Acupuncture point recommendation method, acupuncture point model acquisition method, acupuncture point recommendation device, acupuncture point model acquisition equipment and medium

Abstract

The invention relates to the field of traditional Chinese medicine clinical auxiliary diagnosis, and discloses an acupuncture point recommendation method, an acupuncture point recommendation model obtaining method, an acupuncture point recommendation model obtaining device, acupuncture point recommendation equipment and a medium, and the acupuncture point recommendation model obtaining method comprises the following steps: collecting patient medical record data to form a medical record data sample set; preprocessing the medical record data sample set to obtain a standard medical record data sample set; inputting the standard medical record data sample set into a preset language model, performing preliminary training on the language model based on the standard medical record data sample set, so that the language model establishes a semantic mapping relationship between the patient symptom information and the acupuncture points, and performing parameter fine tuning and strategy optimization on the preliminarily trained language model in sequence, obtaining an acupuncture point recommendation model; according to the acupuncture point recommendation model obtaining method, the accuracy and individual adaptability of acupuncture clinical treatment are improved, the diagnosis and treatment efficiency is optimized, and a scientific and systematic treatment scheme is provided for a patient.

Description

Technical Field

[0001] The present application relates to the technical field of clinical auxiliary diagnosis of traditional Chinese medicine, and in particular to a method for recommending acupuncture points and a method, device, equipment and medium for obtaining acupuncture points model. Background Art

[0002] With increasing work pressure and improved living standards, sub-health issues have gradually become a focus of public attention, and the use of Traditional Chinese Medicine (TCM) acupoints to regulate bodily functions has gradually gained attention. Acupuncture, as an important component of Traditional Chinese Medicine (TCM), is gaining increasing application and recognition worldwide and has been proven to have certain clinical benefits as an adjunctive treatment. However, acupuncture therapy is subject to a degree of subjectivity in clinical practice, relying not only on the practitioner's experience and skill level but also on individual patient differences. The effectiveness of acupuncture is based on the body's self-regulatory function, which means that its effects can vary significantly between patients.

[0003] At the same time, artificial intelligence has made significant progress in assisting disease diagnosis and treatment in recent years, with large language models demonstrating significant progress in clinical predictive modeling. With sufficient training samples, large language models can perform as well as or even better than domain experts in diagnosing certain diseases. However, in the field of acupuncture, direct training of high-precision models faces the challenge of data scarcity due to the difficulty in obtaining large-scale real-world medical records. The emergence of transfer learning offers a potential solution to this problem, namely pre-training on large unlabeled datasets before fine-tuning for specific tasks. Currently, acupuncture bronze man simulation tools exist that can provide a foundation for training and teaching, but these tools have yet to be combined with large models to form systematic, intelligent, and standardized acupuncture recommendations.

[0004] However, traditional acupuncture therapy has some shortcomings in clinical application: (1) Dependence on experience and individual differences: The selection of acupuncture points often depends on the doctor's personal experience and judgment, and lacks objective and unified acupoint selection standards. At the same time, patients' own physical constitutions, tolerance and self-regulation abilities vary greatly, resulting in different effects of the same treatment plan on different patients. (2) Lack of data and insufficient standardization: Clinical data in the field of acupuncture is difficult to collect, label and organize on a large scale, resulting in insufficient training resources for large language models in acupuncture applications. In addition, acupuncture therapy lacks a complete digital standard, making it more difficult to structure the input and output of the model. (3) Lack of intelligent teaching and auxiliary systems: Although the existing acupuncture bronze man simulation practice tool can help medical students or practitioners practice acupuncture techniques, it has not been integrated with the artificial intelligence system, making it difficult to provide learners with personalized and immediate feedback guidance, and it is also impossible to form a standardized and verifiable intelligent recommendation system.

[0005] Therefore, how to use modern artificial intelligence technology to improve the objectivity and standardization of acupuncture treatment and overcome limitations such as subjective dependence and lack of data has become an important issue that needs to be urgently addressed. Summary of the Invention

[0006] In view of this, the embodiments of the present application provide an acupuncture point recommendation method and its model acquisition method, device, equipment and medium, which can effectively solve the problems in the prior art of acupuncture point selection that rely too much on subjective experience, lack of clinical data and lack of intelligent assistance.

[0007] In a first aspect, an embodiment of the present application provides a method for obtaining an acupuncture point recommendation model, comprising: Collecting patient medical record data, and performing format conversion, encoding, and field classification extraction operations on the collected patient medical record data in sequence to form a medical record data sample set; Preprocessing the medical record data sample set to generate a standard medical record data sample set; Inputting the standard medical record data sample set into a preset language model, and preliminarily training the language model based on the standard medical record data sample set, so that the language model establishes a semantic mapping relationship between patient symptom information and acupuncture points; Parameter fine-tuning and strategy optimization are sequentially performed on the preliminarily trained language model to obtain an acupuncture point recommendation model; the acupuncture point recommendation model is used to output a mapping result between the symptom information and the corresponding acupuncture points.

[0008] In some embodiments, the medical record data sample set includes a plurality of initial medical record data entries; and preprocessing the medical record data sample set to generate a standard medical record data sample set includes: Performing semantic analysis on each of the initial medical record data entries in the medical record data sample set to extract key diagnosis and treatment fields; Based on the preset prompt word template, add corresponding prompt words for the key diagnosis and treatment fields; Setting a corresponding priority according to the importance of the key diagnosis and treatment fields in acupuncture diagnosis, and binding the priority to the prompt word; Integrate the key diagnosis and treatment information fields with the prompt words added and bound to the priority to generate a standard medical record data entry; A plurality of the standard medical record data entries are aggregated to form the standard medical record data sample set.

[0009] In some embodiments, inputting the standard medical record data sample set into a preset language model and performing preliminary training on the language model based on the standard medical record data sample set includes: Performing word segmentation and encoding processing on the standard medical record data sample set to generate a numerical representation acceptable to the language model; Inputting the standard medical record data sample set that has undergone the word segmentation and encoding processing into the language model based on the TransformerDecoder architecture, and extracting semantic features from the standard medical record data sample set using position encoding and attention mechanism; Based on the semantic features, the acupuncture point information predicted by the language model is compared with the pre-labeled acupuncture point information, and the error is calculated using a preset loss evaluation standard; Iteratively updating the parameters of the language model through a back-propagation function; When the error during training drops to a preset convergence standard, the language model that has undergone preliminary training is output.

[0010] In some embodiments, fine-tuning and strategy optimization are sequentially performed on the preliminarily trained language model to obtain an acupuncture point recommendation model, including: Decomposing the weight of the language model after preliminary training into an amplitude component and a direction component; Fix the direction component, and update the amplitude component according to a preset weight adjustment strategy, and output a parameter fine-tuning result; Based on the parameter fine-tuning results, a feedback mechanism based on reinforcement learning is constructed. By introducing a strategy reward signal, the language model is guided to perform strategy optimization to form the acupuncture point recommendation model; wherein the reward signal is determined according to the degree of match between the acupuncture point recommendation information and the actual treatment effect.

[0011] In a second aspect, the present invention provides an acupuncture point recommendation method, comprising: In response to the received symptom information of the patient to be diagnosed, extract key diagnosis and treatment information of the symptom information of the patient to be diagnosed, and generate a standardized symptom description; The standardized symptom description is processed using an acupuncture point recommendation model in combination with TCM theory rules and a medical knowledge base to generate acupuncture point recommendation information corresponding to the symptom information; wherein the acupuncture point recommendation model is obtained using the acupuncture point recommendation model acquisition method described in any one of the first aspects.

[0012] In some embodiments, the method further comprises: The acupuncture point recommendation information is mapped onto an acupuncture bronze mannequin corresponding to a three-dimensional human body structure, so as to display the recommended acupuncture points on the acupuncture bronze mannequin; wherein the acupuncture bronze mannequin is provided with a plurality of marking points corresponding to human acupuncture points; the mapping process includes: Converting the acupuncture point recommendation information into mapping data corresponding to the marked points; The mapping data is input into the acupuncture bronze man model to display the acupuncture point recommendation information on the acupuncture bronze man model.

[0013] In some embodiments, the acupuncture point recommendation model is used to process the standardized symptom description in combination with traditional Chinese medicine theory and a medical knowledge base to generate acupuncture point recommendation information corresponding to the symptom information, including: Extracting key diagnostic and treatment information from the patient's symptom information to generate a standardized symptom description; Inputting the standardized symptom description into the acupuncture point recommendation model for semantic analysis to obtain initial acupuncture point recommendation information; The initial acupuncture point recommendation information is calibrated using the TCM theory rules and the medical knowledge base to generate the acupuncture point recommendation information corresponding to the symptom information for recommendation.

[0014] In a third aspect, an embodiment of the present application provides a device for obtaining an acupuncture point recommendation model, comprising: The data collection module is used to collect patient medical record data, and perform format conversion, encoding, and field classification extraction on the collected patient medical record data in sequence to form a medical record data sample set; A data processing module, configured to pre-process the medical record data sample set to generate a standard medical record data sample set; a model training module, configured to input the standard medical record data sample set into a preset language model, and perform preliminary training on the language model based on the standard medical record data sample set, so that the language model establishes a semantic mapping relationship between patient symptom information and acupuncture points; The acupoint generation module is used to perform parameter fine-tuning and strategy optimization on the language model that has been preliminarily trained in sequence to obtain an acupuncture point recommendation model; the acupuncture point recommendation model is used to output the mapping result between the symptom information and the corresponding acupuncture points.

[0015] In a fifth aspect, an embodiment of the present application provides a computer-readable storage medium, and when the computer program is executed on a processor, the method described in any one of the first and second aspects above is implemented.

[0016] The embodiments of the present application have the following beneficial effects: The acupuncture point recommendation model acquisition method of the present application constructs a unified and standard medical record data sample set by format conversion, encoding and field classification extraction of patient medical record data, and further preprocesses to generate a high-quality standard medical record data sample set to ensure the consistency and structuring of data input. Subsequently, the standard medical record data sample set is input into a preset language model and preliminarily trained to establish a semantic mapping relationship between patient symptom information and acupuncture points; then, by sequentially performing parameter fine-tuning and strategy optimization, a highly accurate and robust acupuncture point recommendation model is finally generated. The acupuncture point recommendation model acquisition method of the present application not only overcomes the problem that traditional acupuncture point selection relies too much on the subjective experience of physicians and the lack of clinical data, but also provides a solid foundation for subsequent model applications, significantly improving the objectivity and standardization of acupuncture treatment plans.

[0017] The acupuncture point recommendation method of this application, by responding to the symptom information of the patient to be diagnosed, uses natural language processing technology to extract key diagnosis and treatment information, generates standardized symptom descriptions, and then uses the trained acupuncture point recommendation model to achieve accurate semantic analysis of the patient's symptom information. Combining the theoretical rules of traditional Chinese medicine and the medical knowledge base to calibrate the preliminary recommendation results, the acupuncture point recommendation method of this application can output personalized acupuncture treatment suggestions that conform to clinical practice, overcome the subjective problems caused by differences in doctor experience and non-standard data in the traditional diagnosis process, and effectively improve the objectivity of diagnosis and the accuracy of treatment plans, providing scientific, standardized and personalized auxiliary decision support for the clinical practice of traditional Chinese medicine. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.

[0019] Figure 1 A flow chart of a method for obtaining an acupuncture point recommendation model according to an embodiment of the present application is shown; Figure 2 A schematic diagram of the Llama3 model in a method for obtaining an acupuncture point recommendation model according to an embodiment of the present application is shown; Figure 3 A schematic diagram showing a method for obtaining an acupuncture point recommendation model according to an embodiment of the present application using a rotational position encoding technique to enable the model to perceive sequence positions is shown; Figure 4 A schematic diagram showing a multi-head attention mechanism and a group query attention mechanism in a method for acquiring an acupuncture point recommendation model according to an embodiment of the present application is shown; Figure 5 A schematic diagram showing a DoRA fine-tuning method in a method for acquiring an acupuncture point recommendation model according to an embodiment of the present application is shown; Figure 6 A schematic diagram of a model optimization process based on reinforcement learning in a method for obtaining an acupuncture point recommendation model according to an embodiment of the present application is shown; Figure 7 A flow chart of a method for recommending acupuncture points according to an embodiment of the present application is shown; Figure 8 A structural schematic diagram of an acupuncture point recommendation model acquisition device according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments.

[0021] The components of the embodiments of the present application generally described and illustrated in the drawings herein may be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed application, but rather merely represents selected embodiments of the present application. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative effort are within the scope of protection of the present application.

[0022] Hereinafter, the terms "including", "having" and their cognates used in various embodiments of the present application are intended only to indicate specific features, numbers, steps, operations, elements, components or combinations of the aforementioned items, and should not be understood as excluding the existence of one or more other features, numbers, steps, operations, elements, components or combinations of the aforementioned items or adding the possibility of one or more features, numbers, steps, operations, elements, components or combinations of the aforementioned items. In addition, the terms "first", "second", "third" and the like are only used to distinguish descriptions and should not be understood as indicating or implying relative importance.

[0023] Unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by those skilled in the art to which the various embodiments of the present application belong. The terms (such as those defined in generally used dictionaries) will be interpreted as having the same meaning as in the context of the relevant technical field and will not be interpreted as having an idealized meaning or an overly formal meaning unless clearly defined in the various embodiments of the present application.

[0024] The following describes some embodiments of the present application in detail with reference to the accompanying drawings. In the absence of conflict, the following embodiments and features in the embodiments may be combined with each other.

[0025] Taking into account the problems of excessive reliance on subjective experience, lack of clinical data and lack of intelligent assistance in the existing technology of acupuncture point selection, a method for obtaining an acupuncture point recommendation model is proposed. By collecting and preprocessing patient medical record data, and inputting a standardized medical record data sample set into a preset language model for training and optimization, an acupuncture point recommendation model is obtained. An acupuncture point recommendation method is proposed. When receiving symptom information of patients to be diagnosed, the trained acupuncture point recommendation model is used in combination with traditional Chinese medicine theory and medical knowledge base to generate personalized acupuncture points, thereby improving the accuracy and efficiency of acupuncture treatment.

[0026] Figure 1 A flow chart of a method for obtaining an acupuncture point recommendation model according to an embodiment of the present application is shown. Exemplarily, the method includes the following steps: Step S100 , collecting patient medical record data, and sequentially performing format conversion, encoding, and field classification extraction operations on the collected patient medical record data to form a medical record data sample set.

[0027] For example, real-world patient medical records are collected from hospitals or clinics. This data primarily includes the patient's personal information, physician's diagnosis, current medical history, chief complaint, past medical history, symptoms, and physical condition. This information is typically stored in electronic medical records and can be extracted from various systems, including diagnostic and treatment records. Next, the collected raw medical record data undergoes format conversion and encoding. The goal of format conversion is to convert the diverse text, tables, or other data formats within the raw medical records into a standardized format so that all data meets the model input requirements. Subsequently, encoding is performed to convert the standardized text information into a numerical representation, generating vector data acceptable to the model. The converted and encoded medical record data is then segmented and extracted according to field categories. This process involves segmenting the medical record information into field categories, for example, extracting information such as "medical history," "symptoms," and "diagnosis" and classifying and encoding them. Each data entry will contain multiple structured fields, such as patient name, age, and chief complaint. This structuring process organizes the fragmented text data into a form that is easy to analyze and manipulate, ensuring consistent and high-quality input data for subsequent data analysis and model training. Finally, the processed initial medical record data entries are categorized and organized to form a complete medical record data sample set. This sample set covers a sufficient number of representative patient records, providing strong data support for subsequent deep learning model training.

[0028] Step S200: pre-process the medical record data sample set to generate a standard medical record data sample set.

[0029] In this example, the primary purpose of preprocessing the medical record data sample set is to convert the collected raw medical record data into a unified and standardized data format to facilitate efficient subsequent model training. This process not only improves data consistency and quality but also ensures that the data format fully meets model input requirements, laying a solid foundation for establishing a precise semantic mapping between patient symptom information and acupuncture points.

[0030] In an optional embodiment, step S200 includes the following sub-steps: Step S201 , performing semantic analysis on each initial medical record data entry in the medical record data sample set, and extracting key diagnosis and treatment fields.

[0031] As an example, semantic parsing is performed on each initial medical record in the sample set to extract key diagnosis and treatment fields. These fields include the patient's chief complaint, current medical history, past medical history, the four diagnostic methods of Traditional Chinese Medicine (TCM), and TCM diagnosis. This information is crucial for the subsequent generation of acupuncture points. Semantic parsing involves using natural language processing techniques to convert the information in each field into a form that the model can understand, facilitating subsequent data analysis and efficient language model training.

[0032] Step S202: Add corresponding prompt words to each key diagnosis and treatment field based on a preset prompt word template.

[0033] To help the model better understand the extracted fields, each key diagnosis and treatment field is assigned appropriate prompt words based on a pre-defined prompt word template. These prompt word templates are designed using human annotation and a medical expert knowledge base to guide the model in more accurately identifying the contextual meaning of the fields. For example, for the "Chief Complaint" field, prompt words might include "symptoms" or "patient's perception." This process ensures the model's semantic understanding of diagnosis and treatment information is more accurate.

[0034] Step S203: setting a corresponding priority according to the importance of the key diagnosis and treatment fields in acupuncture diagnosis, and binding the priority with the prompt word.

[0035] During model training, different diagnosis and treatment fields have different degrees of impact on treatment, so it is necessary to set priorities for each field. For example, the chief complaint and current medical history may be more important than past medical history, so these fields have a higher priority. You can also directly set the chief complaint, the four traditional Chinese medicine examinations, and the traditional Chinese medicine diagnosis as high priority. In this way, the model can pay more attention to the fields that have a greater impact on the recommendation results during training. The binding of priorities to prompt words is to ensure that the model gives priority to high-priority fields during the analysis process, thereby improving the accuracy of recommendations.

[0036] Step S204 : integrating the key diagnosis and treatment information fields with added prompt words and bound priorities to generate standard medical record data entries.

[0037] After completing the prompt word and priority binding, all diagnosis and treatment fields are integrated into a standardized medical record data entry. These entries are formatted uniformly to ensure that each record contains the necessary diagnosis and treatment information and is organized according to specific rules. The purpose of this step is to convert the raw medical record data into a structured, unified data format so that it can be smoothly input into the subsequent learning model training.

[0038] Step S205 , aggregating multiple standard medical record data entries to form a standard medical record data sample set.

[0039] Finally, all preprocessed and standardized medical record data entries are aggregated into a standard medical record data sample set. This sample set serves as the foundation for model training. It includes detailed medical records for multiple patients, covering a wide range of possible symptoms and treatments. This standard medical record data sample set serves as input for language model training, providing the necessary information and data support for subsequent model learning.

[0040] Step S300: Input the standard medical record data sample set into a preset language model, and preliminarily train the language model based on the standard medical record data sample set, so that the language model establishes a semantic mapping relationship between patient symptom information and acupuncture points.

[0041] A sample set of standard medical record data is input into a language model based on the Transformer Decoder architecture. The model extracts semantic features from the text through positional encoding and self-attention mechanisms, thereby capturing key information in the patient's symptom description. The model then compares the predicted acupuncture point information with the pre-labeled real information, calculates the error, and iteratively updates the model parameters using the backpropagation algorithm. When the error drops to the preset convergence standard, the model outputs a preliminarily trained language model, which has successfully established a semantic mapping between patient symptom information and acupuncture points, laying the foundation for subsequent model fine-tuning and optimization.

[0042] In an optional embodiment, step S300 includes the following sub-steps: Step S301 : performing word segmentation and encoding processing on the standard medical record data sample set to generate a numerical representation acceptable to the language model.

[0043] Specifically, each medical record is first divided into words or subwords using a pre-set word segmentation algorithm. These text units are then converted into corresponding numerical identifiers using a vocabulary. The resulting numerical vector representations constitute the acceptable input data for the model, ensuring that all input data has a unified and standardized format, providing a foundation for subsequent model training.

[0044] In step S302, the standard medical record data sample set that has been processed by word segmentation and encoding is input into a language model based on the TransformerDecoder architecture, and the semantic features in the standard medical record data sample set are extracted using position encoding and attention mechanism.

[0045] The standard medical record data sample set that has been word segmented and encoded is input into a language model built based on the Transformer Decoder architecture (for example, Llama3). The language model is trained with a large amount of text data and is able to understand the complex structure of natural language, especially the terms and descriptions in the medical field. Figure 2 As shown, the Llama3 model has two architectural styles: a standard architecture, in which input data sequentially enters the attention module, undergoes processing through a multi-layer perceptron (MLP), and then undergoes layer normalization (LayerNorm); and a pre-normalized architecture, in which input data first undergoes root mean square normalization (RMSNorm) before entering the attention module and the multi-layer perceptron (MLP). This design ensures a stable gradient flow and efficient information transfer when processing symptom descriptions, thereby better capturing the complex terminology and descriptions in the medical field.

[0046] Next, if Figure 3 As shown in the figure, the enhanced Transformer architecture is used to improve the model's ability to model meridian data in time series. "Enhanced" refers to structural optimizations made to the basic Transformer architecture, making it more suitable for representing complex positional information. The Transformer is a deep neural network architecture based on the self-attention mechanism, widely used in natural language processing and sequence modeling. Its core advantage lies in its ability to process input sequences in parallel and capture long-range dependencies between different positions.

[0047] To further enhance the Transformer's ability to perceive positional information in sequences, the Rotary Position Embedding (RoPE) mechanism is introduced. This mechanism encodes positional information as rotations between vectors, embedding absolute positions as angles into queries and keys, achieving unified modeling of relative and absolute position relationships. "Rotary" refers to injecting positional information (e.g., 1, 2, 3, etc.) into vectors as rotation angles, "Position" refers to the position index of each element in the sequence, and "Embedding" means encoding this information into a learnable or executable vector form and embedding it into the model.

[0048] For example, RoPE constructs a two-dimensional coordinate system on adjacent dimensions of a vector (e.g., , ), and applies a rotation angle related to position m to this coordinate system , thus obtaining the new coordinates after rotation .

[0049] Where m represents the position index of the element in the input sequence; Represents the parameters of the model (e.g., coefficients or number of fields for positional encoding); Query (query vector) and Key (key vector) are used to calculate attention weights respectively; RoPE applies this position-aware rotation to both the query and key, allowing the attention weights to carry both semantic and spatial position information. In the model, each position-encoded query and key pair is fed into a self-attention mechanism for weighted calculations, generating a contextual representation that incorporates both semantic features and position information. This approach allows the model to preserve the spatial positional attributes of each element when processing long sequences, spatiotemporal data, or complex meridian sequences, significantly improving its perception of local structure and global distribution.

[0050] In step S303 , based on the semantic features, the acupuncture point information predicted by the language model is compared with the pre-labeled acupuncture point information, and the error is calculated using a preset loss evaluation standard.

[0051] The model uses the extracted semantic features to predict acupuncture point information and compares this prediction with pre-labeled true acupuncture point information to calculate the prediction error. This error signal is generated using a preset loss metric (e.g., cross-entropy or mean squared error). This error signal is then propagated through the backpropagation algorithm to the parameters of each layer of the model. These parameters are then updated using gradient descent, thereby continuously reducing the prediction error and optimizing the semantic mapping between symptom information and acupuncture points.

[0052] In addition, if Figure 4 As shown in Figure 2, the Llama3 model also employs a multi-head and grouped-query attention mechanism. In the multi-head approach, each attention head has independent queries, keys, and values ​​to capture information in different subspaces. In the grouped-query approach, multiple queries are merged or grouped to reduce the need for repeated computations on each attention head, thereby lowering computational overhead. This approach retains the expressive power of multi-head attention to a certain extent while reducing the use of computing resources.

[0053] Step S304: iteratively update the parameters of the language model through a back-propagation function.

[0054] Using a backpropagation algorithm, the error signal is transmitted to each layer of the model parameters, and the model parameters are iteratively updated using gradient descent. Each update adjusts the parameters based on the error calculated in the previous round, gradually narrowing the gap between the model output and the actual acupuncture point information, thereby continuously improving the model's ability to learn the semantic mapping between symptom information and acupuncture points.

[0055] Step S305: When the error during training drops to a preset convergence standard, the preliminarily trained language model is output.

[0056] When the training error drops to the preset convergence standard, the training process ends and the preliminarily trained language model is output. At this point, the model has successfully established a semantic mapping relationship between patient symptom information and acupuncture points, laying a solid data foundation for subsequent model fine-tuning and further optimization.

[0057] Step S400 , performing parameter fine-tuning and strategy optimization on the preliminarily trained language model in sequence to obtain an acupuncture point recommendation model.

[0058] After initial training, the trained language model undergoes two stages of fine-tuning to improve recommendation accuracy. First, during the parameter fine-tuning phase, the model decomposes the pre-trained weights into magnitude and direction. The direction is fixed, and only the magnitude is updated, allowing the model to more accurately capture the association between symptoms and acupuncture points. Subsequently, during the strategy optimization phase, a reinforcement learning feedback mechanism is employed to further adjust the model strategy by introducing reward signals based on the degree of match between the model's recommendations and actual treatment outcomes.

[0059] In an optional embodiment, step S400 includes the following sub-steps: Step S401 : Decomposing the weight of the preliminarily trained language model into a magnitude component and a direction component.

[0060] First, the first stage of DoRA fine-tuning (directional optimization fine-tuning) is performed. In this stage, the model parameters are decomposed into magnitude and direction components. DoRA fine-tuning focuses on adjusting the magnitude component to enhance the model's understanding of key information (for example, the correspondence between symptoms and treatments). For example, the training data is diversified and enhanced through methods such as synonym replacement and sentence reorganization. Then, during fine-tuning, error weights are assigned to different data samples to highlight the role of key samples. By adjusting only the magnitude component and freezing the direction component, the model's sensitivity in specific areas (for example, acupuncture point recommendations) is improved while retaining its ability to understand the overall task.

[0061] The magnitude component controls the model's learning intensity for a specific feature in different scenarios, representing the magnitude of the model's response to certain keywords or features. The direction component controls the direction of the model's parameters in different dimensions, determining the direction of the model's response when processing different inputs. The advantage of this decomposition strategy is that by adjusting the magnitude, the model can be made more sensitive in specific areas (for example, acupuncture point recommendations), while fixing the direction component ensures the model's ability to handle the overall problem.

[0062] Step S402 : fix the direction component, adjust the amplitude component according to a preset weight adjustment strategy, and output a parameter fine-tuning result.

[0063] After the parameter decomposition is completed, the direction component will be frozen to ensure that it does not change during the fine-tuning process. Next, the magnitude component will be fine-tuned according to the preset weight adjustment strategy. The core idea of ​​the weight adjustment strategy is to give each data sample different importance and adjust the magnitude component according to the priority of the sample. For example, if certain data samples (such as the patient's chief complaint or TCM four diagnostic information) are considered more important in acupuncture recommendations, then the model will give these samples greater weights during the fine-tuning process. In this way, the model will learn that in specific cases, certain symptom information has a greater impact on acupuncture treatment, thereby enhancing the model's attention to this information. After fine-tuning, the magnitude component of the model will become more adapted to the specific treatment recommendation task.

[0064] For example, if Figure 5 As shown, : Represents the original pre-trained parameter matrix; in the decomposition / initialization step, the original weights of the model are split or initialized into several parts for subsequent finer fine-tuning; amplitude: refers to the quantization of the size of the weights. Direction: It is usually obtained by normalizing the weights of the model, indicating the direction in which the parameters point in the vector space. Freeze: refers to the process of fine-tuning in which certain parameters or components remain unchanged and do not participate in training updates; Trainable: refers to the parameter components that need to be updated according to specific tasks or strategies during the fine-tuning process; Adaptation: refers to the stage in which the model only trains and optimizes the "amplitude" components to adapt to the target task. Merge refers to reintegrating the updates obtained after fine-tuning and adaptation in the previous step with the initially decomposed parts into new weights; W′: represents the new weights after the merger, that is, a parameter matrix that retains the original general capabilities while also having optimization results for the target task (for example, acupuncture recommendations).

[0065] In detail: 1. Decomposition and initialization: from pre-trained weights First, split the original weight into two parts: magnitude and direction. This operation aims to control the "parameter size" and "parameter direction in space" separately, preparing for modifying only part of them later.

[0066] 2. Adaptation: In this stage, Figure 5 Parameters marked "Pretrained Weights (Frozen ①)" are no longer updated, while those marked "Magnitude (Trainable ②)" are fine-tuned based on task requirements. For example, direction ① can be frozen, and only magnitude ② can be adjusted based on weight priority or data sample importance. Alternatively, the opposite operation can be performed, or both magnitude and direction can be partially updated simultaneously. To achieve this update, a loss function is constructed to measure the error between the model output and the desired target:

[0067] in, The total loss function; θ refers to the parameters being fine-tuned (e.g., magnitude m, direction n); is the weighted average loss; is the regularization term; is the sample weight; N is the total number of training samples; is the loss function of a single sample; is the regularization coefficient; is the regularization function.

[0068] Next, the gradient descent method is used to update the trainable parameters. The update formula is:

[0069] in, : represents the model parameters after the tth round of update and the t+1th round of update; ; : Learning rate, which determines the step size of each update; : Loss function parameters The gradient reflects the error direction and magnitude under the current parameters.

[0070] 3. Merge: After the adaptation is completed, the updated trainable part is merged with the frozen part to generate new weights ( ).at this time, It not only inherits the general knowledge from the pre-training stage, but also incorporates specialized adaptation to the target task, resulting in higher accuracy and adaptability.

[0071] In step S403, based on the parameter fine-tuning results, a feedback mechanism based on reinforcement learning is constructed. By introducing a strategy reward signal, the language model is guided to perform strategy optimization to form an acupuncture point recommendation model.

[0072] During the RLAIF fine-tuning process, the model was further optimized using GPT-4. This process uses reinforcement learning (RLAIF) to continuously adjust and optimize the model, gradually improving the accuracy of its recommendations. The RLAIF approach enables the model to continuously improve its decision-making capabilities in complex environments based on feedback from existing AI, and optimizes its behavior through enhanced information feedback. In this embodiment, the reward signal primarily comes from the accuracy of the model's recommendations. For example, if the model's recommended acupuncture points based on symptom information match the patient's actual symptoms, the model receives a positive reward; conversely, if the recommendation is inaccurate, it receives negative feedback.

[0073] In order to guide the model to optimize its strategy, this embodiment also designs a reward function to evaluate the quality of the recommendation results output by the model. This function calculates the reward based on the matching degree between the acupoints recommended by the model and the actual symptoms of the patient. Specifically, the reward function considers the following factors: (1) Symptom matching: the correlation between the acupoints recommended by the model and the patient's symptoms. (2) Diagnostic accuracy: whether the acupuncture treatment plan recommended by the model is consistent with the treatment rules in traditional Chinese medicine theory and the medical knowledge base. (3) User feedback: if the model has been deployed and run, the actual user's satisfaction feedback on the recommended plan. By guiding the reward signal, the model gradually learns to optimize its recommendation decisions when processing different symptom information, thereby improving its overall performance.

[0074] For example, Figure 6As shown in Figure 1, the reinforcement learning process is divided into two main paths: the existing model feedback path and the human feedback path. In the existing model feedback path, responses generated by the model (e.g., r1, r2) are passed to the existing model for scoring. These scores are used to train the reward model (RM from AI Feedback). Through reinforcement learning with the reward model, the model continuously optimizes its recommendation strategy. In the human feedback path, humans rate the responses generated by the model and train the reward model. The reward model is then used to optimize the model strategy, improving its accuracy and adaptability.

[0075] After multiple rounds of reinforcement learning and strategy optimization, the model will gradually master how to make more accurate decisions in complex diagnosis and treatment environments. Ultimately, the optimized acupuncture point recommendation model will be able to recommend the most appropriate acupuncture point treatment plan based on the symptom information provided by the patient.

[0076] Figure 7 A flowchart of the acupuncture point recommendation method according to an embodiment of the present application is shown. Exemplarily, the method includes the following steps: Step S101 : In response to received symptom information of a patient to be diagnosed, extract key diagnosis and treatment information of the symptom information of the patient to be diagnosed, and generate a standardized symptom description.

[0077] After receiving the symptom information submitted by the patients awaiting diagnosis, natural language processing technology is first used to process the input symptom information of the patients awaiting diagnosis. Specifically, the original symptom information is segmented and semantically analyzed to extract key diagnosis and treatment information, such as the patient's chief complaint, current medical history, past medical history, four examinations of traditional Chinese medicine, and traditional Chinese medicine diagnosis. After these key information are extracted, they will be expanded according to the preset prompt word template. The prompt word template is constructed by medical experts and manual annotations to enhance the model's understanding of the implicit meaning in the text. Subsequently, priorities are set according to the importance of each diagnosis and treatment field in acupuncture diagnosis, and these priorities are bound to the corresponding prompt words to ensure that the core information is focused on during subsequent data analysis. Finally, all processed and enhanced key information is integrated into a standardized symptom description in a unified format.

[0078] Step S102: using the acupuncture point recommendation model, combining TCM theory rules and the medical knowledge base to process the standardized symptom description, and generating acupuncture point recommendation information corresponding to the symptom information.

[0079] After the standardized symptom description is generated, the description is sent as input to the acupuncture point recommendation model. The model is obtained using the method of any of the above embodiments, that is, a semantic mapping relationship between the patient's symptom information and the acupuncture points has been established. In actual applications, the model performs semantic analysis on the standardized symptom description and outputs preliminary acupuncture point recommendation results. Subsequently, in order to further improve the accuracy of the recommendation, the preliminary recommendation results are calibrated in combination with the theoretical rules of traditional Chinese medicine and the medical knowledge base. The calibration process includes comparing the model prediction with the description of the acupoint indications in traditional Chinese medicine theory, and adjusting the recommendation plan based on the treatment records in the medical knowledge base. Ultimately, the acupuncture point recommendation information after rule calibration can accurately reflect the patient's symptom characteristics and provide scientific and personalized acupuncture treatment recommendations for clinicians.

[0080] For example, take a patient as an example: the main complaint is "stomach discomfort accompanied by constipation for 2 weeks".

[0081] The patient's medical history revealed that she had experienced stomach discomfort for two weeks without apparent cause, accompanied by acid reflux and heartburn, abdominal distension after eating, and hard stools (with bowel movements occurring every two to three days). This was accompanied by neck aches, lower back pain, intermittent numbness in her left hand, and intermittent coughing (thick sputum). She had no nausea, vomiting, cough, or fever, and her symptoms did not improve significantly with rest. She was admitted for treatment combining traditional Chinese and Western medicine.

[0082] The patient's mental state was good, and his diet and sleep were normal. His stools were dry, and his urination was normal. He had been suffering from type 2 diabetes for over a month and was taking oral metformin and Xiaoke pills to control his blood sugar. He had no history of hypertension, coronary heart disease, hepatitis, tuberculosis, major surgery, or blood transfusions, and his allergy history was unclear.

[0083] The information of the four diagnostic methods of Traditional Chinese Medicine is as follows: Inspection: clear consciousness, ruddy complexion, moderate body shape, dark red tongue with thin white coating, slightly dilated sublingual veins; Auscultation: loud voice; Medical Questions: Stomach discomfort, acid reflux and heartburn, abdominal distension, constipation, neck and back pain, left hand numbness, intermittent cough with sputum (thick sputum); Palpation: The pulse is stringy and slippery.

[0084] Western medical signs and examination results are as follows: The body temperature was 36.5°C, the pulse was 85 beats / min, the respiration rate was 17 breaths / min, and the blood pressure was 120 / 54 mmHg. The abdominal examination showed no tenderness, rebound tenderness, or edema, and the lower limbs were normal.

[0085] The comprehensive diagnosis is as follows: TCM diagnosis: stomach disease, syndrome is spleen and stomach deficiency; Western medicine diagnosis: ① Gastric bloating to be investigated; ② Type 2 diabetes.

[0086] An acupuncture point recommendation model analyzes this symptom information, integrating it with Traditional Chinese Medicine theory and a medical knowledge base to generate recommended acupoints corresponding to the symptoms. Examples include Neiguan (regulates stomach qi and relieves stomach discomfort), Shenmen (calms the mind and regulates spleen and stomach function), Zusanli (tonifies the spleen and stomach, promotes digestion, and relieves constipation), Pishu (strengthens spleen and stomach function and promotes food digestion), and Daheng (relieves abdominal distension and regulates the stomach). This process aims to provide patients with personalized treatment plans through data input and model calculations. Through the acupuncture point recommendation model, symptom information is effectively decoded and compared with theory and data in the medical knowledge base to generate precise acupuncture treatment recommendations, thereby improving clinical treatment outcomes.

[0087] In an optional embodiment, the acupuncture point recommendation method further includes: The acupuncture point recommendation information is mapped to an acupuncture bronze man model corresponding to the three-dimensional human body structure, so as to display the recommended acupuncture points on the acupuncture bronze man model.

[0088] For example, the bronze acupuncture man model is a classic tool in traditional Chinese medicine acupuncture theory, helping acupuncturists understand the distribution of meridians and acupoints in the human body. To visualize acupuncture recommendations, a precise bronze acupuncture man model was constructed using 3D modeling technology. This model not only reflects the basic human anatomy but also features multiple identifiable markers on its surface that correspond to actual acupoints. These markers correspond to specific acupoints on the human body, such as Hegu, Baihui, and Zusanli, which are commonly used acupuncture points. These markers are uniquely coded within the model to match the acupoint data in the acupuncture recommendation results. Through 3D modeling, the bronze man model visually represents a rotatable and interactive virtual human body, allowing users to view, understand, and practice from different angles.

[0089] To effectively display the recommended acupoints, the acupoint recommendations must be converted into mapping data corresponding to identifiable landmarks. Specifically, each recommended acupoint corresponds to one or more specific locations on the bronze mannequin. Using this mapping data, the recommended acupoints can be accurately located at the landmarks on the bronze mannequin.

[0090] Mapping data is generated by matching the coordinates of the recommended acupoints with the markers in the model. The core of this process is to establish a precise mathematical relationship between the output acupoints and the markers on the 3D model, allowing the recommended results to be directly reflected in the model. This process also takes into account the size, shape, and meridian alignment of the acupoints to ensure mapping accuracy and consistency with Traditional Chinese Medicine theory.

[0091] The visual mapping data is input into the acupuncture bronze man model to display the recommended acupuncture points on the acupuncture bronze man model.

[0092] For example, the converted visual mapping data is input into an acupuncture bronze man model, and the recommended acupuncture points are visually displayed on the model. For example, this process is implemented through a graphical user interface (GUI), where users can click on different acupuncture points on the model to view detailed treatment information and operation instructions.

[0093] At this stage, users can not only see the recommended acupoints but also interact with them on the model. For example, clicking on an acupoint on the model will display relevant information such as its name, function, indications, and operation methods, helping medical students, doctors, or patients better understand how to practice acupuncture for treatment.

[0094] In an optional embodiment, step S102 includes: Extract key diagnostic and treatment information from the patient's symptom information and generate standardized symptom descriptions.

[0095] First, key information from the symptoms must be extracted. This information may include the patient's chief complaint, current medical history, and physical condition. Natural language processing technology is used to perform semantic analysis on this information to identify the most relevant diagnostic and treatment elements. This key information is then organized into standardized symptom descriptions for subsequent use in the acupuncture point recommendation model.

[0096] The standardized symptom description is input into the acupuncture point recommendation model for semantic analysis to obtain the initial acupuncture point recommendation results.

[0097] The standardized symptom information was input into the acupuncture point recommendation model. Based on previous training, the model processed the input symptom description through semantic analysis and output preliminary acupuncture point recommendation results, as shown in Table 1. This identifies the most relevant recommended acupoints for the symptoms, acupoint location descriptions, and other symptoms associated with the acupoints.

[0098]

[0099] Table 1 The initial acupuncture point recommendation results are calibrated using preset TCM theoretical rules and medical knowledge base to generate recommended acupoints corresponding to symptom information.

[0100] For example, the initial recommendations are further calibrated based on TCM theory and the medical knowledge base. The TCM theory base provides detailed rules regarding symptoms and treatment options, and, combined with known medical literature, helps the model fine-tune its recommendations. For example, the model may recommend multiple acupoints for certain symptom descriptions. By comparing TCM theory with similar cases in the knowledge base, the model selects the most appropriate acupoint for the final recommendation. This process ensures the medical legitimacy of the recommendations and the reliability of the treatment outcomes.

[0101] Figure 8 A schematic diagram of the structure of the device for obtaining acupuncture point recommendation model according to an embodiment of the present application is shown. Exemplarily, the device 100 includes: The data collection module 110 is used to collect patient medical record data, and perform format conversion, encoding, and field classification extraction on the collected patient medical record data to form a medical record data sample set; The data processing module 120 is used to pre-process the medical record data sample set to generate a standard medical record data sample set; A model training module 130 is configured to input the standard medical record data sample set into a preset language model and perform preliminary training on the language model based on the standard medical record data sample set so that the language model establishes a semantic mapping relationship between patient symptom information and acupuncture points; The acupoint generation module 140 is used to perform parameter fine-tuning and strategy optimization on the language model that has been preliminarily trained in sequence to obtain an acupuncture point recommendation model; the acupuncture point recommendation model is used to output the mapping result between the symptom information and the corresponding acupuncture points.

[0102] It can be understood that the apparatus of this embodiment corresponds to the method of the above embodiment, and the options in the above embodiment are also applicable to this embodiment, so they will not be described again here.

[0103] The present application also provides a computer device. Exemplarily, the computer device includes a processor and a memory, wherein the memory stores a computer program, and the processor runs the computer program, thereby enabling the computer device to execute the functions of each module in the above method or the above apparatus.

[0104] The processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, including at least one of a central processing unit (CPU), a graphics processing unit (GPU), a network processor (NP), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor or any conventional processor, etc., and can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of this application.

[0105] The memory may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), and electrically erasable programmable read-only memory (EEPROM). The memory is used to store computer programs, and the processor can execute the computer programs accordingly after receiving an execution instruction.

[0106] This application also provides a computer-readable storage medium for storing the computer program used in the aforementioned computer device. For example, the computer-readable storage medium may include, but is not limited to, various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0107] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and structure diagrams in the accompanying drawings show the possible architectures, functions and operations of the devices, methods and computer program products according to the multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of the code, and the module, program segment or a part of the code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in an alternative implementation, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the structure diagram and / or flowchart, and the combination of boxes in the structure diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or can be implemented using a combination of dedicated hardware and computer instructions.

[0108] In addition, the functional modules or units in the various embodiments of the present application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0109] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a smart phone, personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application.

[0110] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A method for obtaining an acupuncture point recommendation model, characterized in that: The method comprises: Collecting patient medical record data, and performing format conversion, encoding, and field classification extraction operations on the collected patient medical record data in sequence to form a medical record data sample set; Preprocessing the medical record data sample set to generate a standard medical record data sample set; Inputting the standard medical record data sample set into a preset language model, and preliminarily training the language model based on the standard medical record data sample set, so that the language model establishes a semantic mapping relationship between patient symptom information and acupuncture points; Parameter fine-tuning and strategy optimization are sequentially performed on the preliminarily trained language model to obtain an acupuncture point recommendation model; the acupuncture point recommendation model is used to output a mapping result between the symptom information and the corresponding acupuncture points.

2. The method for obtaining an acupuncture point recommendation model according to claim 1, wherein: The medical record data sample set includes a plurality of initial medical record data entries; the preprocessing of the medical record data sample set to generate a standard medical record data sample set includes: Performing semantic analysis on each of the initial medical record data entries in the medical record data sample set to extract key diagnosis and treatment fields; Based on the preset prompt word template, add corresponding prompt words for the key diagnosis and treatment fields; Setting a corresponding priority according to the importance of the key diagnosis and treatment fields in acupuncture diagnosis, and binding the priority to the prompt word; Integrate the key diagnosis and treatment information fields with the prompt words added and bound to the priority to generate a standard medical record data entry; A plurality of the standard medical record data entries are aggregated to form the standard medical record data sample set.

3. The method for obtaining an acupuncture point recommendation model according to claim 1, wherein: Inputting the standard medical record data sample set into a preset language model and performing preliminary training on the language model based on the standard medical record data sample set includes: Performing word segmentation and encoding processing on the standard medical record data sample set to generate a numerical representation acceptable to the language model; Inputting the standard medical record data sample set that has undergone the word segmentation and encoding processing into the language model based on the Transformer Decoder architecture, and extracting semantic features from the standard medical record data sample set using position encoding and attention mechanism; Based on the semantic features, the acupuncture point information predicted by the language model is compared with the pre-labeled acupuncture point information, and the error is calculated using a preset loss evaluation standard; Iteratively updating the parameters of the language model through a back-propagation function; When the error during training drops to a preset convergence standard, the language model that has undergone preliminary training is output.

4. The method for obtaining an acupuncture point recommendation model according to claim 1, wherein: The method sequentially performs fine-tuning and strategy optimization on the preliminarily trained language model to obtain an acupuncture point recommendation model, including: Decomposing the weight of the language model after preliminary training into an amplitude component and a direction component; Fix the direction component, and update the amplitude component according to a preset weight adjustment strategy, and output a parameter fine-tuning result; Based on the parameter fine-tuning results, a feedback mechanism based on reinforcement learning is constructed. By introducing a strategy reward signal, the language model is guided to perform strategy optimization to form the acupuncture point recommendation model; wherein the reward signal is determined according to the degree of match between the acupuncture point recommendation information and the actual treatment effect.

5. A method for recommending acupuncture points, characterized in that: The method comprises: In response to the received symptom information of the patient to be diagnosed, extract key diagnosis and treatment information of the symptom information of the patient to be diagnosed, and generate a standardized symptom description; The standardized symptom description is processed using an acupuncture point recommendation model in combination with TCM theory rules and a medical knowledge base to generate acupuncture point recommendation information corresponding to the symptom information; wherein the acupuncture point recommendation model is obtained using the acupuncture point recommendation model acquisition method as described in any one of claims 1-4.

6. The acupuncture point recommendation method according to claim 5, characterized in that: The method further comprises: The acupuncture point recommendation information is mapped onto an acupuncture bronze mannequin corresponding to a three-dimensional human body structure, so as to display the recommended acupuncture points on the acupuncture bronze mannequin; wherein the acupuncture bronze mannequin is provided with a plurality of marking points corresponding to human acupuncture points; the mapping process includes: Converting the acupuncture point recommendation information into mapping data corresponding to the marked points; The mapping data is input into the acupuncture bronze man model to display the acupuncture point recommendation information on the acupuncture bronze man model.

7. The acupuncture point recommendation method according to claim 5, characterized in that: The acupuncture point recommendation model is used to process the standardized symptom description in combination with traditional Chinese medicine theory and a medical knowledge base to generate acupuncture point recommendation information corresponding to the symptom information, including: Extracting key diagnostic and treatment information from the patient's symptom information to generate a standardized symptom description; Inputting the standardized symptom description into the acupuncture point recommendation model for semantic analysis to obtain initial acupuncture point recommendation information; The initial acupuncture point recommendation information is calibrated using the TCM theory rules and the medical knowledge base to generate the acupuncture point recommendation information corresponding to the symptom information for recommendation.

8. A device for obtaining a recommended acupuncture point model, characterized in that: The device comprises: The data collection module is used to collect patient medical record data, and perform format conversion, encoding, and field classification extraction on the collected patient medical record data in sequence to form a medical record data sample set; A data processing module, configured to pre-process the medical record data sample set to generate a standard medical record data sample set; a model training module, configured to input the standard medical record data sample set into a preset language model, and perform preliminary training on the language model based on the standard medical record data sample set, so that the language model establishes a semantic mapping relationship between patient symptom information and acupuncture points; The acupoint generation module is used to perform parameter fine-tuning and strategy optimization on the language model that has been preliminarily trained in sequence to obtain an acupuncture point recommendation model; the acupuncture point recommendation model is used to output the mapping result between the symptom information and the corresponding acupuncture points.

9. A computer device, characterized in that: The computer device includes a processor and a memory, wherein the memory stores a computer program, and the processor is configured to execute the computer program to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The device stores a computer program, which, when executed on a processor, implements the method according to any one of claims 1 to 7.

Images