Intelligent acupuncture inquiry method and system based on continuous dialogues

Abstract

The invention relates to an intelligent acupuncture inquiry method and system based on continuous dialogues. A pre-trained candidate symptom prediction model is mainly applied to obtain candidate symptoms according to input questions of a user; wherein the candidate symptom prediction model is embedded in a ChatGPT inquiry model; the ChatGPT inquiry model is also used for determining a question and answer list based on candidate symptoms; interacting with the user according to the question and answer list, and determining an actual symptom; and outputting acupuncture points according to actual symptoms. According to the method, a more accurate auxiliary diagnosis result can be obtained through the continuous dialogue ChatGPT inquiry model.

Description

Technical Field

[0001] The present invention relates to the technical field of traditional Chinese medicine consultation, and more specifically to an intelligent acupuncture consultation method and system based on continuous dialogue. Background Art

[0002] As a traditional medical method, acupuncture has gained wide international recognition and clinical validation worldwide for its unique efficacy, no adverse reactions, and portability. However, the complexity of TCM syndrome differentiation theory and the extensiveness of the knowledge system pose great challenges to acupuncturists in terms of memorization and mastery. Especially in community medical services, acupuncturists are often lacking in qualifications and experience, which limits the application and development of acupuncture treatment in primary medical care.

[0003] In order to solve this problem, in recent years, researchers have begun to explore the introduction of artificial intelligence technology into the field of acupuncture, hoping to assist acupuncturists in making more comprehensive and accurate diagnosis and treatment through intelligent means. Advanced artificial intelligence technologies such as deep neural networks, reinforcement learning, and genetic algorithms have shown great potential in the medical field.

[0004] In particular, ChatGPT, as a natural language processing tool based on the Transformer neural network architecture, can achieve fluent human-computer dialogue through pre-training models, which has extremely high application value in the field of medical consultation. ChatGPT can understand and generate natural language by connecting a large number of corpora for training, which is of great significance for building an intelligent acupuncture consultation system.

[0005] Although ChatGPT technology has made some progress in the medical field, such as in critical care, infectious diseases, and public health, it is still a blank in the field of intelligent Chinese medicine research, especially in the application of intelligent acupuncture consultation. In addition, ChatGPT technology has certain limitations in data security, technical limitations, susceptibility to interference, humanization, machine thinking skills, and equipment requirements. These are all issues that need to be considered when developing an intelligent acupuncture consultation system.

[0006] Currently, there is no effective intelligent acupuncture consultation system that can meet the needs of personalized consultation while ensuring the security of medical data and complying with acupuncture thinking and equipment requirements. Summary of the invention

[0007] In view of this, in order to at least partially solve the above technical problems, the present invention provides a novel intelligent acupuncture consultation method and system based on continuous dialogue to promote the precision and intelligent development of acupuncture treatment.

[0008] In order to achieve the above object, the present invention adopts the following technical solution:

[0009] First, the present application discloses an intelligent acupuncture consultation method based on continuous dialogue, including:

[0010] Applying a pre-trained candidate symptom prediction model to obtain candidate symptoms according to the user's input questions; wherein the candidate symptom prediction model is embedded in the ChatGPT consultation model;

[0011] The ChatGPT consultation model is also used to determine a question and answer list based on candidate symptoms;

[0012] Interact with users based on the question and answer list to determine the actual symptoms;

[0013] And output acupuncture points according to actual symptoms.

[0014] Preferably, determining a question-answer list based on candidate symptoms comprises:

[0015] Constructing a symptom database and a question database, wherein the symptom database stores a mapping relationship between symptoms and symptoms, and the question database stores questions corresponding to the symptoms and candidate answers corresponding to the questions;

[0016] Determine the corresponding symptoms according to the candidate symptoms based on the symptom library, and obtain the corresponding questions and the candidate answers corresponding to the questions according to the symptoms based on the question library;

[0017] The question and the corresponding candidate answer form a question-answer pair, and a question-answer list is generated after removing duplicates.

[0018] Preferably, when interacting with the user according to the question and answer list, judging the user's symptoms includes:

[0019] If the symptoms are judged to be mild, provide guidance to alleviate the user's symptoms;

[0020] When it is judged as an emergency symptom, an emergency plan is output and the nearest hospital is selected for treatment.

[0021] Preferably, the step of determining the actual symptoms comprises:

[0022] S1. According to the question list, confirm with the user whether the symptom indicated by the question exists. If so, update the existing symptom and jump to step S2. If not, jump directly to step S2.

[0023] S2. Determine whether all the questions in the question list have been traversed. If not, return to step S1 and ask the next question. If so, determine the actual symptom based on the current symptom, including inferring the matching symptom based on the symptom library according to the current symptom, and taking the symptom with the highest matching degree as the actual symptom.

[0024] Preferably, when interacting with the user according to the question and answer list, the question and answer list is mapped into 13 categories, including: cold and heat, sweat, pain, taste, ears, eyes, chest, flank, stomach and abdomen, sleep, thirst and drinking water, appetite and food intake, and defecation; and the order of questions and answers is determined according to the ten-question song.

[0025] Preferably, acupuncture literature and acupuncture clinical cases are collected to construct a data set, and the pre-trained ChatGPT consultation model is fine-tuned to output acupuncture points according to actual symptoms.

[0026] Preferably, the candidate symptom prediction model is optimized according to the actual symptoms; and a scoring model is constructed to score the acupuncture points output by the ChatGPT consultation model, and the ChatGPT consultation model is optimized according to the scoring results.

[0027] Preferably, the ChatGPT consultation model is optimized according to the scoring results, including optimizing the automatic acupuncture point recognition model based on reinforcement learning according to the scoring results and the error of the preset scoring constant. The reward agent model expression of the optimization process is:

[0028]

[0029] In the formula, s t represents the transition state at time t; s t+1 represents the transition state at time t+1; p represents the preset scoring constant, ε t+1 Indicates the error between the scoring result at time t+1 and the preset scoring constant, ε t Indicates the error between the scoring result at time t and the preset scoring constant.

[0030] Second, the present application discloses an intelligent acupuncture consultation system based on continuous dialogue, which applies the intelligent acupuncture consultation method based on continuous dialogue as described above, wherein the system includes a client, a ChatGPT consultation model is built in the client, and a pre-trained candidate symptom prediction model is embedded in the ChatGPT consultation model,

[0031] The candidate symptom prediction model obtains candidate symptoms according to the user's input questions;

[0032] The ChatGPT consultation model is also used to determine a question and answer list based on candidate symptoms;

[0033] Interact with users based on the question and answer list to determine the actual symptoms;

[0034] And output acupuncture points according to actual symptoms.

[0035] Preferably, the system further comprises a server, wherein the server stores a symptom library and a question library, wherein the symptom library stores a mapping relationship between symptoms and symptoms, and the question library stores questions corresponding to the symptoms and candidate answers corresponding to the questions;

[0036] The server is used to determine the corresponding symptoms according to the candidate symptoms based on the symptom library, and obtain the corresponding questions and the candidate answers corresponding to the questions according to the symptoms based on the question library;

[0037] The question and the corresponding candidate answer are formed into question-answer pairs, and a question-answer list is generated after removing duplicates.

[0038] It can be seen from the above technical solution that compared with the prior art, the present invention discloses a new type of intelligent acupuncture consultation method and system based on continuous dialogue. The system can combine the ChatGPT pre-trained consultation model and the learnable reinforcement learning scoring model, and through the iterative update and optimization of a large amount of clinical data, to achieve more accurate auxiliary diagnosis results.

[0039] Specifically, compared with the prior art, the beneficial effects of the present application include:

[0040] By combining the ChatGPT pre-training model and reinforcement learning technology, a more humane and efficient acupuncture consultation process is achieved, effectively solving the problem of doctors remembering and mastering complex dialectical theories in traditional acupuncture consultation; at the same time, more precise acupuncture points are determined based on the patient's actual symptoms, improving the pertinence and effectiveness of acupuncture treatment.

[0041] The present invention simplifies the acupuncture consultation process, making acupuncture treatment easier to promote and apply in primary medical services; at the same time, it provides a powerful auxiliary tool for acupuncturists, which helps to improve the diagnosis and treatment level of doctors, especially in community medical services with insufficient qualifications and experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0043] Figure 1 This is a flow chart of the intelligent acupuncture consultation method based on continuous dialogue of the present invention;

[0044] Figure 2 A schematic diagram of a process for determining actual symptoms of the present invention;

[0045] Figure 3This is a schematic diagram of the optimization process of the ChatGPT consultation model of the present invention;

[0046] Figure 4 It is a schematic diagram of the reinforcement learning process of the present invention;

[0047] Figure 5 This is a process diagram for determining actual symptoms by the intelligent acupuncture consultation system based on continuous dialogue of the present invention. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] Embodiment 1

[0050] The embodiment of the present invention discloses an intelligent acupuncture consultation method based on continuous dialogue, such as Figure 1 ,include,

[0051] Applying a pre-trained candidate symptom prediction model to obtain candidate symptoms according to the user's input questions; wherein the candidate symptom prediction model is embedded in the ChatGPT consultation model;

[0052] The ChatGPT consultation model is also used to determine a question and answer list based on candidate symptoms;

[0053] Interact with users based on the question and answer list to determine the actual symptoms;

[0054] And output acupuncture points according to actual symptoms.

[0055] In one embodiment, determining a question-answer list based on candidate symptoms comprises:

[0056] Constructing a symptom database and a question database, wherein the symptom database stores a mapping relationship between symptoms and symptoms, and the question database stores questions corresponding to the symptoms and candidate answers corresponding to the questions;

[0057] Determine the corresponding symptoms according to the candidate symptoms based on the symptom library, and obtain the corresponding questions and the candidate answers corresponding to the questions according to the symptoms based on the question library;

[0058] The question and the corresponding candidate answer form a question-answer pair, and a question-answer list is generated after removing duplicates.

[0059] Furthermore, the user is interacted with based on the question and answer list. During the interaction, the question and answer list is mapped into 13 categories, including: cold and heat, sweat, pain, taste, ears, eyes, chest, flank, stomach and abdomen, sleep, thirst and drinking water, appetite and food intake, and defecation; and the order of each category of questions and answers is determined based on the corresponding content of the "Ten Questions Song". In the actual diagnosis process, the doctor obtains diagnostic information such as symptoms and medical history through dialogue with the patient, and then establishes a mapping between symptoms and syndromes.

[0060] At the same time, during the interaction process, the user's symptoms are judged, including:

[0061] If the symptoms are judged to be mild, provide guidance to alleviate the user's symptoms;

[0062] When it is judged as an emergency symptom, an emergency plan is output and the nearest hospital is selected for treatment.

[0063] Preferably, the step of determining the actual symptoms, such as Figure 2 As shown, including:

[0064] S1. According to the question list, confirm with the user whether the symptom indicated by the question exists. If so, update the existing symptom and jump to step S2. If not, jump directly to step S2.

[0065] S2. Determine whether all the questions in the question list have been traversed. If not, return to step S1 and ask the next question. If so, determine the actual symptom based on the current symptoms, including the current symptoms determined based on T questions and answers, reversely inferring matching symptoms based on the symptom library, and taking the symptom with the highest matching degree as the actual symptom.

[0066] According to an embodiment of the present invention, the question-answer database is defined as D = {r u |u∈U}, the candidate acupuncture syndrome set is defined as S={s1,s2,…,s d}, s represents a single candidate syndrome, and d represents the number of all candidate syndromes. For a single patient's consultation answer u∈U, the diagnosis conclusion of the tth consultation (candidate acupuncture syndrome), where U is the set of patient consultation answers, and T is the number of consultation inquiries of patient u. u =(m1,m2,…,m T ) is the question sequence of acupuncture consultation for patient u, m T Represents the patient's answer to the Tth consultation.

[0067] Given an acupuncture medical case question-answering database D, patient u and the patient's multiple consultation diagnosis conclusions r u (candidate acupuncture syndromes of the previous t visits), comprehensively predict the diagnostic conclusions of multiple acupuncture consultation dialogues (candidate acupuncture syndromes for the t+1th time), and finally select the best acupuncture prescription from the question-and-answer system and recommend it to the doctor.

[0068] In an exemplary embodiment, a candidate symptom identification model is constructed to achieve the effect of determining actual symptoms based on candidate symptoms. When applied, it is only necessary to make the ChatGPT consultation model directly call the candidate symptom identification model.

[0069] In one embodiment, acupuncture literature and acupuncture clinical cases are collected to construct a data set, and the pre-trained ChatGPT consultation model is fine-tuned to improve the accuracy of outputting acupuncture points based on actual symptoms.

[0070] In order to further optimize the above technical solution, the candidate symptom prediction model is optimized according to the actual symptoms; and a scoring model is constructed to score the acupuncture points output by the ChatGPT consultation model, and the ChatGPT consultation model is optimized according to the scoring results.

[0071] Among them, the scoring model needs to construct a question-scoring training data set for pre-training. Figure 3 As shown; the data set includes a question library and corresponding question scores, and the question library transforms existing input questions through a genetic algorithm: synonym replacement, antonym replacement, sentence segmentation (deletion), etc., to generate new questions (child individuals) based on existing questions (parent individuals) in the database.

[0072] Preferably, 50 modern medical records are used to construct a scoring model training data set according to the specifications of natural language processing.

[0073] Furthermore, the ChatGPT consultation model is optimized according to the scoring results, including optimizing the automatic acupuncture point recognition model based on reinforcement learning according to the scoring results and the error of the preset scoring constant. The reward agent model expression of the optimization process is:

[0074]

[0075] In the formula, s t represents the transition state at time t; s t+1 represents the transition state at time t+1; p represents the preset scoring constant, ε t+1 Indicates the error between the scoring result at time t+1 and the preset scoring constant, ε t Indicates the error between the scoring result at time t and the preset scoring constant.

[0076] The coordinate error expression is:

[0077] ε=||x pre -x target ||

[0078] Among them, x pre Indicates the scoring result; x targetRepresents the preset scoring constant.

[0079] In this embodiment, the potential-based reward shaping term p can increase the reward signal and guide the training process without misleading the automatic acupuncture point recognition model to learn suboptimal policies. In the deep reinforcement learning problem, if ε t+1 If it is less than p, the manipulation is considered successful, a reward of +1 is obtained, and the iteration is terminated; otherwise, the reward is -1, and the iterative optimization continues.

[0080] In this embodiment, the reinforcement learning process is as follows: Figure 4 ,include,

[0081] Initialize two neural networks, one as the evaluation network and the other as the action network;

[0082] The evaluation network receives the current state of the environment (the question input by the user) and outputs the predicted value of each possible action (acupuncture position);

[0083] The action network is used to select an action based on the value information provided by the evaluation network, and apply it to the current state of the environment to obtain a new state and actual acupoint coordinates; the reward is calculated by calculating the error between the scoring result and the preset scoring constant, and the reward signal is used to update the evaluation network, and iterate continuously until the evaluation network can obtain the scoring result with a small enough error, or reaches the preset number of training times. In this way, the various parts of the reinforcement learning system work closely together to learn how to effectively identify acupuncture points.

[0084] In this embodiment, the temporal difference method is further used to learn and update the evaluation network, that is, by comparing the state and value function of two consecutive time points to estimate the change of the value function, the difference between the predicted value and the actual value (TD error) is reduced. At the same time, based on the Q value output by the evaluation network, the particle swarm optimization algorithm is used to update the speed and position of the particles to find the optimal solution of the action network; and finally, specific action suggestions are generated according to the updated action network, and the actions proposed by the action network are executed.

[0085] In this application, reinforcement learning is divided into two parts, namely forward signal calculation and particle swarm reinforcement calculation. In the forward signal calculation, the evaluation network is predicted by using the time difference prediction method, while the action network is optimized by the particle swarm optimization algorithm, so that the internal reinforcement signal is used as the fitness function of the action network;

[0086] The evaluation network provides more effective internal reinforcement signals to the action network, enabling it to take more appropriate actions. The internal reinforcement signals enable the action network and the evaluation network to learn at each step without having to wait for the arrival of external reinforcement signals, thereby greatly accelerating the learning of the two networks.

[0087] Furthermore, the evaluation network models the environment to make multi-step predictions of external reinforcement signals.

[0088] Embodiment 2

[0089] This embodiment discloses an intelligent acupuncture consultation system based on continuous dialogue, which includes a client, a ChatGPT consultation model is built in the client, and a pre-trained candidate symptom prediction model is embedded in the ChatGPT consultation model.

[0090] Preferably, the system also includes a server, which stores a symptom library and a question library. The symptom library stores the mapping relationship between symptoms and symptoms, and the question library stores questions corresponding to the symptoms and candidate answers corresponding to the questions.

[0091] In one embodiment, the candidate symptom prediction model obtains candidate symptoms based on a user's input question;

[0092] The ChatGPT consultation model is used to upload candidate symptoms to the server, which is used to determine the corresponding symptoms according to the candidate symptoms based on the symptom library, and obtain corresponding questions and candidate answers corresponding to the questions based on the symptoms based on the question library;

[0093] and forming question-answer pairs with the questions and corresponding candidate answers, and generating a question-answer list after removing duplicates;

[0094] Then, we interact with the user based on the question and answer list to determine the actual symptoms;

[0095] And output acupuncture points according to actual symptoms.

[0096] In an exemplary embodiment, Figure 5 Shown; including:

[0097] 1) Input candidate syndromes to be identified to the client, and the client transmits the candidate syndromes to the server and requests a list of questions. After receiving the request, the server obtains the symptoms of each candidate syndrome from the syndrome database of acupuncture consultation, obtains the corresponding questions based on the symptoms, and then retrieves the candidate answers to the questions from the question database. Finally, the questions and their candidate answers are organized into a question list in order, and duplicate questions are removed to form a question and answer list, which is returned to the client;

[0098] 2) After receiving the question list, the client interacts with the patient. Perform the following operations on each item in the question and answer list in order to determine the patient's symptoms: Give the patient a question and ask the patient to choose the option that suits his or her situation from the candidate answers corresponding to the question. Then the client transmits the candidate syndrome and the patient's symptoms to the server and requests syndrome identification;

[0099] 3) After receiving the request, the server completes the acupuncture syndrome identification. The server calls the syndrome identification model interface to identify candidate syndromes: query the symptom set of each candidate syndrome from the syndrome library of acupuncture consultation, and match it with the patient's symptoms. Candidate syndromes that meet the matching criteria are confirmed syndromes, and they are arranged in descending order of matching degree. The one with the highest matching degree is the main syndrome and the corresponding acupuncture point is given as the answer. The server returns the confirmed syndrome to the client;

[0100] 4) Use the returned syndromes to train the pre-trained consultation model based on ChatGPT, and then fine-tune the parameters of the entire acupuncture consultation model to obtain better answers to acupuncture syndromes.

[0101] The present invention provides an innovative intelligent diagnosis method and system for the field of acupuncture, which will help promote the precision and intelligent development of acupuncture treatment and improve the application value of acupuncture in primary medical care.

[0102] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0103] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An intelligent acupuncture consultation method based on continuous dialogue, characterized in that: Applying a pre-trained candidate symptom prediction model to obtain candidate symptoms according to the user's input questions; wherein the candidate symptom prediction model is embedded in the ChatGPT consultation model; The ChatGPT consultation model is also used to determine a question and answer list based on candidate symptoms; Interact with users based on the question and answer list to determine the actual symptoms; And output acupuncture points according to actual symptoms.

2. The intelligent acupuncture consultation method based on continuous dialogue according to claim 1, characterized in that: Determine a list of questions and answers based on candidate symptoms, the steps include: Constructing a symptom database and a question database, wherein the symptom database stores a mapping relationship between symptoms and symptoms, and the question database stores questions corresponding to the symptoms and candidate answers corresponding to the questions; Determine the corresponding symptoms according to the candidate symptoms based on the symptom library, and obtain the corresponding questions and the candidate answers corresponding to the questions according to the symptoms based on the question library; The question and the corresponding candidate answer form a question-answer pair, and a question-answer list is generated after removing duplicates.

3. The intelligent acupuncture consultation method based on continuous dialogue according to claim 2, characterized in that: When interacting with users based on the question and answer list, judge the user's symptoms, including: If the symptoms are judged to be mild, provide guidance to alleviate the user's symptoms; When it is judged as an emergency symptom, an emergency plan is output and the nearest hospital is selected for treatment.

4. The intelligent acupuncture consultation method based on continuous dialogue according to claim 2, characterized in that: Steps to determine the actual symptoms include: S1. According to the question list, confirm with the user whether the symptom indicated by the question exists. If so, update the existing symptom and jump to step S2. If not, jump directly to step S2. S2. Determine whether all the questions in the question list have been traversed. If not, return to step S1 and ask the next question. If so, determine the actual symptom based on the current symptom, including inferring the matching symptom based on the symptom library according to the current symptom, and taking the symptom with the highest matching degree as the actual symptom.

5. The intelligent acupuncture consultation method based on continuous dialogue according to claim 1, characterized in that: When interacting with users based on the question and answer list, the question and answer list is mapped into 13 categories, including: cold and heat, sweat, pain, taste, ears, eyes, chest, flanks, stomach and abdomen, sleep, thirst and drinking water, appetite and food intake, and defecation; and the order of questions and answers is determined based on the Ten Questions Song.

6. The intelligent acupuncture consultation method based on continuous dialogue according to claim 1, characterized in that: Acupuncture literature and clinical cases were collected to construct a dataset, and the pre-trained ChatGPT consultation model was fine-tuned to output acupuncture points based on actual symptoms.

7. The intelligent acupuncture consultation method based on continuous dialogue according to claim 1, characterized in that: Optimize candidate symptom prediction models based on actual symptoms; And construct a scoring model for scoring the acupuncture points output by the ChatGPT consultation model, and optimize the ChatGPT consultation model according to the scoring results.

8. The intelligent acupuncture consultation method based on continuous dialogue according to claim 7, characterized in that: The ChatGPT consultation model is optimized according to the scoring results, including optimizing the automatic acupuncture point recognition model based on reinforcement learning according to the error between the scoring results and the preset scoring constant. The reward agent model expression of the optimization process is: In the formula, s t represents the transition state at time t; s t+1 represents the transition state at time t+1; p represents the preset scoring constant, ε t+1 Indicates the error between the scoring result at time t+1 and the preset scoring constant, ε t Indicates the error between the scoring result at time t and the preset scoring constant.

9. An intelligent acupuncture consultation system based on continuous dialogue, characterized in that: The intelligent acupuncture consultation method based on continuous dialogue described in any one of claims 1 to 8 is applied, comprising a client, wherein the client has a built-in ChatGPT consultation model, wherein the ChatGPT consultation model has a built-in pre-trained candidate symptom prediction model, The candidate symptom prediction model obtains candidate symptoms according to the user's input questions; The ChatGPT consultation model is also used to determine a question and answer list based on candidate symptoms; Interact with users based on the question and answer list to determine the actual symptoms; And output acupuncture points according to actual symptoms.

10. The intelligent acupuncture consultation system based on continuous dialogue according to claim 9, characterized in that: The system further includes a server, wherein the server stores a symptom database and a question database, wherein the symptom database stores a mapping relationship between symptoms and symptoms, and the question database stores questions corresponding to the symptoms and candidate answers corresponding to the questions; The server is used to determine the corresponding symptoms according to the candidate symptoms based on the symptom library, and obtain the corresponding questions and the candidate answers corresponding to the questions according to the symptoms based on the question library; The question and the corresponding candidate answer are formed into question-answer pairs, and a question-answer list is generated after removing duplicates.

Images