Traditional Chinese medicine tuina acupoint compatibility and treatment decision support system based on knowledge graph

Abstract

This application relates to the field of medical data processing technology, specifically to a knowledge graph-based TCM massage acupoint combination and treatment decision support system. The system constructs a TCM massage knowledge graph integrating entities of symptoms, diseases, acupoints, techniques, and therapeutic effects through a knowledge graph construction module, assigning weights based on quantified features to the edges of relationships between entities. A reasoning and recommendation engine module calculates the probability distribution of related diseases in the graph based on the input patient symptoms. Based on the probability distribution and the associations between acupoints in the graph, candidate acupoint combination schemes are generated and a comprehensive recommendation score is calculated. Finally, high-scoring recommended schemes and associated techniques are output. A user interaction module provides a symptom entity input interface and displays the recommended acupoint combination schemes and massage technique entities. This transforms implicit massage combination experience into a computable and reasonable graph model, achieving intelligent, precise, and interpretable recommendation of acupoint combination schemes.

Description

Technical Field

[0001] This application belongs to the field of medical data processing technology, specifically relating to a knowledge graph-based TCM massage acupoint matching and treatment decision support system. Background Technology

[0002] The acupoint combination schemes in Traditional Chinese Medicine (TCM) massage rely heavily on the physician's personal experience and tacit knowledge. Their inherent patterns are scattered across various literature and practice records, making them difficult to express and quantify systematically and structurally. Existing auxiliary methods, such as simple statistics based on databases or fixed rule bases, can only achieve superficial correlation analysis between acupoints and symptoms. They cannot characterize the complex, multidimensional network relationships between "symptoms-disease-acupoint-manipulation," and even less can they quantify and evaluate the synergistic effects (combination gain) between different acupoint combinations. This results in a lack of interpretable, reasonable, and precise intelligent combination recommendations for clinical practice. Therefore, there is an urgent need for a decision support solution that can transform TCM massage experience and knowledge into a computable model and support complex reasoning. Summary of the Invention

[0003] To address the aforementioned issues, this application provides a knowledge graph-based TCM massage acupoint combination and treatment decision support system, the system comprising: The knowledge graph construction module is used to extract symptom entities, disease entities, acupoint entities, massage technique entities, and efficacy entities from multi-source TCM massage data as knowledge graph nodes, and extract the association relationships between each entity to construct the edges of each node of the knowledge graph. The edges of each node are configured with weights determined based on the quantitative relationship features of each node. The reasoning and recommendation engine module is used to receive the symptom entity features of the current patient, calculate the probability distribution of disease entities associated with the symptom entity features based on the knowledge graph, generate candidate acupoint matching schemes according to the probability distribution and the knowledge graph, give a comprehensive score to the candidate acupoint matching schemes, and output recommended acupoint matching schemes and associated massage techniques entities according to the comprehensive score. The user interaction module is used to provide a symptom entity input interface and display the recommended acupoint combination scheme and the associated massage technique entity.

[0004] In one implementation, the knowledge graph construction module includes: The data preprocessing submodule is used to clean and standardize the terminology of TCM massage text data to obtain preprocessed text data. The entity relationship extraction submodule is used to identify and extract the symptom entity, the disease entity, the acupoint entity, the massage technique entity, the therapeutic effect entity, and the relationship between each entity from the preprocessed text data. The feature calculation and graph generation submodule is used to convert each entity into a numerical vector node, calculate the weight of the edge of each node based on the initial connection probability between each entity and the centrality of the acupoint entity, and generate a weighted knowledge graph.

[0005] In one implementation, the feature calculation and map generation submodule includes: The vectorization submodule is used to convert the acupoint entity into a numerical vector using the TF-IDF algorithm or word embedding model, wherein the symptom description, meridian attributes and main diseases associated with the acupoint entity constitute the feature set of the acupoint entity. The edge weight calculation submodule is used to calculate the first type of edge weight connecting entities of different categories and the second type of edge weight connecting acupoint entities of the same category; the first type of edge weight is determined based on the centrality of the associated acupoint entities and the initial connection probability between each entity; the second type of edge weight is determined based on the point mutual information value between two acupoint entities.

[0006] In one implementation, the edge weight calculation submodule includes: The centrality calculation submodule is used to calculate the centrality of acupoint entities in disease entity clusters. The calculation is based on the mean of the probability of association between the acupoint entity and each disease entity in its disease entity cluster and the variance of each association probability in the cluster.

[0007] In one implementation, the edge weight calculation submodule includes: The point mutual information value calculation submodule is used to calculate the point mutual information value between two acupoint entities. The calculation is based on the co-occurrence frequency of the two acupoint entities and the occurrence frequency of each acupoint entity.

[0008] In one implementation, the reasoning and recommendation engine module includes: The disease probability inference submodule is used to calculate the probability distribution of each related disease entity by traversing the edges between symptom entities and disease entities in the knowledge graph based on the input symptom entities. When the input is multiple symptom entities, the probability distribution is obtained by summing and normalizing the association probabilities between each symptom entity and the disease entity. The matching generation and scoring submodule is used to select candidate acupoint entities according to the probability distribution, generate candidate acupoint matching schemes based on the edge weight relationship between acupoint entities in the knowledge graph, and call the evaluation model to calculate a comprehensive score for each candidate scheme. The recommended output submodule is used to sort the comprehensive scores and output recommended acupoint combination schemes and associated massage techniques based on the sorting results.

[0009] In one embodiment, the evaluation model is a neural network model or a rule fusion model, and its input features include at least: the association probability between the acupoint entity and the inferred disease entity, the centrality feature of the acupoint entity in the knowledge graph, and the historical therapeutic data associated with the acupoint entity.

[0010] In one implementation, the user interaction module includes a visual interface for graphically displaying entities, relationships, and reasoning paths related to the recommendation scheme in the knowledge graph.

[0011] In one embodiment, the system further includes: The knowledge graph storage module is used to persistently store the knowledge graph and provide graph query services. The feedback learning module is used to receive clinical efficacy feedback on the recommended treatment and update the edge weights or entity attributes of each node in the knowledge graph based on the feedback data.

[0012] In one implementation, the system is deployed as a cloud server architecture, the user interaction module is a web application or mobile application running on the client, and the user interaction module communicates with the reasoning and recommendation engine module and the knowledge graph storage module in the cloud server through a network interface.

[0013] The embodiments of this application have the following beneficial effects: This application provides a knowledge graph-based TCM massage acupoint combination and treatment decision support system. It constructs a TCM massage knowledge graph integrating entities of symptoms, diseases, acupoints, techniques, and therapeutic effects through a knowledge graph construction module, and assigns weights based on quantified features to the relationship edges between entities. A reasoning and recommendation engine module calculates the probability distribution of related diseases in the graph based on the input patient symptoms. Based on the probability distribution and the associations between acupoints in the graph, candidate acupoint combination schemes are generated and a comprehensive recommendation score is calculated. Finally, high-scoring recommended schemes and associated techniques are output. A user interaction module provides a symptom entity input interface and displays the recommended acupoint combination schemes and massage technique entities. This application transforms implicit massage combination experience into a computable and reasonable graph model, realizing intelligent, precise, and interpretable recommendation of acupoint combination schemes.

[0014] Other features and advantages of this application will be described in detail in the following detailed description section. Attached Figure Description

[0015] To more clearly illustrate the implementation schemes of this application, the accompanying drawings used in the implementation schemes will be briefly introduced below. It should be understood that the accompanying drawings only show some implementation schemes of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from the accompanying drawings without creative effort.

[0016] Figure 1 This is a block diagram illustrating a knowledge graph-based TCM massage acupoint matching and treatment decision support system according to an exemplary embodiment.

[0017] Figure 2 This is a block diagram illustrating a knowledge graph construction module according to an exemplary embodiment.

[0018] Figure 3 This is a block diagram illustrating a feature calculation and map generation submodule according to an exemplary embodiment.

[0019] Figure 4 This is a block diagram illustrating an edge weight calculation submodule according to an exemplary embodiment.

[0020] Figure 5 This is a block diagram illustrating yet another edge weight calculation submodule according to an exemplary embodiment.

[0021] Figure 6 This is a block diagram illustrating a reasoning and recommendation engine module according to an exemplary embodiment.

[0022] Figure 7 This is a block diagram illustrating a user interaction module according to an exemplary embodiment.

[0023] Figure 8 This is a block diagram illustrating yet another knowledge graph-based TCM massage acupoint matching and treatment decision support system, according to an exemplary embodiment. Detailed Implementation

[0024] The specific embodiments of this application will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit this application.

[0025] It should be understood that the term "comprising" and its variations as used herein are open-ended, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the following description.

[0026] It should be noted that the concepts of "first" and "second" mentioned in this application are only used to distinguish different devices, modules, or units, and are not used to limit the order of functions performed by these devices, modules, or units or their interdependencies. The modifications of "one" and "multiple" mentioned in this application are illustrative rather than restrictive, and those skilled in the art should understand that, unless explicitly stated in the context, they should be understood as "one or more". In the description of this application, unless otherwise stated, "multiple" means two or more, and other quantifiers are similar; "at least one item", "one item or multiple items", or similar expressions refer to any combination of these items, including any combination of single or multiple items.

[0027] Although operations or steps are described in a specific order in the accompanying drawings in the embodiments of this application, this should not be construed as requiring these operations or steps to be performed in the specific order or serial order shown, or requiring all of the shown operations or steps to be performed to obtain the desired result. In the embodiments of this application, these operations or steps may be performed serially; they may be performed in parallel; or a portion of these operations or steps may be performed.

[0028] The names of the messages or information exchanged between multiple devices in the embodiments of this application are for illustrative purposes only and are not intended to limit the scope of these messages or information. It is understood that before using the technical solutions of the various embodiments of this application, the type, scope of use, and usage scenarios of the personal information involved in this application should be disclosed to the user and their authorization obtained in accordance with relevant laws and regulations through appropriate means.

[0029] First, the application scenarios of this application will be explained. Traditional Chinese massage, as an important component of the TCM treatment system, regulates the flow of Qi and blood in the meridians and improves the function of internal organs through the selection of specific acupoints and combinations of techniques. It is widely used in various bone and joint diseases, internal medicine diseases, soft tissue injuries, and sub-health conditioning. In long-term clinical practice, different symptoms and diseases often correspond to combinations of multiple acupoints and massage techniques. These combinations mainly rely on the experience of traditional physicians or classical literature, exhibiting characteristics such as tacit knowledge, difficulty in quantification, and fragmented structure.

[0030] With the accelerated modernization and intelligentization of Traditional Chinese Medicine (TCM), researchers are increasingly exploring the use of information technology to structure and organize TCM knowledge. In recent years, knowledge graph technology has been widely applied in medicine, industry, and other fields. Its "entity-relationship-attribute" structure can clearly express complex knowledge relationships, while also allowing for reasoning and analysis by combining quantitative and semantic features. Therefore, using knowledge graphs to express multi-dimensional relationships such as "symptoms-acupoints-massage techniques-diseases" provides a new and feasible path for the systematization, structuring, and intelligentization of TCM massage.

[0031] However, current research, while employing databases, statistical analysis, or rule bases to summarize the frequency of acupoint usage and symptom associations, largely remains at the stage of simple co-occurrence statistics or knowledge extraction based on single rules, failing to support more complex massage compatibility logic. Traditional massage acupoint compatibility relies on accumulated experience and lacks clear mathematical expression; for example, whether different acupoint combinations have a synergistic enhancement effect (compatibility gain) cannot be directly derived, and reliable quantitative indicators are lacking. Massage therapy often involves multiple symptoms, multiple acupoint combinations, and multiple techniques, essentially constituting a multidimensional, multi-level relational network. Existing methods struggle to simultaneously encompass information at multiple levels, such as "association strength," "similarity," and "efficacy mechanism." Existing decision support methods are mostly based on rules or simple retrieval, unable to combine structural relationships between knowledge for comprehensive reasoning. For example: which acupoints are most likely associated with symptoms? Which acupoint combinations have a greater synergistic effect? ​​How is the optimal massage path for a certain disease derived? There is a lack of interpretable, quantifiable, and reasonable intelligent decision-making systems for massage. Therefore, there is an urgent need for a decision support solution that can transform the experience and knowledge of traditional Chinese massage into a computable model and support complex reasoning. The following describes this application with specific embodiments.

[0032] Figure 1 This is a block diagram illustrating a knowledge graph-based TCM massage acupoint combination and treatment decision support system according to an exemplary embodiment. Figure 1 As shown in the figure, this application provides a knowledge graph-based TCM massage acupoint combination and treatment decision support system, which may include the following modules: The knowledge graph construction module is used to extract symptom entities, disease entities, acupoint entities, massage technique entities, and efficacy entities from multi-source TCM massage data as knowledge graph nodes, and extract the relationships between each entity to construct the edges of each node in the knowledge graph. The edges of each node are configured with weights determined based on the quantitative relationship features of each node.

[0033] The reasoning and recommendation engine module is used to receive the symptom entity features of the current patient, calculate the probability distribution of disease entities associated with the symptom entity features based on the knowledge graph, generate candidate acupoint combination schemes according to the probability distribution and the knowledge graph, give a comprehensive score to the candidate acupoint combination schemes, and output recommended acupoint combination schemes and associated massage techniques entities according to the comprehensive score.

[0034] The user interaction module is used to provide a symptom entity input interface and display the recommended acupoint combination scheme and the associated massage technique entity.

[0035] For example, this application provides a knowledge graph-based TCM massage acupoint matching and treatment decision support system. This system mainly includes a knowledge graph construction module, a reasoning and recommendation engine module, and a user interaction module. The knowledge graph construction module is responsible for extracting structured information from multi-source TCM massage data, abstracting entities such as symptoms, diseases, acupoints, massage techniques, and therapeutic effects into nodes in the graph. By analyzing the co-occurrence frequency, semantic similarity, and association strength between entities, it assigns quantitative weights to the edges between nodes, thereby constructing a weighted, computable TCM massage knowledge graph. The reasoning and recommendation engine module is the intelligent core of the entire system. It receives patient symptoms input by the user, performs graph traversal and probability calculations using the constructed knowledge graph, infers the most likely disease distribution, and generates multiple candidate acupoint matching schemes based on this. Each candidate scheme is scored and ranked using a comprehensive evaluation model (such as a neural network or rule fusion model) that combines structural features (e.g., acupoint centrality, matching association strength), semantic features, and historical therapeutic data from the graph. Ultimately, this module outputs the highest-scoring acupoint combination and its associated recommended massage techniques. The user interaction module provides an intuitive input and display interface where doctors or users can input symptoms, and the system will clearly present the recommended treatment plan, key evidence, and related reasoning paths in the knowledge graph in graphical or list format.

[0036] This application transforms the tacit knowledge in traditional Chinese massage, which relies on personal experience, into a structured, quantifiable, and reasonable explicit knowledge network. Through the construction and application of this knowledge graph, the system can simulate the diagnostic and acupoint selection process of experienced physicians, achieving automated decision support from symptom to disease probability inference, and then to precise acupoint combination recommendations. This not only significantly lowers the learning and application threshold for young doctors or primary care physicians and improves the scientific rigor and standardization of treatment plans, but also enhances the interpretability and credibility of the decision-making process through visualized reasoning paths, providing an effective technical tool for the standardization and intelligent development of traditional Chinese massage.

[0037] Figure 2This is a block diagram illustrating a knowledge graph construction module according to an exemplary embodiment. For example... Figure 2 As shown, the knowledge graph construction module may include the following modules: The data preprocessing submodule is used to clean and standardize the terminology of TCM massage text data to obtain preprocessed text data.

[0038] The entity relationship extraction submodule is used to identify and extract the symptom entity, the disease entity, the acupoint entity, the massage technique entity, the therapeutic effect entity, and the relationship between each entity from the preprocessed text data.

[0039] The feature calculation and graph generation submodule is used to convert each entity into a numerical vector node, calculate the weight of the edge of each node based on the initial connection probability between each entity and the centrality of the acupoint entity, and generate a weighted knowledge graph.

[0040] For example, this application embodiment further refines the knowledge graph construction process in the system, specifically including three sub-modules: data preprocessing, entity relation extraction, and feature calculation and graph generation. The system first cleans and standardizes the collected TCM massage text data (such as textbooks, literature, and case records), removing irrelevant information and normalizing different terms expressing the same concept into standard entities to ensure data quality and semantic consistency. Subsequently, the entity relation extraction sub-module uses natural language processing technology to automatically identify and extract key entities such as symptoms, diseases, acupoints, techniques, and therapeutic effects from the preprocessed text. It also extracts various relationships between these entities, such as "symptom-acupoint selection," "acupoint-compatibility," and "technique-effect," providing raw materials of "nodes" and "edges" for graph construction. Finally, the feature calculation and graph generation sub-module undertakes the tasks of structuring and quantifying knowledge. It transforms the extracted entities into numerical vector representations and assigns weight values ​​to the connection edges between different nodes in the knowledge graph based on the initial connection probability between entities (such as the prior probability that a symptom leads to a disease) and quantitative features such as the acupoint centrality obtained through complex calculations, thereby generating a weighted knowledge graph that can be used for probabilistic reasoning and intelligent recommendation.

[0041] For example, relevant materials on Traditional Chinese Medicine (TCM) massage are obtained from databases, such as patient case studies, textbooks, and journal articles. Feature word extraction is performed on the acquired data to obtain node data. This includes TCM massage textbook texts (acupoint location, techniques, indications), journal articles and guidelines (treatment plans, clinical evidence), and case studies and clinical records (actual treatment combinations, efficacy evaluation). Electronic literature databases (such as CNKI, PubMed TCM sections), hospital HIS / EMR structured fields (after anonymization / authorization), and OCR are used to perform structured recognition on scanned textbook texts. HTML tags, annotation text, meaningless table of contents information, OCR typos, and variant characters are removed. The goal is to unify synonyms into unique identifiers (IDs), for example: "lower back pain" ←→ "lower back pain", "massage method" ←→ "massage".

[0042] The above text data is abstracted from text entities into knowledge graph nodes. Node abstraction follows a unified rule: each type of entity corresponds to one node type, and all nodes must have a unique ID and standardized attributes. The node abstraction process is the same process by which the NLP language model converts the recognized text data into feature word vectors, as detailed in Table 1 below.

[0043] Table 1 Each type of node has fixed attributes: Acupoint node attributes: Name, Meridian to which it belongs, Location description, Efficacy description (retained from the original text); Symptom node attributes: symptom category, primary / secondary symptom tags; Disease node attributes: ICD code, TCM disease mechanism classification (e.g., "wind-cold-dampness"); Technique node attributes: operation method (pointing, pressing, kneading, etc.), force characteristics; Treatment endpoint attributes: Indicator type (VAS score, ROM, severity).

[0044] Raw information such as symptoms, syndromes, acupoints, massage techniques, and compatibility rules are obtained from textbooks, guidelines, literature, and case studies. This data is then cleaned, desensitized, and standardized in terminology to ensure comprehensive and reliable input data. Standardization establishes a unified semantic foundation, ensuring consistency and traceability in subsequent extraction and modeling. Key entities (acupoints, symptoms, syndromes, therapeutic effects, techniques) and their corresponding relationships (e.g., "symptom → acupoint selection," "acupoint → compatibility," "technique → effect") are extracted from the text. Natural language is converted into structured information, laying the foundation for the atlas; relationship extraction constitutes the "skeleton" of reasoning paths and compatibility mechanisms.

[0045] This refined approach ensures a complete transformation chain from raw, unstructured TCM texts to a structured, computable knowledge system. Data preprocessing and terminology standardization lay the foundation for high-quality data, entity relation extraction realizes the transformation of knowledge from natural language to a machine-understandable form, and feature computation and weighted graph generation endow traditional empirical rules with mathematical expression and computational capabilities. The entire construction process is systematic and automated, capable of continuously integrating new clinical data and literature knowledge, enabling the constructed knowledge graph to have dynamic evolution and self-improvement capabilities, providing a solid, reliable, and scalable knowledge foundation for higher-level intelligent decision-making.

[0046] Figure 3 This is a block diagram illustrating a feature calculation and map generation submodule according to an exemplary embodiment. For example... Figure 3 As shown, the feature calculation and map generation submodule may include the following modules: The vectorization submodule is used to convert the acupoint entity into a numerical vector using the TF-IDF algorithm or word embedding model, wherein the symptom description, meridian attributes and indications associated with the acupoint entity constitute the feature set of the acupoint entity.

[0047] The edge weight calculation submodule is used to calculate the first type of edge weight connecting entities of different categories and the second type of edge weight connecting acupoint entities of the same category; the first type of edge weight is determined based on the centrality of the associated acupoint entities and the initial connection probability between each entity; the second type of edge weight is determined based on the point mutual information value between two acupoint entities.

[0048] For example, this scheme focuses on the core aspects of feature computation in the knowledge graph generation process, particularly the vectorized representation of nodes and the precise calculation of edge weights. Regarding vectorized representation, the system employs the TF-IDF algorithm or a word embedding model (such as Word2Vec) to transform each acupoint entity into a numerical vector. The transformation is based on the rich contextual information associated with the acupoint, such as which symptoms it is often described in literature as relieving, which meridian it belongs to, and which diseases it treats. These textual descriptions collectively constitute the feature set of the acupoint, which is transformed by the algorithm into a vector that represents its semantic and functional characteristics. In terms of edge weight calculation, the system distinguishes between two types of edges and uses different calculation methods. For edges connecting entities of different categories (such as symptom nodes and acupoint nodes), their weight is determined by the centrality of the associated acupoint entity and the initial connection probability between the two entities (e.g., the product of centrality and the initial connection probability between the two entities), reflecting the strength and importance of the association between cross-category entities. For edges connecting acupoint entities of the same category (i.e., edges representing acupoint compatibility relationships), their weights are determined by calculating the point mutual information value between the two acupoint entities. The point mutual information value can effectively distinguish whether two acupoints frequently and synergistically co-occur or merely co-occur occasionally, thereby quantifying the strength of the compatibility association between acupoints.

[0049] For example, the TF-IDF algorithm or Word2Vec model is used to convert nodes (symptoms, acupoints) into numerical vectors. Each acupoint is treated as a document, and its associated symptoms, meridians, and indications are used as feature words to construct a TF-IDF vector, which is then used as input for K-means clustering.

[0050] Based on the acquired node data, a massage knowledge graph is constructed by analyzing the co-occurrence frequency and similarity of nodes. The distribution of the current patient's disease information within the knowledge graph is then used to recommend acupoints. Building upon multi-source TCM massage-related knowledge, standardized data collection and text processing are used to structurally extract symptoms, syndromes, acupoints, massage techniques, and their combination rules from literature, textbooks, guidelines, and clinical cases from natural language. A computable knowledge graph is then constructed under a unified terminology system, presenting the complete knowledge chain of "symptoms—syndromes—acupoints—combinations—therapeutic effects" in massage treatment in the form of nodes and relationships. Relying on this structured carrier of the knowledge graph, the system further quantifies the structural features (occurrence frequency, co-occurrence relationships, node similarity, graph centrality, structural neighborhood, etc.) and textual semantic features (such as acupoint descriptions, therapeutic effect semantic vectors, and clinical context embedding), transforming traditional empirical acupoint combination rules into a feature system with computable attributes.

[0051] The constructed node data is analyzed for similarity between nodes, and semantic similarity is obtained through node feature vectors. Traditional Chinese medicine terminology has semantic implicitness (e.g., "dispersing wind and cold" and "benefiting qi and promoting blood circulation"), which cannot be captured by frequency alone. The obtained [number] [item] is calculated... The data of the first node and the first node Cosine similarity between nodes (Cos( , (Existing technologies will not be elaborated upon.) Compare nodes of the same type. Perform K-means clustering analysis based on the similarity between the nodes obtained above, grouping similar node data into the same cluster. Nodes mainly include symptom, disease, and acupoint information. Statistically analyze the obtained historical patient information, constructing the frequency of acupoints under disease information based on symptom and disease information. Analyze the disease information corresponding to symptom information, the acupoint information corresponding to disease information, the treatment methods corresponding to acupoint information, and finally the therapeutic effect of the treatment (S→D→A→M→E).

[0052] For example, the first The symptoms may be the first The probability of being caused by this disease It can be obtained from the following formula: Formula 1 in, The number of possible diseases D corresponding to the symptom S obtained. This represents the number of patients with the current symptom S. When this value is 0, the calculation uses a preset correction parameter as the denominator. The preset correction parameter is set to 1 and can be adjusted according to the specific implementation environment.

[0053] Two entities (symptom entities) and disease entity The initial connection probability P( , It can be obtained from the following formula: P( , )= Formula 2 in, For the first The symptoms may be the first The probability of being caused by a certain disease.

[0054] Formula 3 in, This indicates the number of symptom j data points within the same cluster that are mapped to the number of disease i data points within the same cluster. This represents the total number of symptom information entries obtained; this value is not zero. This indicates the number of disease types in the disease information cluster.

[0055] This application's embodiments transform acupoint entities into vectors rich in semantics, enabling the system to deeply understand the functional characteristics of each acupoint at the textual level and supporting clustering and analysis based on semantic similarity. Furthermore, refined and differentiated calculation of edge weights allows the constructed knowledge graph to not only reflect whether there are relationships between entities but also accurately characterize the strength and nature of those relationships. In particular, the use of point mutual information to quantify acupoint compatibility relationships directly addresses the core but difficult-to-quantify empirical knowledge of "synergistic compatibility" in traditional Chinese massage, transforming it into a calculable indicator. This provides crucial data support for subsequent intelligent acupoint combination recommendations, significantly improving the rationality and clinical value of the recommendation results.

[0056] Figure 4 This is a block diagram illustrating an edge weight calculation submodule according to an exemplary embodiment. For example... Figure 4 As shown, the edge weight calculation submodule may include the following modules: The centrality calculation submodule is used to calculate the centrality of acupoint entities in disease entity clusters. The calculation is based on the mean of the probability of association between the acupoint entity and each disease entity in its disease entity cluster and the variance of each association probability in the cluster.

[0057] For example, this scheme elaborates on the specific calculation method of the key feature of acupoint centrality. Acupoint centrality is used to measure the prevalence and importance of a certain acupoint in treatment plans for a group of similar diseases (disease clusters). The calculation process first analyzes the probability of the target acupoint being selected when associated with each specific disease within the disease cluster, based on knowledge graphs or historical data. Then, the average value of these probabilities is calculated across the entire disease cluster. The higher the average value, the more commonly used the acupoint is in the treatment of this type of disease. Simultaneously, the variance of these probabilities within the cluster is calculated. The smaller the variance, the more stable and consistent the selection probability of the acupoint for different diseases within the cluster, and the more obvious its characteristic as a core treatment acupoint for this type of disease. Ultimately, the centrality of an acupoint is determined by both the average probability and the variance-adjusted value; typically, it is designed so that the higher the average probability and the smaller the variance, the larger the centrality value.

[0058] For example, the centrality of acupoint entities within disease entity clusters. It can be obtained from the following formula: Formula 4 in, Acupoints Within the disease entity cluster and disease entity The mean of the association probability, For the probability of association Variance within the cluster, Let e ​​be an exponential function with the natural number e as its base, and let e be the correlation probability. The method for obtaining it can be referred to Formula 1.

[0059] This calculation method has profound clinical significance. It simulates the "focusing on the primary symptoms and using the primary acupoints" approach in Traditional Chinese Medicine (TCM) acupoint selection. If an acupoint appears frequently and consistently in the treatment of a large category of similar diseases (such as various types of lower back pain), it is likely to be a "key acupoint" or "primary acupoint" for treating this type of disease. By quantifying this "universality" and "stability" mathematically, the system can automatically identify and assign higher centrality weights to these key acupoints. In subsequent recommendation processes, acupoints with high centrality will receive higher priority, which helps ensure that the recommended treatment closely adheres to the core pathogenesis of the disease, conforms to the basic principles of TCM syndrome differentiation and treatment, and improves the clinical relevance and expected stability of the therapeutic effects of the recommendations.

[0060] Figure 5 This is a block diagram illustrating yet another edge weight calculation submodule according to an exemplary embodiment. For example... Figure 5 As shown, the edge weight calculation submodule may include the following modules: The point mutual information value calculation submodule is used to calculate the point mutual information value between two acupoint entities. The calculation is based on the co-occurrence frequency of the two acupoint entities and the occurrence frequency of each acupoint entity.

[0061] For example, this scheme specifically defines a method for quantifying the strength of the compatibility association between two acupoint entities—calculating point mutual information values. The calculation process is based on statistical information from the entire knowledge base or historical dataset. First, it is necessary to statistically analyze the frequency of co-occurrence of two specific acupoints (e.g., acupoint A and acupoint B) in the same treatment plan or case record, and calculate their co-occurrence probability. Second, it is necessary to statistically analyze the probability of acupoint A appearing alone, and the probability of acupoint B appearing alone. The point mutual information value is the comparison between the probability of co-occurrence and the product of their individual probabilities. If the two acupoints are frequently used together, their co-occurrence probability will be much greater than the product of their individual probabilities, and the calculated point mutual information value will be positive and large, indicating a significant synergistic association or compatibility habit between the two. Conversely, if the co-occurrence probability is close to or equal to the product of their individual probabilities, the point mutual information value is close to zero or negative, indicating a weak association or that their co-occurrence may be accidental.

[0062] For example, two acupoint entities Point mutual information value It can be obtained from the following formula: Formula 5 in, Two acupoint entities co-occurrence frequency, These are two acupoint entities. Their respective frequencies of occurrence It should be a positive minimum value to prevent the denominator from being zero; generally, it can be taken as 0.01.

[0063] This application employs point mutual information to calculate the edge weights between acupoints. Its advantage lies in effectively filtering out spurious associations that merely occur accidentally in a large number of schemes due to their shared commonness, truly capturing "drug pairs" or "acupoint pairs" relationships with clinical synergistic significance. This solves the bias problem that may arise from relying solely on co-occurrence frequency statistics, making the connections between acupoints in the knowledge graph more reflective of the essence of TCM compatibility. Based on this quantitative indicator, when generating candidate acupoint combinations, the system can prioritize combining acupoint pairs with high point mutual information values, thereby recommending more clinically common and theoretically more synergistically effective combination schemes, improving the rationality and professionalism of the recommendation results.

[0064] Figure 6 This is a block diagram illustrating a reasoning and recommendation engine module according to an exemplary embodiment. Figure 6 As shown, the reasoning and recommendation engine module may include the following modules: The disease probability inference submodule is used to calculate the probability distribution of each related disease entity by traversing the edges between symptom entities and disease entities in the knowledge graph based on the input symptom entities. When the input consists of multiple symptom entities, the probability distribution is obtained by summing and normalizing the association probabilities between each symptom entity and the disease entity.

[0065] The matching generation and scoring submodule is used to select candidate acupoint entities according to the probability distribution, generate candidate acupoint matching schemes based on the edge weight relationship between acupoint entities in the knowledge graph, and call the evaluation model to calculate a comprehensive score for each candidate scheme.

[0066] The recommended output submodule is used to sort the comprehensive scores and output recommended acupoint combination schemes and associated massage techniques based on the sorting results.

[0067] For example, this scheme details the core workflow of the reasoning and recommendation engine module, namely, how to generate final treatment recommendations based on input symptoms. The process begins with disease probability inference: when a user inputs one or more symptoms, the system locates these symptom nodes in the knowledge graph and traverses and calculates along the "symptom-disease" edges. For a single symptom, the system calculates the conditional probability based on the historical association frequency between the symptom and each disease; for multiple symptoms, the system sums the disease probabilities associated with each symptom, then performs normalization, ultimately obtaining a series of possible diseases and their corresponding probability distributions. This simulates the process of a doctor making differential diagnoses based on symptom clusters. The next stage is the combination generation and scoring phase: based on the above disease probability distribution, the system prioritizes acupoints strongly correlated with high-probability diseases as candidates, and combines these candidate acupoints into different combination schemes (such as single acupoints, double acupoints, or small combinations) based on the edge weights between acupoint nodes in the knowledge graph (especially edges reflecting combination relationships). For each candidate combination scheme, the system calls an evaluation model. The model's input features integrate multiple information sources, including the strength of the association between acupoints and the inferred disease, the centrality of each acupoint in the atlas, and efficacy data recorded in historical treatments. The evaluation model (which can be a neural network or a rule-based model) processes these features to calculate a comprehensive score for each solution. Finally, the system ranks all candidate solutions by their comprehensive scores and outputs the one or more solutions with the highest scores, along with their associated massage techniques (obtained based on the "acupoint-technique" relationship), to the user.

[0068] For example, disease entities probability distribution It can be obtained from the following formula: Formula 6 in, XD represents the number of symptoms obtained, and XD represents the number of disease types obtained.

[0069] This application's embodiments realize a complete, automated reasoning chain from symptom input to treatment plan output. Its core value lies in transforming vague symptom descriptions into clear disease probability distributions, and then generating and selecting the most reasonable treatment recommendations based on structured knowledge networks and quantitative feature evaluations. The entire process is not only highly efficient, but also, due to the integration of probabilistic inference, graph structure analysis, and multi-feature comprehensive evaluation, the final recommendation results possess clinical relevance, compatibility, and efficacy predictability, providing physicians with powerful digital decision support.

[0070] In one possible implementation, the evaluation model is a neural network model or a rule fusion model, and its input features include at least: the association probability between the acupoint entity and the inferred disease entity, the centrality feature of the acupoint entity in the knowledge graph, and the historical therapeutic data associated with the acupoint entity.

[0071] For example, this scheme further clarifies the specific composition of the comprehensive scoring model used to evaluate candidate acupoint combination schemes. This evaluation model can employ existing neural network models or existing models that integrate multiple rules. Regardless of the form, its input features must include several key dimensions; the primary feature is the association probability between the acupoint entity and the disease entity inferred by the system, which directly reflects the acupoint's targeted strength for the currently suspected disease. The second important feature is the centrality of the acupoint entity calculated in the knowledge graph, which characterizes the universality and core position of the acupoint in treating a certain type of disease; acupoints with high centrality often have a more fundamental role. The third key feature is the historical efficacy data associated with the acupoint entity, such as quantitative indicators like the improvement in patients' pain scores (VAS) and range of motion (ROM) after using the acupoint in past cases, which provides empirical support for the recommended scheme.

[0072] By integrating these multi-dimensional features into the evaluation model, the system can comprehensively assess candidate treatment combinations. Association probability ensures the alignment of the combination with the current diagnosis; centrality incorporates the experiential wisdom of traditional Chinese medicine's "primary acupoint selection," enhancing the classic nature and stability of the combination; historical efficacy data incorporates evidence-based medicine principles, ensuring that recommendations are grounded in clinical evidence. This comprehensive evaluation mechanism avoids the bias that can arise from a single indicator, ensuring that high-scoring combinations are not only targeted at the current symptoms and conform to the principles of traditional Chinese medicine treatment, but are also supported by clinical efficacy data. This significantly improves the clinical credibility and acceptability of the recommended combinations, making the system's decision support more comprehensive and reliable.

[0073] Figure 7 This is a block diagram illustrating a user interaction module according to an exemplary embodiment. For example... Figure 7 As shown, the user interaction module may include a visual interface for graphically displaying entities, relationships, and reasoning paths related to the recommendation scheme in the knowledge graph.

[0074] For example, this scheme demonstrates the visualization function of the system's user interaction module. The system provides a graphical user interface that not only allows users to easily input symptom descriptions but also dynamically displays knowledge graph fragments closely related to the current recommended solution in an intuitive graphical form. For instance, when the system outputs a recommended acupoint combination scheme, the visualization interface can present a partial graph that clearly shows the symptom nodes input by the user, the disease nodes inferred by the system, the recommended acupoint nodes, and the connections between them. Simultaneously, the main reasoning path upon which the system generated the recommendation (such as "symptom -> disease -> acupoint -> technique") is also highlighted in the graph.

[0075] This visualization design has significant application value. It presents the complex calculation and reasoning processes within the system to doctors or users in the form of a clear knowledge network diagram. This not only makes the recommendation results more intuitive and easier to understand, but more importantly, it provides "explainability" for the decision-making process. Doctors can clearly see which symptoms are associated with which diseases, which acupoints were selected based on the characteristics of the diseases, and how the acupoints are combined. This transparent decision-making process helps doctors understand and verify the system's recommendation logic, building trust in the intelligent system. Simultaneously, it also serves as an efficient teaching tool, helping learners intuitively understand the complex network relationships between symptoms, diseases, and acupoints in Traditional Chinese Medicine massage, promoting the dissemination and inheritance of knowledge.

[0076] Figure 8 This is a block diagram illustrating yet another knowledge graph-based TCM massage acupoint matching and treatment decision support system, according to an exemplary embodiment. For example... Figure 8 As shown, the system may further include the following modules: The knowledge graph storage module is used to persistently store the knowledge graph and provide graph query services.

[0077] The feedback learning module is used to receive clinical efficacy feedback on the recommended treatment and update the edge weights or entity attributes of each node in the knowledge graph based on the feedback data.

[0078] For example, this solution adds continuous learning and evolution capabilities to the system, primarily through a knowledge graph storage module and a feedback learning module. The knowledge graph storage module is responsible for persistently storing the constructed, weighted knowledge graph and providing efficient graph query and access services, ensuring the inference engine can quickly obtain the required data. The feedback learning module is key to the system's intelligent iteration. In clinical practice, doctors adopt the system's recommended treatment plans and record the actual therapeutic effects into the system. The feedback learning module collects this clinical therapeutic feedback data and updates and optimizes the knowledge graph based on it. For instance, if an acupoint combination plan shows excellent efficacy in multiple practices, the system can correspondingly increase the weights of the edges between related acupoints and diseases, and between acupoints; conversely, if a plan is ineffective, its related weights may be decreased. Simultaneously, new therapeutic data can also be used as entity attributes to enrich the information of acupoint or technique nodes.

[0079] By introducing a feedback learning mechanism, the system transforms from a static knowledge base application into a dynamically evolving intelligent system. It continuously feeds the latest clinical practice results back into the knowledge graph, revising and refining the original knowledge model and recommendation logic. This closed loop of "practice-feedback-optimization" allows the system to keep pace with clinical developments, absorb new treatment experiences, and its recommendation performance is expected to continuously improve with increased usage time and the accumulation of feedback data. This aligns with the characteristic of Traditional Chinese Medicine (TCM) knowledge originating from and evolving through practice, thus constructing a truly vibrant and adaptable intelligent auxiliary decision-making system for TCM.

[0080] In one possible implementation, the system is deployed as a cloud server architecture, the user interaction module is a web application or mobile application running on the client, and the user interaction module communicates with the reasoning and recommendation engine module and the knowledge graph storage module in the cloud server through a network interface.

[0081] For example, this scheme describes a typical deployment architecture for the system, employing a combination of cloud servers and client applications. The system's core computing and data resources, including the knowledge graph construction module, the inference and recommendation engine module, and the database storing the complete knowledge graph, are all deployed on cloud servers. This cloud deployment ensures powerful computing capabilities, large-scale knowledge graph storage space, and convenient centralized maintenance and updates. Users access the system through a web browser or a dedicated mobile application (App), which constitutes the system's user interaction module. Users input symptoms through the client, and requests are sent to the cloud server via a network interface. Upon receiving the request, the cloud-based inference engine invokes the knowledge graph to perform calculations, generates recommendation schemes, and then returns the results to the client via the network, where they are presented to the user through the client interface.

[0082] This cloud-client architecture offers several advantages. First, it lowers the barrier to entry for users; doctors and learners can access the powerful system functions simply through a regular computer browser or mobile app, without needing to install complex software or maintain a massive database locally. Second, it facilitates unified data management and synchronized knowledge updates; a single update to the knowledge graph in the cloud allows all client users to immediately enjoy the latest knowledge services and improved recommendation algorithms. Third, this architecture facilitates remote diagnosis and treatment support and internet-based health management; primary care physicians and patients can easily access professional advice from the cloud system, promoting the sharing and dissemination of high-quality traditional Chinese medicine massage resources, aligning with the development trend of "Internet + Healthcare."

[0083] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these modifications and improvements all fall within the protection scope of this application.

Claims

1. A knowledge graph-based TCM massage acupoint combination and treatment decision support system, characterized in that, The system includes: The knowledge graph construction module is used to extract symptom entities, disease entities, acupoint entities, massage technique entities, and efficacy entities from multi-source TCM massage data as knowledge graph nodes, and extract the association relationships between each entity to construct the edges of each node of the knowledge graph. The edges of each node are configured with weights determined based on the quantitative relationship features of each node. The reasoning and recommendation engine module is used to receive the symptom entity features of the current patient, calculate the probability distribution of disease entities associated with the symptom entity features based on the knowledge graph, generate candidate acupoint matching schemes according to the probability distribution and the knowledge graph, give a comprehensive score to the candidate acupoint matching schemes, and output recommended acupoint matching schemes and associated massage techniques entities according to the comprehensive score. The user interaction module is used to provide a symptom entity input interface and display the recommended acupoint combination scheme and the associated massage technique entity.

2. The knowledge graph-based TCM massage acupoint matching and treatment decision support system according to claim 1, characterized in that, The knowledge graph construction module includes: The data preprocessing submodule is used to clean and standardize the terminology of TCM massage text data to obtain preprocessed text data. The entity relationship extraction submodule is used to identify and extract the symptom entity, the disease entity, the acupoint entity, the massage technique entity, the therapeutic effect entity, and the relationship between each entity from the preprocessed text data. The feature calculation and graph generation submodule is used to convert each entity into a numerical vector node, calculate the weight of the edge of each node based on the initial connection probability between each entity and the centrality of the acupoint entity, and generate a weighted knowledge graph.

3. The knowledge graph-based TCM massage acupoint matching and treatment decision support system according to claim 2, characterized in that, The feature calculation and map generation submodule includes: The vectorization submodule is used to convert the acupoint entity into a numerical vector using the TF-IDF algorithm or word embedding model, wherein the symptom description, meridian attributes and main diseases associated with the acupoint entity constitute the feature set of the acupoint entity. The edge weight calculation submodule is used to calculate the first type of edge weight connecting entities of different categories and the second type of edge weight connecting acupoint entities of the same category; the first type of edge weight is determined based on the centrality of the associated acupoint entities and the initial connection probability between each entity; the second type of edge weight is determined based on the point mutual information value between two acupoint entities.

4. The knowledge graph-based TCM massage acupoint matching and treatment decision support system according to claim 3, characterized in that, The edge weight calculation submodule includes: The centrality calculation submodule is used to calculate the centrality of acupoint entities in disease entity clusters. The calculation is based on the mean of the probability of association between the acupoint entity and each disease entity in its disease entity cluster and the variance of each association probability in the cluster.

5. The knowledge graph-based TCM massage acupoint matching and treatment decision support system according to claim 3, characterized in that, The edge weight calculation submodule includes: The point mutual information value calculation submodule is used to calculate the point mutual information value between two acupoint entities. The calculation is based on the co-occurrence frequency of the two acupoint entities and the occurrence frequency of each acupoint entity.

6. The knowledge graph-based TCM massage acupoint matching and treatment decision support system according to claim 1, characterized in that, The reasoning and recommendation engine module includes: The disease probability inference submodule is used to calculate the probability distribution of each related disease entity by traversing the edges between symptom entities and disease entities in the knowledge graph based on the input symptom entities. When the input is multiple symptom entities, the probability distribution is obtained by summing and normalizing the association probabilities between each symptom entity and the disease entity. The matching generation and scoring submodule is used to select candidate acupoint entities according to the probability distribution, generate candidate acupoint matching schemes based on the edge weight relationship between acupoint entities in the knowledge graph, and call the evaluation model to calculate a comprehensive score for each candidate scheme. The recommended output submodule is used to sort the comprehensive scores and output recommended acupoint combination schemes and associated massage techniques based on the sorting results.

7. The knowledge graph-based TCM massage acupoint matching and treatment decision support system according to claim 6, characterized in that, The evaluation model is a neural network model or a rule fusion model, and its input features include at least: the association probability between the acupoint entity and the inferred disease entity, the centrality feature of the acupoint entity in the knowledge graph, and the historical therapeutic data associated with the acupoint entity.

8. The knowledge graph-based TCM massage acupoint matching and treatment decision support system according to claim 1, characterized in that, The user interaction module includes a visual interface for displaying, in graphical form, the entities, relationships, and reasoning paths related to the recommendation scheme in the knowledge graph.

9. The knowledge graph-based TCM massage acupoint matching and treatment decision support system according to claim 1, characterized in that, The system also includes: The knowledge graph storage module is used to persistently store the knowledge graph and provide graph query services. The feedback learning module is used to receive clinical efficacy feedback on the recommended treatment and update the edge weights or entity attributes of each node in the knowledge graph based on the feedback data.

10. The knowledge graph-based TCM massage acupoint matching and treatment decision support system according to claim 9, characterized in that, The system is deployed in a cloud server architecture. The user interaction module is a web application or mobile application running on the client. The user interaction module communicates with the reasoning and recommendation engine module and the knowledge graph storage module in the cloud server through a network interface.

Images