Deep learning-based traditional chinese medicine school inheritance teaching system

Abstract

This invention discloses a deep learning-based teaching system for the inheritance of traditional Chinese medicine schools, belonging to the field of artificial intelligence technology. It includes: a data acquisition and preprocessing module, a three-level vertical deep feature extraction module, a traditional Chinese medicine school inheritance teaching model training module, a personalized teaching interaction module, and a teaching effect evaluation and model iteration module. The data acquisition and preprocessing module is used to collect multi-source heterogeneous raw data related to traditional Chinese medicine schools, perform standardized preprocessing, and output a standardized dataset that can be used for feature extraction. It breaks through the bottlenecks of traditional inheritance models, realizing the universal accessibility of high-quality resources: it overcomes the inherent limitations of the traditional "master-apprentice" model, such as low inheritance efficiency, insufficient standardization, easy fragmentation and loss of academic thought, and severe restrictions imposed by geographical location and mentorship resources. It digitizes and systematically integrates scarce traditional Chinese medicine school inheritance resources, achieving the universal dissemination of high-quality academic resources without geographical barriers, and significantly improving the coverage and efficiency of traditional Chinese medicine school inheritance.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a teaching system for the inheritance of traditional Chinese medicine based on deep learning. Background Technology

[0002] Schools of traditional Chinese medicine are the core carriers of the academic development of traditional Chinese medicine. Different schools (such as the Typhoid Fever School, the Febrile Disease School, the Hejian School, and the Yishui School) have formed their own unique academic thoughts, diagnostic logic systems, and clinical medication characteristics, which are the core treasure trove for the inheritance and innovation of traditional Chinese medicine.

[0003] The current inheritance and teaching of traditional Chinese medicine suffer from the following core pain points and technical deficiencies:

[0004] Traditional transmission models have inherent limitations: the mainstream "master-apprentice" oral transmission model suffers from problems such as low transmission efficiency, insufficient standardization, easy fragmentation and loss of academic ideas, and severe limitations due to geographical and mentorship resources, making it impossible to achieve universal dissemination of high-quality academic transmission resources.

[0005] The existing digital teaching systems are superficial in function: Most existing TCM digital teaching systems are mainly based on the accumulation of resources such as electronic medical books and teaching videos. They can only realize basic content display and retrieval. They cannot deeply explore and systematically present the core academic ideas, diagnostic logic system, and clinical medication characteristics of different TCM schools. They are unable to restore the core thinking path of the school's inheritance and cannot achieve the goal of "teaching a man to fish".

[0006] The application of artificial intelligence technology suffers from both gaps in application scenarios and technical deficiencies: Existing AI applications in the field of Traditional Chinese Medicine (TCM) are mostly concentrated in clinical scenarios such as assisted diagnosis and prescription recommendation, with very few dedicated AI systems for the teaching and inheritance of TCM schools of thought. Furthermore, current technologies for feature extraction from TCM text data often employ single-level or multi-branch parallel feature extraction architectures, which cannot adapt to the three-tiered progressive structure of TCM school knowledge—"basic entities - systemic connections - thought inheritance"—failing to achieve hierarchical feature transmission and in-depth mining. This makes it difficult to accurately capture the core academic differences and key points of inheritance among different schools, resulting in a lack of targeted teaching content and insufficient personalized teaching capabilities, hindering the accurate, efficient, and systematic inheritance of TCM school academic thought.

[0007] To address the many shortcomings of existing technologies, this invention proposes a deep learning-based teaching system for the inheritance of traditional Chinese medicine schools. Through an innovative three-level vertically progressive deep feature extraction architecture, it accurately mines the hierarchical knowledge and thinking characteristics of traditional Chinese medicine schools, constructs a unique inheritance teaching model, and solves the core pain points of existing technologies. Summary of the Invention

[0008] Purpose of the invention: The purpose of this invention is to provide a teaching system for the inheritance of traditional Chinese medicine based on deep learning; it can solve the problems of the inherent limitations of the traditional inheritance model, the superficial functions of the existing digital teaching system, and the lack of application scenarios and technical defects of artificial intelligence technology in the existing technology.

[0009] Technical solution: To solve the above-mentioned technical problems, according to one aspect of the present invention, more specifically, a deep learning-based traditional Chinese medicine school inheritance teaching system, including: a data acquisition and preprocessing module, a three-level vertical deep feature extraction module, a traditional Chinese medicine school inheritance teaching model training module, a personalized teaching interaction module, and a teaching effect evaluation and model iteration module;

[0010] Data acquisition and preprocessing module: used to collect multi-source heterogeneous raw data related to traditional Chinese medicine schools, perform standardized preprocessing, and output a standardized dataset that can be used for feature extraction;

[0011] The three-level vertical depth feature extraction module is used to extract basic semantic and TCM entity features from the preprocessed standardized text dataset of the Chinese Medical School and output the first-level features. Based on the first-level features, the module extracts the core knowledge system association features of the Chinese Medical School and outputs the second-level features. Based on the second-level features, the module extracts the inheritance thinking path and teaching adaptation features of the Chinese Medical School and outputs the third-level features.

[0012] Training module for the teaching model of the inheritance of traditional Chinese medicine schools: used to construct and train a multi-task fusion teaching model of the inheritance of traditional Chinese medicine schools based on third-level features;

[0013] Personalized teaching interaction module: Used to provide learners with a full-process teaching service for the inheritance of traditional Chinese medicine schools based on the completed training teaching model.

[0014] Teaching effectiveness evaluation and model iteration module: used to collect teaching data and evaluate its effectiveness, and to iteratively optimize the system based on the evaluation results.

[0015] Furthermore, the raw data collected by the data acquisition and preprocessing module includes, but is not limited to, original medical texts from various dynasties, medical records of famous doctors, clinical records, academic works, lecture manuscripts, audio and video transcripts, records of prescriptions and drug combinations, cases of syndrome differentiation and treatment, and historical materials on the academic lineage of schools of thought. The preprocessing includes: data cleaning, text standardization, TCM-specific word segmentation, and dataset partitioning. Data cleaning: deduplication, noise reduction, missing value completion, irrelevant content removal, and filtering of garbled characters and special symbols from the raw data. Text standardization: conversion of traditional Chinese characters to simplified Chinese characters and standardization of variant characters and ancient and modern characters. Based on the national standards "Classification and Code of Diseases and Syndromes in Traditional Chinese Medicine" and "Clinical Terminology of Traditional Chinese Medicine," standardized alignment of TCM terminology is achieved, solving the problems of synonyms and polysemy in TCM terminology. TCM-specific word segmentation: accurate word segmentation of TCM terminology is achieved through a word segmentation model fine-tuned based on a large-scale TCM corpus, combined with a dictionary of terminology specific to the target school. Dataset partitioning: the preprocessed dataset is divided into training, validation, and test sets according to a preset ratio for subsequent model training and validation.

[0016] Furthermore, the three-level vertical depth feature extraction module includes: a first-level feature extraction unit, a second-level feature extraction unit, and a third-level feature extraction unit;

[0017] The first-level feature extraction unit employs a TCM BERT pre-trained language model based on the full TCM corpus of the "Chinese Medical Encyclopedia," with fine-tuning for the specific corpus of the target TCM school. The fine-tuned model performs token-level encoding on the input text, extracting the basic semantic embedding vectors of the text. Simultaneously, through the built-in Named Entity Recognition (NER) submodule, it accurately identifies and extracts core TCM entities from the text, including but not limited to disease names, syndrome types, etiologies and pathogenesis, treatment principles and methods, names of Chinese medicines, names of prescriptions, properties and meridian tropism, names of acupoints, names of physicians, names of schools of thought, and keywords of academic propositions. The output is the first-level feature.

[0018] The second-level feature extraction unit: Based on the entity and semantic features in the first-level features, it constructs a knowledge graph specific to the target Chinese medicine school, using the extracted core TCM entities as graph nodes and the semantic associations and logical causal relationships between entities as graph edges. The graph edges include, but are not limited to, the correspondence between "syndrome type and treatment principle," the compatibility between "prescription and Chinese medicine," the causal relationship between "etiology and pathogenesis," the attribution relationship between "physician" and "academic proposition," and the lineage relationship between "school of thought" and "historical transmission." The constructed knowledge graph of the Chinese medicine school is trained end-to-end using a graph attention network (GAT). Through the attention mechanism, the weights of different entity associations are adaptively adjusted to accurately extract the embedding features of nodes, the association features of edges, and the structural features of subgraphs in the graph. This extracts systematic knowledge features, including but not limited to the core academic system, dialectical logic framework, prescription compatibility rules, and academic transmission lineage of the target Chinese medicine school, and outputs these as the second-level features.

[0019] The third-level feature extraction unit employs a Transformer decoder architecture with a masking mechanism, combined with a temporal convolutional network (TCN), to deeply encode the second-level features. Through a masking learning mechanism, it simulates the entire process of syndrome differentiation and treatment thinking, including etiology collection, pathogenesis analysis, syndrome differentiation judgment, treatment principle determination, prescription compatibility, and modification. It extracts the syndrome differentiation thinking path features specific to the school of thought, and divides the learning progression nodes from beginner to intermediate to advanced levels based on the knowledge system hierarchy of traditional Chinese medicine. It extracts the knowledge weight features and learning path features corresponding to different learning stages, and extracts teaching adaptation features corresponding to different learning objectives, learning foundations, and learning progress, based on the needs of teaching scenarios. The output is the third-level features.

[0020] Furthermore, the training module of the National Medical School Inheritance Teaching Model adopts a multi-task learning architecture, sharing underlying feature weights and adapting to different teaching sub-tasks, including: academic thought interpretation sub-task, dialectical thinking simulation sub-task, learning path planning sub-task, clinical teaching training sub-task, and school knowledge question-and-answer sub-task. The model training strategy uses third-level features as the core input, and standardized teaching content corresponding to the National Medical School, clinical thinking annotation data of famous doctors, and teaching effect feedback data as labels. It uses a joint loss function combining cross-entropy loss function and contrastive learning loss function to train the model end-to-end. The model hyperparameters are adjusted through the validation set, and the generalization ability of the model is verified through the test set, finally obtaining the trained National Medical School Inheritance Teaching Model.

[0021] Furthermore, the personalized teaching interaction module includes: a unit for systematic interpretation of the core academic ideas of the school of thought, a unit for teaching dialectical thinking in a step-by-step manner, a unit for personalized learning path planning, and a unit for intelligent question answering and clinical simulation.

[0022] The unit on the systematic interpretation of the core academic thought of a school of thought: This unit provides a systematic and hierarchical interpretation of the academic origins, core propositions, lineage of medical practitioners throughout history, representative medical books and case studies of the target school of thought, and fully presents the academic development and core characteristics of the school of thought.

[0023] The step-by-step diagnostic thinking teaching unit restores the diagnostic and treatment thinking path of the medical schools and provides learners with step-by-step thinking training from beginner to expert, including training in pathogenesis analysis, syndrome identification, treatment principle and method formulation, and prescription and drug compatibility and modification.

[0024] Personalized learning path planning unit: Based on the learner's learning foundation, learning progress, and test results, a unique learning path and teaching content are matched.

[0025] Intelligent Q&A and Clinical Simulation Unit: Provides 24 / 7 intelligent Q&A service with exclusive knowledge of the school of thought, as well as highly realistic clinical scenario-based training in syndrome differentiation and treatment, providing real-time feedback on learning effectiveness and targeted optimization suggestions.

[0026] Furthermore, the teaching effectiveness evaluation and model iteration module collects learners' learning data, test results, and teaching feedback to construct a teaching effectiveness evaluation index system that includes three core dimensions: knowledge mastery, dialectical thinking ability, and clinical application ability. Based on the evaluation results, the parameters of the three-level vertical deep feature extraction module and the weights of the inheritance teaching model are iteratively optimized to continuously improve the system's teaching effectiveness and scenario adaptability, forming a complete closed loop of "teaching-evaluation-optimization".

[0027] According to another aspect of the present invention, a deep learning-based teaching method for the inheritance of traditional Chinese medicine schools is provided. This method is implemented based on the aforementioned deep learning-based teaching system for the inheritance of traditional Chinese medicine schools, and specifically includes the following steps:

[0028] S1. Collect multi-source raw data related to the Chinese medical school and perform standardized preprocessing to obtain a standardized Chinese medical school text dataset.

[0029] S2. A unidirectional, progressive, three-level vertical deep feature extraction architecture is adopted to perform hierarchical feature extraction on the standardized text dataset, specifically as follows:

[0030] S21. First-level feature extraction: Extract basic semantic and TCM entity features from the standardized text dataset and output the first-level features;

[0031] S22. Second-level feature extraction: Based on the first-level features, extract the associated features of the core knowledge system of traditional Chinese medicine and output the second-level features;

[0032] S23. Third-level feature extraction: Based on the second-level features, extract the inheritance thinking path and teaching adaptation features of traditional Chinese medicine schools, and output the third-level features;

[0033] S3. Based on the third-level features, construct and train a multi-task fusion teaching model for the inheritance of traditional Chinese medicine schools;

[0034] S4. Based on the completed training model for the inheritance of traditional Chinese medicine schools, provide learners with teaching services for the inheritance of traditional Chinese medicine schools.

[0035] S5. Collect teaching data and evaluate teaching effectiveness, and iteratively optimize the feature extraction architecture and teaching model based on the evaluation results.

[0036] Furthermore, in step S21, a BERT model pre-trained on a large-scale corpus of traditional Chinese medicine is used to fine-tune the corpus specific to the target country's medical school, extract the basic semantic features and core entity features of traditional Chinese medicine, and output the first-level features.

[0037] Furthermore, in step S22, firstly, based on the entity and semantic features in the first-level features, a unique knowledge graph of the target Chinese medical school is constructed. Then, a graph attention network is used to train the knowledge graph, extract systematic knowledge association features, and output the second-level features.

[0038] Furthermore, in step S23, a Transformer decoder with a masking mechanism is used in conjunction with a temporal convolutional network to extract features of dialectical thinking path, learning progression features, and teaching adaptation features, and output the third-level features.

[0039] Beneficial effects: Breaking through the bottlenecks of traditional inheritance models and realizing the universal access to high-quality resources: It breaks through the natural limitations of the traditional "master-apprentice" model, which is characterized by low inheritance efficiency, insufficient standardization, easy fragmentation and loss of academic ideas, and severe restrictions by region and mentorship resources. It digitizes and systematizes the scarce traditional Chinese medicine school inheritance resources, realizes the universal dissemination of high-quality academic resources without regional barriers, and greatly improves the coverage and efficiency of traditional Chinese medicine school inheritance.

[0040] This upgrade transforms teaching from "content accumulation" to "the inheritance of thought": It solves the core problem that existing TCM digital teaching systems can only display resources and have superficial functions. It not only systematically presents the academic origins, core propositions and inheritance of the TCM school, but also fully restores the school's unique diagnostic and treatment thinking path through in-depth technology mining, truly realizing the inheritance of the core thinking of the TCM school and achieving the teaching goal of "teaching a man to fish".

[0041] Filling the gap in AI applications and addressing the deficiencies in feature extraction technology: This invention fills the gap in the specialized application of AI technology in the field of traditional Chinese medicine school inheritance and teaching. Through a three-level vertically progressive deep feature extraction architecture, it perfectly adapts to the three-layer progressive knowledge structure of traditional Chinese medicine schools, namely "basic entities - system association - thought inheritance". It realizes hierarchical feature transfer and in-depth mining, which can accurately capture the core academic differences and inheritance points of different schools. It solves the technical deficiency of existing single-level / multi-branch parallel feature extraction architectures that cannot deeply mine the core characteristics of schools.

[0042] We construct a closed-loop teaching system that balances standardization and personalization: Based on a multi-task integrated learning architecture, it adapts to the teaching needs of various scenarios. At the same time, through personalized teaching interaction modules, it can match exclusive learning paths and content based on learners' foundation, progress, and goals. Through a complete closed-loop mechanism of "teaching-assessment-optimization", the system model can be continuously iterated and optimized based on teaching effectiveness. While ensuring the standardization and high-fidelity inheritance of the school's academic thought, it fully meets the personalized learning needs of different learners.

[0043] To achieve the replicable inheritance and innovation support of the academic thought of different schools of traditional Chinese medicine: Through standardized data preprocessing and precise feature extraction, the core academic knowledge and dialectical thinking characteristics of different schools of traditional Chinese medicine are fully preserved, avoiding the fragmentation and loss of academic knowledge during the inheritance process; at the same time, the systematic digital resources and AI models also provide standardized data and technical support for the subsequent academic innovation, cross-school comparative research and clinical application expansion of schools of traditional Chinese medicine. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of the system principle;

[0045] Figure 2 This is a flowchart illustrating the method. Detailed Implementation

[0046] To make the technical solution of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0047] Example 1

[0048] The deep learning-based teaching system for the inheritance of the Typhoid Fever School, as described in this embodiment, fully comprises the five core modules of this invention: a data acquisition and preprocessing module, a three-level vertical deep feature extraction module, a Typhoid Fever School inheritance teaching model training module, a personalized teaching interaction module, and a teaching effectiveness evaluation and model iteration module. Each module achieves unidirectional, progressive data transmission through standardized data interfaces, while simultaneously enabling bidirectional parameter optimization and model iteration through the evaluation and iteration module, forming a complete system closed loop.

[0049] II. Specific Implementation Details of Each Core Module

[0050] 1. Implementation of the data acquisition and preprocessing module

[0051] This module is designed for the typhoid fever school of thought, completing the acquisition of multi-source heterogeneous data and standardized preprocessing throughout the entire process. The specific implementation is as follows:

[0052] Original Data Collection: Collect all relevant original data from the Shanghan School, including: ① Original medical texts from various dynasties: original texts of "Shanghan Lun" and "Jinkui Yaolue" as well as annotations by famous scholars such as Cheng Wuji's "Annotations on Shanghan Lun", Ke Qin's "Shanghan Laisu Ji", and You Yi's "Shanghan Guanzhu Ji"; ② Medical Cases and Clinical Records of Famous Physicians: clinical medical cases, records of the application of classical prescriptions, and complete cases of syndrome differentiation and treatment by physicians from the Shanghan School such as Zhang Zhongjing, Cheng Wuji, Fang Youzhi, Yu Jiayan, and Ke Qin; ③ Academic Works and Historical Materials on Transmission: academic works, academic propositions, and historical materials on the academic transmission of the Shanghan School, including the schools of error correction, maintenance of old theories, and syndrome differentiation and treatment; ④ Teaching-related Data: lecture manuscripts, teaching materials, transcribed texts of audio and video courses, records of classical prescription combinations, and data from the syndrome differentiation training question bank.

[0053] Data cleaning: The collected raw text data is deduplicated to remove duplicate medical records and annotations; noise reduction is performed to remove prefaces, postscripts, and non-medical content that are unrelated to the academic content of the Typhoid Fever School; missing medical record diagnosis records and prescription information are manually annotated and supplemented; and garbled characters, special symbols, and non-Chinese characters are filtered out.

[0054] Text standardization: First, the traditional Chinese characters were converted to simplified Chinese characters. Variant characters and ancient and modern characters in ancient books were normalized and unified into the standard characters used in modern Chinese. Based on the national standards "Classification and Code of Diseases and Syndromes in Traditional Chinese Medicine" and "Clinical Diagnosis and Treatment Terminology in Traditional Chinese Medicine", the TCM terms in the text were standardized and aligned. Synonyms of terms such as "Taiyang disease" and "Yangming disease" were normalized to solve the problem of multiple synonyms and multiple meanings of TCM terms. The "Standardized Terminology Database of the Typhoid Fever School" was constructed.

[0055] Traditional Chinese Medicine (TCM) Specialized Word Segmentation: The BERT word segmentation model, finely tuned based on the full TCM corpus of the "Chinese Medical Encyclopedia," is combined with the "Dictionary of Exclusive Terms of the Shanghan School" (covering terms related to the Six Channels Differentiation, names of classic prescriptions, and keywords of the school's exclusive academic propositions) constructed in this embodiment to accurately segment standardized text, avoiding the incorrect segmentation of TCM specialized terms by general word segmentation models, and fully recognizing exclusive specialized terms such as "Taiyang Zhongfeng Syndrome," "Mahuang Decoction," and "Six Channels Differentiation."

[0056] Dataset partitioning: The preprocessed Typhoid Fever School text dataset is divided into training set, validation set, and test set according to a preset ratio of 8:1:1, which are used for subsequent model training, hyperparameter tuning, and model generalization ability verification, respectively.

[0057] 2. Implementation of the three-level vertical depth feature extraction module

[0058] This module strictly adopts the unidirectional progressive three-level vertical architecture described in this invention, and completes feature extraction in three units step by step. The specific implementation is as follows:

[0059] First-level feature extraction unit (basic semantics and TCM entity feature extraction)

[0060] The TCM BERT pre-trained language model, which is pre-trained based on the full TCM corpus of "Chinese Medical Classics", was used. It was then fine-tuned based on the corpus-specific language set of the Shanghan School in this embodiment. The fine-tuning epoch was set to 10, the learning rate was set to 2e-5, and the batch_size was set to 32 to complete the model fine-tuning.

[0061] The TCM BERT model, after fine-tuning, performs token-level encoding on the input standardized text, extracting a 768-dimensional textual basic semantic embedding vector. Simultaneously, through the model's built-in Named Entity Recognition (NER) submodule, it accurately identifies and extracts core TCM entities from the Shanghan School, including disease names (such as Taiyang disease and Yangming disease, which fall under the Six Channels), syndrome types (such as Taiyang Zhongfeng syndrome and Yangming Fushi syndrome), etiology and pathogenesis, treatment principles and methods, names of Chinese medicines, names of prescriptions, names of physicians, and keywords of academic propositions. The semantic embedding vector is then fused with the entity features to output the first-level features.

[0062] Second-level feature extraction unit (core knowledge system related feature extraction)

[0063] The first step involves constructing a "Knowledge Graph Exclusive to the Shanghan School" based on the core TCM entities and semantic features extracted from the first-level features, using the identified core TCM entities of the Shanghan School as graph nodes and the semantic associations and logical causal relationships between entities as graph edges. Specifically, the graph edges include: the correspondence between "syndrome type - treatment principle", the compatibility between "prescription - Chinese herbal medicine", the causal relationship between "etiology - pathogenesis - syndrome type", the attribution relationship between "physicians - academic propositions", the lineage relationship between "school of thought - historical inheritance", and the correspondence between "prescription and syndrome - modification and alteration".

[0064] The second step involves end-to-end training of the constructed knowledge graph of the Shanghan School using a graph attention network (GAT). The GAT network is configured with two attention layers, each with eight attention heads, and the output feature dimension is set to 256. Through the attention mechanism, the weights of different entity associations are adaptively adjusted. For the core "Six Channels Differentiation" system of the Shanghan School, the association weights between "disease-syndrome-prescription-medicine" are strengthened. The embedding features of nodes, the association features of edges, and the structural features of subgraphs in the graph are accurately extracted. This allows for the extraction of the systematic knowledge features of the core academic system, the diagnostic logic framework, the rules of prescription compatibility, and the academic inheritance of the Shanghan School, and the output of the second-level features.

[0065] The third-level feature extraction unit (inherited thinking path and teaching adaptation feature extraction)

[0066] A Transformer decoder architecture with a masking mechanism is adopted, combined with a temporal convolutional network (TCN) to perform deep encoding of the second-level features. The Transformer decoder is set to 6 layers, and the masking mechanism adopts causal masking to simulate the temporal process of dialectical treatment. The TCN network is set to 3 layers of dilated convolution with dilation coefficients of 1, 2 and 4 to capture long-distance thought path dependence.

[0067] Through a masked learning mechanism, the entire diagnostic and treatment thought process of the Shanghan Lun school—"etiology collection → pathogenesis analysis → six-channel syndrome differentiation → treatment principle determination → classical prescription compatibility → modification and adjustment"—is simulated to extract the unique characteristics of the six-channel syndrome differentiation thought process of the Shanghan Lun school. Based on the knowledge system hierarchy of the Shanghan Lun school, learning progression nodes from beginner to intermediate to advanced levels are divided, and knowledge weight features and learning path features corresponding to different learning stages are extracted. Combined with the needs of teaching scenarios, teaching adaptation features corresponding to different learning objectives, learning foundations, and learning progress are extracted, and finally, the third-level features are output.

[0068] 3. Implementation of the Training Module of the Teaching Model for the Tradition of the Typhoid Fever School

[0069] This module adopts the multi-task learning architecture described in this invention, sharing underlying feature weights, and is adapted to 5 teaching sub-tasks: academic thought interpretation sub-task, dialectical thinking simulation sub-task, learning path planning sub-task, clinical teaching training sub-task, and school of thought knowledge Q&A sub-task.

[0070] Model training implementation: The third-level features are used as the core input, and the standardized teaching content of the Typhoid Fever School, the clinical thinking annotation data of famous doctors, and the teaching effect feedback data are used as the labels for the corresponding sub-tasks. A joint loss function combining the cross-entropy loss function and the contrastive learning loss function is adopted, with the cross-entropy loss weight set to 0.7 and the contrastive learning loss weight set to 0.3. The model is trained end-to-end.

[0071] During training, the epoch was set to 20, the optimizer was AdamW, the learning rate was set to 3e-5, and the batch size was set to 16. Hyperparameters such as the number of network layers, the number of attention heads, and the learning rate were adjusted using the validation set. The generalization ability of the model was verified using the test set. When the accuracy of each subtask on the test set reached more than 90%, the model training was completed, and the final usable teaching model for the inheritance of the Shanghan School was obtained.

[0072] 4. Implementation of Personalized Interactive Teaching Modules

[0073] This module, based on a trained teaching model for the transmission of the Typhoid Fever School, provides learners with a full-process transmission teaching service. It consists of four core units, and the specific implementation is as follows:

[0074] Systematic Interpretation of the Core Academic Thoughts of the Typhoid Fever School

[0075] This book provides a systematic and hierarchical visual interpretation of the academic origins, core tenets (the Six Channels Differentiation System, the theory of formula-symptom correspondence, etc.), lineage of physicians throughout history, characteristics of subordinate branches and schools, representative medical books, and classic medical cases of the Typhoid Fever School. It presents the academic lineage of the Typhoid Fever School from the Han Dynasty to the Qing Dynasty using a timeline, the complete system of the Six Channels Differentiation using mind maps, and the core academic tenets and debates of physicians throughout history using structured content, thus comprehensively showcasing the academic development and core characteristics of the Typhoid Fever School.

[0076] Step-by-step dialectical thinking teaching unit

[0077] This course reconstructs the thought process of the Six Channels Differentiation and Treatment of the Shanghan School of Medicine, providing learners with a step-by-step training approach from beginner to expert. The beginner stage includes basic training in pathogenesis analysis and identification of Six Channels syndromes; the intermediate stage includes training in formulating treatment principles and methods and the rules of classical prescription compatibility; and the expert stage includes training in modifying classical prescriptions and developing diagnostic thinking skills for difficult medical cases. It comprehensively covers all the thought processes and key points of the Shanghan School's diagnostic and treatment approach, allowing learners to gradually master the core diagnostic thinking of the Shanghan School through step-by-step training.

[0078] Personalized learning path planning unit

[0079] First, we collect learners' initial information, including their learning background (zero knowledge / TCM enthusiasts / TCM students / licensed TCM practitioners), learning goals (interest-based learning / professional exams / apprenticeship learning / clinical skills improvement), and available learning time. We obtain the learners' knowledge baseline through initial assessment. Based on the baseline data, we match learners with personalized learning paths and teaching content. At the same time, we dynamically adjust the learning content and pace based on the learners' real-time learning progress and unit test results.

[0080] Intelligent Question Answering and Clinical Simulation Unit

[0081] Built-in 24 / 7 intelligent question-and-answer engine, based on a trained model, provides learners with Q&A services specific to the Shanghan Lun (Treatise on Cold Damage) school of thought. It can accurately answer various questions regarding the original text of the Shanghan Lun, medical theories, key points of the Six Channels Differentiation, contraindications of classical prescriptions, and case analysis. Simultaneously, it features highly realistic simulated clinical scenarios and a multi-disciplinary clinical case library related to the Shanghan Lun school, simulating real-world clinical situations. This allows learners to complete the entire process of clinical diagnosis training. The system provides real-time scoring and feedback on operations, pointing out deviations in diagnostic thinking and offering targeted optimization suggestions.

[0082] 5. Implementation of the Teaching Effectiveness Evaluation and Model Iteration Module

[0083] This module continuously collects all learning data from learners within the system, including learning duration, content completion rate, unit test results, clinical simulation training scores, Q&A interaction data, and teaching satisfaction feedback, to construct the three-core-dimensional teaching effectiveness evaluation index system described in this invention:

[0084] Knowledge mastery dimension: It includes four secondary indicators: mastery of the original text, understanding of academic ideas, accuracy of core terms, and mastery of prescription compatibility.

[0085] The dimension of dialectical thinking ability includes three secondary indicators: accuracy rate of pathogenesis analysis, accuracy rate of syndrome differentiation, and matching degree of treatment principle formulation.

[0086] Clinical application ability dimension: It includes three secondary indicators: completeness of clinical diagnosis, rationality of prescription and drug combination, and suitability of addition, subtraction and modification.

[0087] The weights of each indicator are determined using the analytic hierarchy process (AHP), and learners' learning outcomes are quantitatively scored to generate personalized learning effectiveness evaluation reports. Simultaneously, evaluation data from all learners is aggregated to analyze weaknesses in the system's teaching content and model. Based on the evaluation results, the network parameters of the three-level vertical deep feature extraction module and the weights of the inherited teaching model are fine-tuned and iterated to continuously improve the system's teaching effectiveness and scenario adaptability, forming a complete closed loop of "teaching-evaluation-optimization".

[0088] III. Implementation Process of Corresponding Teaching Methods

[0089] The deep learning-based teaching method for the inheritance of the Typhoid Fever School in this embodiment is fully implemented based on the above system and strictly follows the steps described in this invention. The specific process is as follows:

[0090] S1. Collect multi-source raw data related to the Typhoid Fever School, complete data cleaning, text standardization, TCM professional word segmentation, and dataset partitioning as standardized preprocessing to obtain a standardized Typhoid Fever School text dataset;

[0091] S2. A unidirectional, progressive, three-level vertical deep feature extraction architecture is adopted to perform hierarchical feature extraction on standardized text datasets:

[0092] S21. First-level feature extraction: The BERT model, pre-trained on a large-scale TCM corpus, is fine-tuned for the corpus specific to the Shanghan School to extract the basic semantic features and core TCM entity features of the text and output the first-level features.

[0093] S22. Second-level feature extraction: Based on the first-level features, construct a knowledge graph specific to the Typhoid Fever School, train the knowledge graph using a graph attention network (GAT), extract systematic knowledge association features, and output the second-level features;

[0094] S23. Third-level feature extraction: Using a Transformer decoder with masking mechanism combined with a temporal convolutional network (TCN), the features of the diagnostic thinking path of the Typhoid Fever School, learning progression features, and teaching adaptation features are extracted, and the third-level features are output.

[0095] S3. Based on the third-level features, a multi-task learning architecture teaching model for the inheritance of the Typhoid Fever School is constructed. The joint loss function is used for end-to-end training. The generalization ability is verified by adjusting the parameters on the validation set and verifying the test set, and the trained teaching model is obtained.

[0096] S4. Based on the trained model, through personalized teaching interaction modules, learners are provided with a full-process teaching service including systematic academic interpretation, step-by-step thinking training, personalized learning path planning, intelligent question answering and clinical simulation.

[0097] S5. Collect all teaching data and evaluate teaching effectiveness based on a three-dimensional indicator system of knowledge mastery, dialectical thinking ability, and clinical application ability. Based on the evaluation results, iteratively optimize the feature extraction architecture and teaching model to form a complete closed loop.

[0098] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A deep learning-based teaching system for the inheritance of traditional Chinese medicine schools, characterized in that: include: The module includes: data acquisition and preprocessing module, three-level vertical depth feature extraction module, traditional Chinese medicine school inheritance teaching model training module, personalized teaching interaction module, and teaching effect evaluation and model iteration module. Data acquisition and preprocessing module: used to collect multi-source heterogeneous raw data related to traditional Chinese medicine schools, perform standardized preprocessing, and output a standardized dataset that can be used for feature extraction; The three-level vertical depth feature extraction module is used to extract basic semantic and TCM entity features from the preprocessed standardized text dataset of the Chinese Medical School and output the first-level features. Based on the first-level features, the module extracts the core knowledge system association features of the Chinese Medical School and outputs the second-level features. Based on the second-level features, the module extracts the inheritance thinking path and teaching adaptation features of the Chinese Medical School and outputs the third-level features. Training module for the teaching model of the inheritance of traditional Chinese medicine schools: used to construct and train a multi-task fusion teaching model of the inheritance of traditional Chinese medicine schools based on third-level features; Personalized teaching interaction module: Used to provide learners with a full-process teaching service for the inheritance of traditional Chinese medicine schools based on the completed training teaching model. Teaching effectiveness evaluation and model iteration module: used to collect teaching data and evaluate its effectiveness, and to iteratively optimize the system based on the evaluation results.

2. The deep learning-based traditional Chinese medicine school inheritance teaching system according to claim 1, characterized in that: The raw data collected by the data acquisition and preprocessing module includes, but is not limited to, original medical texts from various dynasties, medical records of famous doctors, clinical records, academic works, lecture manuscripts, audio and video transcripts, records of prescriptions and drug combinations, cases of syndrome differentiation and treatment, and historical materials on the academic lineage of schools of thought. The preprocessing includes: data cleaning, text standardization, TCM-specific word segmentation, and dataset partitioning. Data cleaning: deduplication, noise reduction, missing value completion, irrelevant content removal, and filtering of garbled characters and special symbols from the raw data. Text standardization: conversion of traditional Chinese characters to simplified Chinese characters and standardization of variant characters and ancient and modern characters. Based on the corresponding national standards, standardization and alignment of TCM terminology are achieved, solving the problems of synonyms and polysemy in TCM terminology. TCM-specific word segmentation: accurate word segmentation of TCM terminology is achieved through a word segmentation model finely tuned based on a large-scale TCM corpus, combined with a dictionary of terminology specific to the target school. Dataset partitioning: the preprocessed dataset is divided into training, validation, and test sets according to a preset ratio for subsequent model training and validation.

3. The deep learning-based traditional Chinese medicine school inheritance teaching system according to claim 1, characterized in that: The three-level vertical depth feature extraction module includes: a first-level feature extraction unit, a second-level feature extraction unit, and a third-level feature extraction unit; The first-level feature extraction unit employs a TCM BERT pre-trained language model based on a full TCM corpus of relevant literature, fine-tuned for the specific corpus of the target TCM school. The fine-tuned model performs token-level encoding on the input text, extracting the basic semantic embedding vector of the text. Simultaneously, through the built-in Named Entity Recognition (NER) submodule, it accurately identifies and extracts core TCM entities in the text, including but not limited to disease names, syndrome types, etiologies and pathogenesis, treatment principles and methods, names of Chinese medicines, names of prescriptions, properties and meridian tropism, names of acupoints, names of physicians, names of schools of thought, and keywords of academic propositions. The output is the first-level feature. The second-level feature extraction unit: Based on the entity and semantic features in the first-level features, it constructs a knowledge graph specific to the target Chinese medicine school, using the extracted core TCM entities as graph nodes and the semantic associations and logical causal relationships between entities as graph edges. The graph edges include, but are not limited to, the correspondence between "syndrome type and treatment principle," the compatibility between "prescription and Chinese medicine," the causal relationship between "etiology and pathogenesis," the attribution relationship between "physician" and "academic proposition," and the lineage relationship between "school of thought and historical inheritance." The constructed knowledge graph of the Chinese medicine school is trained end-to-end using a graph attention network (GAT). Through the attention mechanism, the weights of different entity associations are adaptively adjusted to accurately extract the embedding features of nodes, the association features of edges, and the structural features of subgraphs in the graph. This extracts systematic knowledge features, including but not limited to the core academic system, dialectical logic framework, prescription and medicine compatibility rules, and academic inheritance lineage of the target Chinese medicine school, and outputs these as the second-level features. The third-level feature extraction unit employs a Transformer decoder architecture with a masking mechanism, combined with a temporal convolutional network (TCN), to deeply encode the second-level features. Through a masking learning mechanism, it simulates the entire process of syndrome differentiation and treatment thinking, including etiology collection, pathogenesis analysis, syndrome differentiation judgment, treatment principle determination, prescription compatibility, and modification. It extracts the syndrome differentiation thinking path features specific to the school of thought, and divides the learning progression nodes from beginner to intermediate to advanced levels based on the knowledge system hierarchy of traditional Chinese medicine. It extracts the knowledge weight features and learning path features corresponding to different learning stages, and extracts teaching adaptation features corresponding to different learning objectives, learning foundations, and learning progress, based on the needs of teaching scenarios. The output is the third-level features.

4. The deep learning-based traditional Chinese medicine school inheritance teaching system according to claim 1, characterized in that: The training module of the National Medical School Inheritance Teaching Model adopts a multi-task learning architecture, sharing underlying feature weights and adapting to different teaching sub-tasks, including: academic thought interpretation sub-task, dialectical thinking simulation sub-task, learning path planning sub-task, clinical teaching training sub-task, and school knowledge question-and-answer sub-task. The model training strategy uses third-level features as the core input, and standardized teaching content corresponding to the National Medical School, clinical thinking annotation data of famous doctors, and teaching effect feedback data as labels. It uses a joint loss function combining cross-entropy loss function and contrastive learning loss function to train the model end-to-end. The model hyperparameters are adjusted through the validation set, and the generalization ability of the model is verified through the test set, finally obtaining the trained National Medical School Inheritance Teaching Model.

5. The deep learning-based traditional Chinese medicine school inheritance teaching system according to claim 1, characterized in that: The personalized teaching interaction module includes: a unit for systematic interpretation of the core academic ideas of the school of thought, a unit for teaching dialectical thinking in a step-by-step manner, a unit for personalized learning path planning, and a unit for intelligent question answering and clinical simulation. The unit on the systematic interpretation of the core academic thought of a school of thought: This unit provides a systematic and hierarchical interpretation of the academic origins, core propositions, lineage of medical practitioners throughout history, representative medical books and case studies of the target school of thought, and fully presents the academic development and core characteristics of the school of thought. The step-by-step diagnostic thinking teaching unit restores the diagnostic and treatment thinking path of the medical schools and provides learners with step-by-step thinking training from beginner to expert, including training in pathogenesis analysis, syndrome identification, treatment principle and method formulation, and prescription and drug compatibility and modification. Personalized learning path planning unit: Based on the learner's learning foundation, learning progress, and test results, a unique learning path and teaching content are matched. Intelligent Q&A and Clinical Simulation Unit: Provides 24 / 7 intelligent Q&A service with exclusive knowledge of the school of thought, as well as highly realistic clinical scenario-based training in syndrome differentiation and treatment, providing real-time feedback on learning effectiveness and targeted optimization suggestions.

6. The deep learning-based traditional Chinese medicine school inheritance teaching system according to claim 1, characterized in that: The teaching effectiveness evaluation and model iteration module collects learners' learning data, test results, and teaching feedback to construct a teaching effectiveness evaluation index system that includes three core dimensions: knowledge mastery, dialectical thinking ability, and clinical application ability. Based on the evaluation results, the parameters of the three-level vertical depth feature extraction module and the weights of the inheritance teaching model are iteratively optimized to continuously improve the system's teaching effectiveness and scenario adaptability, forming a complete closed loop of "teaching-evaluation-optimization".

7. A teaching method for the inheritance of traditional Chinese medicine schools based on deep learning, characterized by: This method is implemented based on the deep learning-based traditional Chinese medicine school inheritance teaching system described in any one of claims 1-6, and specifically includes the following steps: S1. Collect multi-source raw data related to the Chinese medical school and perform standardized preprocessing to obtain a standardized Chinese medical school text dataset. S2. A unidirectional, progressive, three-level vertical deep feature extraction architecture is adopted to perform hierarchical feature extraction on the standardized text dataset, specifically as follows: S21. First-level feature extraction: Extract basic semantic and TCM entity features from the standardized text dataset and output the first-level features; S22. Second-level feature extraction: Based on the first-level features, extract the associated features of the core knowledge system of traditional Chinese medicine and output the second-level features; S23. Third-level feature extraction: Based on the second-level features, extract the inheritance thinking path and teaching adaptation features of traditional Chinese medicine schools, and output the third-level features; S3. Based on the third-level features, construct and train a multi-task fusion teaching model for the inheritance of traditional Chinese medicine schools; S4. Based on the completed training model for the inheritance of traditional Chinese medicine schools, provide learners with teaching services for the inheritance of traditional Chinese medicine schools. S5. Collect teaching data and evaluate teaching effectiveness, and iteratively optimize the feature extraction architecture and teaching model based on the evaluation results.

8. The deep learning-based teaching method for the inheritance of traditional Chinese medicine schools according to claim 7, characterized in that: In step S21, a BERT model pre-trained on a large-scale corpus of traditional Chinese medicine is used to fine-tune the corpus specific to the target Chinese medicine school, extract the basic semantic features and core entity features of the text, and output the first-level features.

9. The deep learning-based teaching method for the inheritance of traditional Chinese medicine schools according to claim 7, characterized in that: In step S22, firstly, based on the entity and semantic features in the first-level features, a knowledge graph exclusive to the target Chinese medical school is constructed. Then, a graph attention network is used to train the knowledge graph, extract systematic knowledge association features, and output the second-level features.

10. The deep learning-based teaching method for the inheritance of traditional Chinese medicine schools according to claim 7, characterized in that: In step S23, a Transformer decoder with a masking mechanism is used in conjunction with a temporal convolutional network to extract features of dialectical thinking path, learning progression features and teaching adaptation features, and output the third-level features.

Images