A doctor-patient communication simulation training method and system based on an emotion recognition model

Abstract

The application provides a doctor-patient communication simulation training method and system based on an emotion recognition model, relates to the technical field of machine learning, and the method is characterized in that: a patient simulation model is used to simulate a doctor-patient communication scene, and communication scene data is obtained; an emotion classification algorithm and a tone analysis algorithm are used to construct an emotion recognition model; the communication scene data is input into the emotion recognition model, and the self-adaptive loss function is used for training, so that the trained emotion recognition model is obtained; the trained emotion recognition model is used to analyze the simulated doctor-patient communication data, and emotion recognition results are obtained, so that the simulation training of the doctor-patient communication is completed. The application solves the problem of communication simulation training under different patient personalities and emotional states.

Description

Technical Field

[0001] This specification relates to the field of machine learning technology, and in particular to a doctor-patient communication simulation training method and system based on an emotion recognition model. Background Art

[0002] Doctor-patient communication is a crucial aspect of healthcare. For medical students entering clinical practice or inexperienced clinicians, effective patient communication is a skill that urgently needs improvement. However, traditional doctor-patient communication training relies primarily on real-life case simulations and instructor guidance. This has limitations in communication contexts and significant individual differences, making it difficult to systematically and standardizedly train medical students. Summary of the Invention

[0003] In response to the above-mentioned deficiencies in the prior art, the present invention provides a doctor-patient communication simulation training method and system based on an emotion recognition model, which solves the problem of communication simulation training under different patient personalities and emotional states.

[0004] In order to achieve the above-mentioned purpose, the technical solution adopted by the present invention is: a doctor-patient communication simulation training method based on an emotion recognition model, comprising:

[0005] S1: Use the patient simulation model to simulate the doctor-patient communication scenario and obtain communication scenario data;

[0006] S2: Use emotion classification algorithm and intonation analysis algorithm to build emotion recognition model;

[0007] S3: inputting the communication scenario data into the emotion recognition model, and training the model using an adaptive loss function to obtain a trained emotion recognition model;

[0008] S4: Analyze the simulated doctor-patient communication data using the trained emotion recognition model to obtain emotion recognition results, thereby completing the simulated training of doctor-patient communication.

[0009] The beneficial effects of the present application are: simulating a doctor-patient communication scene by using a patient simulation model, obtaining communication scene data, analyzing the simulated doctor-patient communication data by using an emotion recognition model, and obtaining emotion recognition results. (1) The emotion fluctuation information in different rounds of the conversation is automatically extracted by the deep learning model, and is combined with the emotion label to construct a training data set with more accurate emotion annotation and time sequence; (2) By fusing voice and visual data (such as facial expressions) with conversation text data, more dimensional information is provided for emotion recognition, so that more accurate and robust emotion classification data is obtained; (3) Through the collection of personalized emotion data, the model can be customized for different patients, improving the recognition accuracy of the model for individual emotional responses of patients; (4) When the data flows through multiple neural network layers, the features of different modalities (such as emotional words, voice tone, and facial expression features) enter their respective network layers for deep mining. Finally, all features are fused through a cross-modal attention mechanism to ensure that the contribution of each modality to emotion classification is fully reflected; (5) Using an adaptive attention mechanism, the weights of emotional words, tone changes, or facial expressions in the conversation can be automatically adjusted according to the emotional fluctuations of the patient, so that when angry or anxious emotions appear, the model can give more weight to the corresponding emotional signals, thereby improving the accuracy of emotion recognition; (6) The multi-layer emotion reasoning mechanism helps the model better distinguish complex emotions (such as the subtle differences between anxiety and anger), improving the delicacy and accuracy of emotion recognition.

[0010] Further, the patient simulation model comprises:

[0011] An input layer for pre-processing patient basic data, personality type data, emotion parameters, and communication scene information to obtain input data;

[0012] A hidden layer for performing nonlinear transformation on the input data to obtain personality characteristic data;

[0013] An output layer for simulating a doctor-patient communication scene based on the personality characteristic data to obtain communication scene data; wherein the communication scene data includes communication scene text data, communication scene voice data, and communication scene image data.

[0014] In this way, (1) the combination of personalized patient models and multiple emotional states allows students to conduct doctor-patient communication training in various complex situations and develop their ability to cope with different patient personalities and emotions. This optimization enables medical students to train in a more realistic and complex environment, improving their clinical adaptability and communication skills; (2) real-time adjustment of the communication situation during interaction makes the simulation scenario more adaptive and diverse, thereby providing more challenging training scenarios and improving the flexibility and communication skills of students.

[0015] Further, the emotion recognition model comprises:

[0016] A text processing module for pre-processing the communication scenario text data to obtain denoised text data;

[0017] An emotion feature extraction module for analyzing the emotional words of the denoised text data to obtain emotion features;

[0018] An emotion classification module for classifying the emotion features based on the communication scenario image data to obtain an emotion classification result;

[0019] A tone analysis module for tone analysis of the communication scenario voice data to obtain a tone analysis result; wherein the tone analysis result comprises a tone of voice emotional intensity and a language emotional response score;

[0020] A score calculation module for using a multi-layer emotion reasoning and classifier to integrate the emotion classification result and the tone analysis result to obtain an emotion recognition result.

[0021] Further, the analysis of the emotional words of the denoised text data to obtain emotion features comprises:

[0022] Using an emotional dictionary method to identify emotional words in the denoised text data and mapping them to corresponding emotional categories to obtain emotional text data;

[0023] Performing intensity calculation on the emotional text data to obtain emotion features:

[0024]

[0025] Wherein, Sentiment Score represents the emotional intensity score, i.e. emotional text data, Strength represents the emotional intensity of the words, and w i represents the emotional words in the denoised text data.

[0026] Further, the expression of the adaptive loss function is:

[0027]

[0028] wherein L(0) represents the adaptive loss function result, a i represents the dynamic weight of each emotion category sample, L1 represents the classification loss function, represents the label, y i represents the prediction result, b i represents the regularization item weight, Reg represents the regularization item, 0 represents the parameter set of the model, and N represents the number of emotion category samples.

[0029] Further, the expression of the emotion recognition result is:

[0030]

[0031] wherein Emotion Score represents the emotion recognition result, TextSentiment Score represents the text sentiment score, Tone Score represents the tone sentiment intensity, a, b and g respectively represent the weight coefficients of the corresponding data, and Speech RateScoreEmotionScore represents the language emotion reaction intensity.

[0032] Further, the S3 comprises:

[0033] The emotion recognition model is preliminarily trained by using a large-scale general data set, and an initial emotion recognition model is obtained.

[0034] The communication scenario data is input into the initial emotion recognition model, and the trained emotion recognition model is obtained by using an adaptive loss function.

[0035] A doctor-patient communication simulation training system based on an emotion recognition model comprises:

[0036] An acquisition module is configured to simulate a doctor-patient communication scene by using a patient simulation model, and obtain communication scenario data.

[0037] A model construction module is configured to construct an emotion recognition model by using an emotion classification algorithm and a tone analysis algorithm.

[0038] A training module is configured to input the communication scenario data into the emotion recognition model, and train the emotion recognition model by using an adaptive loss function, so as to obtain a trained emotion recognition model.

[0039] An analysis module is configured to analyze simulated doctor-patient communication data by using the trained emotion recognition model, obtain an emotion recognition result, and complete simulation training of the doctor-patient communication. BRIEF DESCRIPTION OF DRAWINGS

[0040] The present specification will be further illustrated in the manner of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not restrictive, and in these embodiments, the same numbers represent the same structures, in which:

[0041] Figure 1 is a module schematic diagram of a doctor-patient communication simulation training system based on an emotion recognition model according to some embodiments of the present specification;

[0042] Figure 2 is an exemplary flowchart of a doctor-patient communication simulation training method based on an emotion recognition model according to some embodiments of the present specification. DETAILED DESCRIPTION

[0043] The specific embodiments of the present application are described below to facilitate the understanding of the present application for those skilled in the art, but it should be clear that the present application is not limited to the scope of the specific embodiments. For those skilled in the art, it is obvious that various changes are within the spirit and scope of the present application defined and determined by the appended claims, and all the applications utilizing the concept of the present application are within the scope of protection.

[0044] Embodiment One

[0045] Figure 1 is a module schematic diagram of a doctor-patient communication simulation training system based on an emotion recognition model according to some embodiments of the present specification.

[0046] In some embodiments, the doctor-patient communication simulation training system based on an emotion recognition model can include an acquisition module, a model construction module, a training module, and an analysis module.

[0047] The acquisition module is configured to simulate a doctor-patient communication scenario using a patient simulation model to obtain communication scenario data.

[0048] The model construction module is configured to construct an emotion recognition model using an emotion classification algorithm and a tone analysis algorithm.

[0049] The training module is configured to input the communication scenario data into the emotion recognition model and train the emotion recognition model using an adaptive loss function to obtain a trained emotion recognition model.

[0050] The analysis module is configured to analyze simulated doctor-patient communication data using the trained emotion recognition model to obtain emotion recognition results and complete the simulation training of doctor-patient communication.

[0051] In some embodiments, the doctor-patient communication simulation training system based on an emotion recognition model can be used to perform a doctor-patient communication simulation training method based on an emotion recognition model, including: S1: simulating a doctor-patient communication scene by using a patient simulation model to obtain communication scene data; S2: constructing an emotion recognition model by using an emotion classification algorithm and a tone analysis algorithm; S3: inputting the communication scene data into the emotion recognition model, training by using an adaptive loss function, and obtaining a trained emotion recognition model; and S4: analyzing simulated doctor-patient communication data by using the trained emotion recognition model to obtain an emotion recognition result, and completing the simulation training of doctor-patient communication.

[0052] In some embodiments of the present specification, the processor performs a doctor-patient communication simulation training method based on an emotion recognition model by using a doctor-patient communication simulation training system based on an emotion recognition model. In this way, (1) the emotion fluctuation information of different rounds in the dialogue can be automatically extracted by the deep learning model, combined with the emotion label, and a training data set with more accurate emotion labeling and time sequence is constructed; (2) by fusing voice and visual data (such as facial expressions) with dialogue text data, more dimensional information is provided for emotion recognition, so that more accurate and robust emotion classification data is obtained; (3) through the collection of personalized emotion data, the model can be customized for different patients, improving the recognition accuracy of the model for individual emotional responses of patients; (4) when the data flows through multiple neural network layers, the features of different modalities (such as emotion words, voice tone, and facial expression features) enter their respective network layers for deep mining. Finally, all features are fused through a cross-modal attention mechanism to ensure that the contribution of each modality to emotion classification is fully reflected; (5) by using an adaptive attention mechanism, the weights of emotion words, tone changes, or facial expressions in the dialogue can be automatically adjusted according to the emotional fluctuations of the patient, so that when an angry emotion or an anxious emotion appears, the model can give more weight to the corresponding emotional signals, thereby improving the accuracy of emotion recognition; (6) the multi-layer emotion reasoning mechanism helps the model to better distinguish complex emotions (such as the subtle difference between anxiety and anger), improving the delicacy and accuracy of emotion recognition.

[0053] Embodiment Two

[0054] Figure 2 is an exemplary flowchart of a doctor-patient communication simulation training method based on an emotion recognition model according to some embodiments of the present specification. As shown in Figure 2 , the flow includes the following steps. In some embodiments, the flow can be performed by a processor.

[0055] S1: Simulate a doctor-patient communication scene by using a patient simulation model to obtain communication scene data.

[0056] The patient simulation model is a neural network model for simulating patient conditions in a doctor-patient communication scenario.

[0057] In some embodiments, the patient simulation model is used to simulate patient conditions for constructing a doctor-patient communication scenario. The type of simulated doctor-patient communication scenario can be various. For example, the type of simulated doctor-patient communication scenario can include a deep neural network model.

[0058] In some embodiments, the input of the patient simulation model can include patient basic data, personality type data, emotional parameters, and communication scenario information, and the output of the patient simulation model can be communication scenario data.

[0059] In some embodiments, the structure of the patient simulation model is as follows:

[0060] The patient simulation model includes an input layer, a hidden layer, and an output layer. The output of the input layer is used as the input of the hidden layer, the output of the hidden layer is used as the input of the output layer, and the output of the output layer is used as the final output of the patient simulation model.

[0061] The input layer is used to preprocess the patient basic data, personality type data, emotional parameters, and communication scenario information to obtain input data. The input of the input layer can include patient basic data, personality type data, emotional parameters, and communication scenario information, and the output can include input data.

[0062] The patient basic data is the basic information of the patient character. For example, the patient basic data can include basic information such as age, gender, health status, etc.

[0063] The personality type data is a binary vector representation of a personality type. For example, the personality type data can include binary vector representations of ISTJ (introverted, sensing, thinking, judging), ISFJ (introverted, sensing, feeling, judging), INFJ (introverted, intuition, feeling, judging), INTJ (introverted, intuition, thinking, judging), ISTP (introverted, sensing, thinking, perception), ISFP (introverted, sensing, feeling, perception), INFP (introverted, intuition, feeling, perception), INTP (introverted, intuition, thinking, perception), ESTP (extroverted, sensing, thinking, perception), ESFP (extroverted, sensing, feeling, perception), ENFP (extroverted, intuition, feeling, perception), ENTP (extroverted, intuition, thinking, perception), ESTJ (extroverted, sensing, thinking, judgment), ESFJ (extroverted, sensing, feeling, judgment), ENFJ (extroverted, intuition, feeling, judgment), and ENFJ (extroverted, intuition, thinking, judgment) personality types.

[0064] In some embodiments, the processor can obtain the personality type data by one-hot encoding the personality type.

[0065] The emotion parameter is a parameter reflecting an emotional state. For example, the emotion parameter can include anger, mildness, anxiety, etc.

[0066] In some embodiments, the processor can obtain the emotion parameter by associating the external stimulus and one-hot encoding.

[0067] The communication scenario information is the clinic scenario information. For example, the communication scenario information can include initial diagnosis, treatment, follow-up, etc.

[0068] In some embodiments, the processor can obtain the communication scenario information by one-hot encoding.

[0069] In some embodiments, the processor can obtain the preset patient basic data, personality type data, emotion parameter and communication scenario information based on the external input device and the interface circuit.

[0070] The input data is the normalized patient basic data, personality type data, emotion parameter and communication scenario information.

[0071] In some embodiments, the processor can normalize the patient basic data, personality type data, emotion parameter and communication scenario information to obtain the input data.

[0072] The hidden layer is used for nonlinear transformation of the input data to obtain the character feature data. The input of the hidden layer can include the input data, and the output can include the character feature data.

[0073] The character feature data is data reflecting the character and emotional state of the patient.

[0074] In some embodiments, the processor can use the neurons of the hidden layer to pass information by weighting and activation function, extract the complex pattern of the character features and emotional reactions of the patient, form an abstract representation of the patient's communication reaction after passing, and increase the complexity of the emotion and communication strategy layer by layer to obtain the character feature data.

[0075] The output layer is used for simulating a doctor-patient communication scenario based on the character feature data to obtain the communication scenario data. The input of the output layer can include the character feature data, and the output can include the communication scenario data.

[0076] In some embodiments, the patient simulation model can be trained by a plurality of labeled first training samples. For example, the plurality of labeled first training samples can be input into an initial patient simulation model, a first loss function can be constructed based on the labels and the results of the initial patient simulation model, and the parameters of the initial patient simulation model can be iteratively updated based on the first loss function by using the Adam optimizer and the Dropout technique to perform small batch gradient descent or other methods. When a preset condition is met, the model training is completed, and a trained patient simulation model is obtained. The preset condition can be, for example, convergence of the loss function, or the number of iterations reaching a threshold.

[0077] The first loss function can include a cross-entropy loss function and a mean square error.

[0078] In some embodiments, the first training samples include historical patient basic data, corresponding personality type data, corresponding emotional parameters, and historical communication scenario information. The labels can be corresponding communication scenario data. The labels can be manually annotated.

[0079] In this way, (1) the combination of personalized patient models and multiple emotional states allows trainees to conduct medical communication training in various complex situations and develop their ability to cope with different patient personalities and emotions. This optimization enables medical students to train in a more realistic and complex environment, improving their clinical adaptability and communication skills; (2) real-time adjustment of the communication scenario during the interaction makes the simulation scenario more adaptive and diverse, thereby providing more challenging training scenarios and improving the flexibility and communication skills of the trainees.

[0080] The communication scenario data is patient emotion-related data reflecting the simulated medical communication. For example, the communication scenario data can include communication scenario text data, communication scenario voice data, and communication scenario image data.

[0081] In some embodiments, the processor can utilize an adaptive attention mechanism to dynamically adjust the attention weight based on the importance of the features of the input data, giving more attention to the key parts of the emotion.

[0082] S2: Utilize an emotion classification algorithm and a tone analysis algorithm to construct an emotion recognition model.

[0083] The emotion recognition model is a neural network model for simulating and recognizing the simulated medical communication scenario.

[0084] In some embodiments, the emotion recognition model is used to simulate and recognize the simulated medical communication scenario.

[0085] In some embodiments, the input of the emotion recognition model can be the communication scenario data, and the output of the emotion recognition model can be the emotion recognition result.

[0086] In some embodiments, the structure of the sentiment recognition model is as follows:

[0087] The sentiment recognition model includes a text processing module, a sentiment feature extraction module, a sentiment classification module, a tone analysis module, and a score calculation module. The output of the text processing module is input to the sentiment feature extraction module, the output of the sentiment feature extraction module is input to the sentiment classification module, the output of the sentiment classification module and the output of the tone analysis module are input to the score calculation module, and the output of the score calculation module is the final output of the sentiment recognition model.

[0088] The text processing module is used to preprocess the communication scenario text data to obtain denoised text data. The input of the text processing module can include communication scenario text data, and the output can include denoised text data. The text processing module can be a convolutional neural network.

[0089] In some embodiments, the processor can remove noise, word segmentation, stemming, lemmatization, and stop word removal from the communication scenario text data to obtain denoised text data. For example, the processor can remove irrelevant words, symbols, punctuation marks, etc. from the text. For example, remove mood words such as "ah", "um", "er", etc.; split the dialogue text into individual words or sub-word units. Assuming the input text is T = "I am feeling upset" T = "I am feeling upset", it is converted to T' = {"I", "am", "feeling", "upset"} T' = {"I", "am", "feeling", "upset"} by word segmentation; reduce the word to the base form. For example, "running" is reduced to "run"; remove common words in the text that are not essential to sentiment analysis, such as "the", "a", "an", etc.

[0090] The sentiment feature extraction module is used to analyze the sentiment words of the denoised text data to obtain sentiment features. The input of the sentiment feature extraction module can include denoised text data, and the output can include sentiment features. The sentiment feature extraction module can be a convolutional neural network.

[0091] The sentiment feature is a feature that reflects the intensity of positive, negative, or neutral sentiment.

[0092] In some embodiments, the expression of the sentiment feature can be:

[0093]

[0094] where Sentiment Score represents the sentiment intensity score, i.e. the sentiment text data, Strength represents the intensity of the word sentiment, and w iThe sentiment words in the denoised text data are represented.

[0095] In some embodiments, the processor can identify the sentiment words in the denoised text data using a sentiment dictionary method, and map to corresponding sentiment categories to obtain sentiment text data; and perform intensity calculation on the sentiment text data to obtain sentiment features.

[0096] The sentiment text data is data reflecting sentiment category scores.

[0097] The emotion classification module is configured to classify the sentiment features based on the communication scene image data to obtain an emotion classification result. The input of the emotion classification module can include the communication scene image data and the sentiment features, and the output can include the emotion classification result. The emotion classification module can be a convolutional neural network.

[0098] The emotion classification result is a result reflecting the emotion type of the text. For example, the emotion analysis result can include a text sentiment score.

[0099] The text sentiment score is a sentiment score corresponding to the communication scene text data.

[0100] The tone analysis module is configured to perform tone analysis on the communication scene voice data to obtain a tone analysis result. The input of the tone analysis module can include the communication scene voice data, and the output can include the tone analysis result. The tone analysis module can be an LSTM network.

[0101] The tone analysis result is a result reflecting the emotional state of the communication scene voice data. For example, the tone analysis result can include a tone sentiment intensity and a language emotional response score.

[0102] The tone sentiment intensity is sentiment intensity information based on pitch frequency and tone amplitude analysis.

[0103] In some embodiments, the processor can perform pitch frequency and tone amplitude analysis on the communication scene voice data to obtain frequency distribution in the audio signal and volume change of the voice, and obtain the tone sentiment intensity.

[0104] The language emotional response score is a score obtained based on speech rate analysis. For example, a faster speech rate is usually associated with emotions such as anxiety, excitement, etc., while a slower speech rate may indicate sadness or fatigue.

[0105] The score calculation module is configured to use a multi-layer sentiment reasoning and classifier to integrate the emotion classification result and the tone analysis result to obtain an emotion recognition result. The input of the score calculation module can include the emotion classification result and the tone analysis result, and the output can include the emotion recognition result.

[0106] The emotion recognition result is a result of quantifying the sentiment score in the communication scene data.

[0107] In some embodiments, the expression of the emotion recognition result can be:

[0108]

[0109] wherein Emotion Score represents the emotion recognition result, TextSentiment Score represents the text sentiment score, Tone Score represents the tone sentiment intensity, a, β and γ respectively represent the weight coefficients of the corresponding data, Speech Rate Score represents the language emotion reaction intensity.

[0110] S3: inputting the communication scenario data into the emotion recognition model, training by using an adaptive loss function, and obtaining a trained emotion recognition model.

[0111] In some embodiments, the processor can implement S3 based on the following steps: preliminarily training the emotion recognition model by using a large-scale general data set, and obtaining an initial emotion recognition model; inputting the communication scenario data into the initial emotion recognition model, training by using an adaptive loss function, and obtaining a trained emotion recognition model.

[0112] In some embodiments, the process of training by using the adaptive loss function can include: inputting a plurality of second training samples with labels into the initial emotion recognition model, constructing an adaptive loss function by using the labels and the results of the initial emotion recognition model, and iteratively updating the parameters of the initial emotion recognition model based on the adaptive loss function by using gradient descent or other methods. When a preset condition is met, the model training is completed, and a trained emotion recognition model is obtained. The preset condition can be that the loss function converges, the number of iterations reaches a threshold, etc.

[0113] In some embodiments, the expression of the adaptive loss function can be:

[0114]

[0115] wherein L(θ) represents the adaptive loss function result, a i represents the dynamic weight of each emotion category sample, L1 represents the classification loss function, represents the label, y i represents the prediction result, β i represents the regularization item weight, Reg represents the regularization item, θ represents the parameter set of the model, and N represents the number of emotion category samples.

[0116] In some embodiments, the second training sample includes. The label can be. The label can be manually annotated.

[0117] S4: Analyzing the simulated doctor-patient communication data using the trained emotion recognition model to obtain emotion recognition results and complete the simulation training of doctor-patient communication.

[0118] In some embodiments, the processor can preprocess the communication scenario text data to obtain denoised text data, analyze the emotional words of the denoised text data to obtain emotional features, classify the emotional features based on the communication scenario image data to obtain emotion classification results, analyze the intonation of the communication scenario voice data to obtain intonation analysis results, use a multi-layer emotion reasoning and classifier to integrate the emotion classification results and the intonation analysis results to obtain emotion recognition results, and complete the simulation training of doctor-patient communication.

[0119] In some embodiments of the present specification, the processor simulates a doctor-patient communication scenario using a patient simulation model to obtain communication scenario data, and analyzes the simulated doctor-patient communication data using an emotion recognition model to obtain emotion recognition results. (1) The deep learning model automatically extracts emotion fluctuation information in different rounds of the conversation, and combines it with emotion labels to build a training data set with more accurate and time-series emotion annotations; (2) By fusing voice and visual data (such as facial expressions) with conversation text data, more dimensional information is provided for emotion recognition, resulting in more accurate and robust emotion classification data; (3) Through the collection of personalized emotion data, the model can be customized for different patients, improving the recognition accuracy of the model for individual emotional responses of patients; (4) When data flows through multiple neural network layers, features of different modalities (such as emotional words, voice intonation, and facial expression features) are respectively input into their own network layers for deep mining. Finally, all features are fused through a cross-modal attention mechanism to ensure that the contribution of each modality to emotion classification is fully reflected; (5) Using an adaptive attention mechanism, the weights of emotional words, tone changes, or facial expressions in the conversation can be automatically adjusted according to the emotional fluctuations of the patient, so that when angry or anxious emotions appear, the model can give more weight to the corresponding emotional signals, thereby improving the accuracy of emotion recognition; (6) The multi-layer emotion reasoning mechanism helps the model better distinguish complex emotions (such as the subtle differences between anxiety and anger), improving the delicacy and accuracy of emotion recognition.

Claims

1. A doctor-patient communication simulation training method based on an emotion recognition model, characterized in that: include: S1: Use the patient simulation model to simulate the doctor-patient communication scenario and obtain communication scenario data; The patient simulation model includes: The input layer is used to pre-process the patient's age, gender, health status, personality type data, emotional parameters and communication scenario information to obtain input data; A hidden layer, configured to perform a nonlinear transformation on the input data to obtain personality characteristic data; An output layer is used to simulate a doctor-patient communication scenario based on the personality characteristic data to obtain communication scenario data; wherein the communication scenario data includes communication scenario text data, communication scenario voice data, and communication scenario image data; S2: Build an emotion recognition model using the emotion classification algorithm and the intonation analysis algorithm; the emotion recognition model includes: A text processing module, configured to pre-process the communication scenario text data to obtain denoised text data; An emotional feature extraction module is used to analyze the emotional words in the denoised text data to obtain emotional features; An emotion classification module, configured to classify the emotion features based on the communication scenario image data to obtain an emotion classification result; A tone analysis module is used to perform tone analysis on the communication scenario voice data to obtain a tone analysis result; wherein the tone analysis result includes the emotional intensity of the tone and the language emotional response score; A score calculation module, configured to utilize multi-layer sentiment reasoning and classifiers to combine the sentiment classification results and the intonation analysis results to obtain a sentiment recognition result; S3: Input the communication scenario data into the emotion recognition model and train it using an adaptive loss function to obtain a trained emotion recognition model; the expression of the adaptive loss function is: ； in, Represents the result of the adaptive loss function, represents the dynamic weight of each emotion category sample, represents the classification loss function, Indicates a label, Represents the prediction result, represents the regularization term weight, represents the regularization term, represents the set of parameters of the model, Indicates the number of emotion category samples; S4: Analyze the simulated doctor-patient communication data using the trained emotion recognition model to obtain emotion recognition results, thereby completing the simulated training of doctor-patient communication.

2. The doctor-patient communication simulation training method based on the emotion recognition model according to claim 1 is characterized in that: The sentiment features obtained by analyzing the sentiment words of the denoised text data include: Using the sentiment dictionary method to identify sentiment words in the denoised text data, and mapping them to corresponding sentiment categories to obtain sentiment text data; Perform intensity calculation on the emotional text data to obtain emotional features: ； in, Represents the sentiment intensity score, i.e., sentiment text data, Indicates the emotional intensity of the word, Represents the sentiment words in the denoised text data.

3. The doctor-patient communication simulation training method based on the emotion recognition model according to claim 1 is characterized in that: The expression of the emotion recognition result is: ； in, represents the emotion recognition result, represents the sentiment score of the text, Indicates the emotional intensity of the tone. 、 and Represent the weight coefficients of the corresponding data, Indicates the intensity of emotional response of language.

4. The doctor-patient communication simulation training method based on the emotion recognition model according to claim 1 is characterized in that: The S3 includes: Preliminarily training the emotion recognition model using a large-scale general dataset to obtain an initial emotion recognition model; The communication scenario data is input into the initial emotion recognition model, and trained using an adaptive loss function to obtain a trained emotion recognition model.

5. A doctor-patient communication simulation training system based on an emotion recognition model, used to execute the doctor-patient communication simulation training method based on an emotion recognition model according to any one of claims 1 to 4, characterized in that: include: An acquisition module is used to simulate doctor-patient communication scenarios using a patient simulation model to obtain communication scenario data; The model building module is used to build an emotion recognition model using the emotion classification algorithm and the intonation analysis algorithm; A training module, configured to input the communication scenario data into the emotion recognition model and perform training using an adaptive loss function to obtain a trained emotion recognition model; The analysis module is used to use the trained emotion recognition model to analyze the simulated doctor-patient communication data, obtain emotion recognition results, and complete the simulation training of doctor-patient communication.

Images