processing of TCM diagnosis and treatment case resources, this may mean mapping the output into classification labels (such as disease type, treatment plan, etc.), regression values ​​(such as treatment effect scores), or other forms of task-specific outputs.

[0057] During implementation, the parameters of the model need to be adjusted according to the specific task and data set, such as the dimension of the embedding vector, the number of heads in the multi-head attention layer, the number of layers and units in the feedforward neural network layer, etc. The model parameters can be optimized by iterative training on the training set, and the model performance can be evaluated using loss functions such as the cross entropy loss function (for classification tasks) or the mean square error loss function (for regression tasks). During the training process, various optimization algorithms (such as Adam, SGD, etc.) can be used to update the model parameters, and the generalization ability of the model can be monitored through the validation set to avoid overfitting. Finally, the trained model can be evaluated on the test set and applied to the actual TCM diagnosis and treatment case resource processing task.

[0058] The training generation module uses an incremental calculation method to update the training questions, and defines the size of the time window according to actual needs, such as a day, a week, and other time periods. The newly added TCM diagnosis and treatment case resources are added to the data set within the time window, integrated with the data in the current time window, and the integrated data is used as a new training data set. The self-attention deep learning model is retrained using the new training data set to absorb the diagnostic logic in the newly added cases and improve the performance and predictive ability of the model in the field of TCM diagnosis and treatment cases. Based on the updated self-attention deep learning model, new knowledge point training questions are generated based on the knowledge points and logic involved in the TCM diagnosis and treatment cases, and added to the training question bank to update the training questions.

[0059] Training through incremental computing can achieve rapid integration of new data and historical data, real-time updating of training data sets and rapid iteration of models, improve model performance and intelligence, and provide users with more accurate and comprehensive diagnosis and treatment knowledge and decision-making support functions.

[0060] The role-based access control (RBAC) model authorizes and manages users based on different roles, permissions, user group divisions, and other factors to achieve control and management of user access rights. A network boundary firewall is set up in the system and network architecture design to filter and control malicious traffic and illegal access requests from external networks based on IP address and port number information to protect the system's network security. A server front-end firewall is set up inside the network boundary firewall to filter and control illegal access to internal resources by internal users or applications based on application layer information to prevent malicious access and attacks. A server firewall is set up inside the host to prevent malware and attackers from attacking host vulnerabilities by monitoring the host's network traffic and process behavior to ensure the security of the host.

[0061] Through the above three-layer firewall and RBAC authority management model, the system security and data confidentiality can be effectively guaranteed, and various network attacks and illegal access can be prevented. At the same time, the system availability and reliability can be improved, ensuring the normal operation and stability of the system.

[0062] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. An integrated interactive system of TCM remote diagnosis and treatment and online education, characterized in that: The system comprises: A remote diagnosis and treatment module, which is used for learning users to remotely watch the diagnosis and treatment process in real time or regularly, and record the diagnosis and treatment experience and questions. The remote diagnosis and treatment module obtains the video, audio and data information of the remote diagnosis and treatment process in real time through wireless communication, and uses an asynchronous interactive method to realize the discussion of the diagnosis and treatment process of the learning users. The remote diagnosis and treatment module records the behavior data, diagnosis and treatment experience and questions of the learning users in the remote viewing of the diagnosis and treatment process in real time by building a front-end user interface, and transmits it to the cloud server for storage and processing; A personalized online education module, which performs intelligent Q&A and personalized course recommendations based on the learning user's behavior data, diagnosis and treatment experience, and questions. The personalized online education module includes an intelligent Q&A unit, a behavior analysis unit, a personalized course recommendation unit, and a learning progress management unit; A cloud resource management module, which connects the electronic medical record platform and the remote diagnosis and treatment platform through the application programming interface API and the data synchronization engine to achieve real-time update and sharing of TCM diagnosis and treatment case resources, and adopts the content management architecture CMS to achieve classified management and retrieval of medical resources and knowledge resources; A training generation module, wherein the training generation module uses a self-attention deep learning model to analyze the diagnostic logic of TCM diagnosis and treatment case resources, and generates knowledge point training questions based on the diagnostic logic of TCM diagnosis and treatment case resources, and then classifies and stores the generated knowledge point training questions into a training question bank, wherein the knowledge point training questions include multiple-choice questions, fill-in-the-blank questions, short-answer questions, and simulated practice training questions; Security maintenance module, which uses the role-based access control (RBAC) model to manage user access rights and uses a three-layer firewall to protect data security, kernel platform security, and service security; The output end of the remote diagnosis and treatment module is connected to the input end of the personalized online education module, the output end of the remote diagnosis and treatment module is connected to the input end of the cloud resource management module, the output end of the cloud resource management module is connected to the input end of the training generation module, the output end of the training generation module is bidirectionally connected to the personalized online education module, and the security maintenance module works throughout the whole process.

2. According to claim 1, the integrated interactive system of TCM remote diagnosis and treatment and online education is characterized in that: The intelligent question-answering unit adopts a large natural language model to perform semantic analysis on the questions raised by the learning users during the remote viewing of the diagnosis and treatment process, and adopts a multi-way recall mechanism to perform knowledge retrieval, matching and inference of answers in the cloud medical resources, knowledge resources and training question bank. The behavior analysis unit adopts a deep hierarchical clustering model to construct a learning user portrait. The deep hierarchical clustering model identifies the learning user's learning pattern, knowledge level, learning difficulties, questions and interests based on the learning user's behavior data and diagnosis and treatment experience. The personalized course recommendation unit adopts an improved COSMO search recommendation algorithm to recommend personalized course resources to the learning users. The learning progress management unit captures the learning user's completion status of courses or exercises in real time through a learning tracking engine, and dynamically adjusts the learning plan according to the learning user's learning progress and goals.

3. According to claim 2, the integrated interactive system of TCM remote diagnosis and treatment and online education is characterized in that: The working method of the deep hierarchical clustering model includes the following steps: Step 1: Collect the behavioral data and treatment experience of learning users, including at least viewing time, learning records and interactive feedback, and clean, organize and standardize the collected data information; Step 2: Use multi-layer nonlinear transformation to extract semantic features, spatiotemporal features, and attention features from the collected data information, and perform hash coding on the extracted features to convert high-dimensional features into low-dimensional data; Step 3: Input the hash-coded low-dimensional data into a hierarchical clustering algorithm for cluster analysis, and calculate the similarity between the input data and the central clusters of learning mode, knowledge level, learning difficulty, question point and interest point. The hierarchical clustering algorithm hierarchically clusters the data according to the learning mode, knowledge level, learning difficulty and question point to form a cluster structure. If the similarity between the input data and the central clusters of learning mode, knowledge level, learning difficulty, question point and interest point is greater than the similarity low threshold, the input data is classified; Step 4: Generate learning user portraits based on clustering results.

4. According to claim 2, the integrated interactive system of TCM remote diagnosis and treatment and online education is characterized in that: The improved COSMO search recommendation algorithm searches for course resources that match the learning mode, knowledge level, learning difficulties, questions and interests of the learning users based on the learning user portrait and historical learning data. The learning user set is , represents the i-th learning user in the system, i represents the sequence number of the learning user, , the set of course resources to be recommended is , represents the jth recommended course resource in the system, j represents the sequence number of the recommended course resource, , the user profile of the i-th learning user is , the attribute of the jth course resource to be recommended is , and They represent the user profile of the i-th learning user and the k-th factor affecting the matching degree in the attributes of the j-th course resource to be recommended, respectively. , n represents the number of factors affecting the matching degree between the user profile of the i-th learning user and the attributes of the j-th recommended course resource, and the recommended course resource With learning users The calculation formula for personalized matching is: In formula (1), represents the matching degree between the i-th learning user and the j-th recommended course resource, k represents the factor affecting the matching degree, represents the weight of factor k in the overall matching degree, It represents the matching function between the i-th learning user and the j-th recommended course resource affected by factor k. The matching function calculation formula between the i-th learning user and the j-th recommended course resource affected by factor k is: In formula (2), represents the kth factor feature vector affecting the matching degree in the user profile of the i-th learning user, represents the modulus of the feature vector of the kth factor affecting the matching degree in the user profile of the i-th learning user, represents the kth factor feature vector that affects the matching degree in the attributes of the jth recommended course resource, Represents the modulus of the kth eigenvector of the factor that affects the degree of matching in the attributes of the jth course resource to be recommended.

5. According to claim 1, the integrated interactive system of TCM remote diagnosis and treatment and online education is characterized in that: The content management architecture CMS creates categories of medical resources and knowledge resources through a content editor, labels the medical resources and knowledge resources based on the created categories, and learns users to search and view medical resources and knowledge resources matching keywords through search engines.

6. The integrated interactive system of TCM tele-diagnosis and online education according to claim 1 is characterized in that: The self-attention deep learning model includes an input layer, a multi-head attention layer, a linear mapping layer, a residual connection layer, a position encoding layer, a feedforward neural network layer and an output layer. The self-attention deep learning model includes the following steps: S1, receiving TCM diagnosis and treatment case resources through the input layer, and converting the received text data into an embedding vector; S2, using the position encoding layer to encode the position information of the input sequence, wherein the position encoding layer adds the order information of the input vocabulary through position embedding learning or fixed position encoding; S3, obtaining the attention weights between the words in the input sequence through the multi-head attention layer to obtain a weighted vector, wherein the multi-head attention layer captures the sequence attention weights between different positions in the input sequence by performing multiple self-attention operations in parallel, and generates a weighted sum representation of the input sequence embedding vector; S4, adjusting the dimension and size of the attention weighted vector through the linear mapping layer, wherein the linear mapping layer maps the output of the multi-head attention layer to a dimension and size suitable for the residual connection layer and the feedforward neural network layer through a linear transformation; S5, using the residual connection layer to perform feature extraction, the residual connection layer adds the vector with adjusted dimension and size to the original input sequence, and performs nonlinear transformation through an activation function; S6, extracting and transforming features through the feedforward neural network layer, wherein the feedforward neural network layer further performs nonlinear transformation and extraction on the features extracted in the previous layer by stacking multiple fully connected layers; S7. Mapping the output of the feedforward neural network layer to the task space through the output layer to obtain an output result.

7. The integrated interactive system of TCM tele-diagnosis and online education according to claim 1 is characterized in that: The training generation module adopts an incremental calculation method to update the training questions. The incremental calculation method adds the newly added TCM diagnosis and treatment case resources to the data set within the time window by regularly sliding the time window forward, thereby integrating the newly added data with the data within the current time window, and forming a new training data set based on the integrated data set. The new training data set is then used to retrain the self-attention deep learning model to absorb the diagnostic logic in the newly added cases, and new knowledge point training questions are generated based on the updated self-attention deep learning model and added to the training question bank.

8. The integrated interactive system of TCM tele-diagnosis and online education according to claim 1 is characterized in that: The role-based access control (RBAC) model implements control and management of user access rights by authorizing user roles, permissions, and user groups. The three-layer firewall includes a network boundary firewall, a server front-end firewall, and a server firewall. The network boundary firewall filters and controls malicious traffic and illegal access requests from external networks based on network layer information of IP addresses and port numbers. The server front-end firewall filters and controls illegal access to internal resources by internal users or applications based on application layer information. The server firewall prevents malware and attackers from attacking host vulnerabilities by monitoring the network traffic and process behavior of the host.

Images