Dermatology department resident physician standardized training teaching process management method

Abstract

The invention relates to the technical field of teaching management, in particular to a dermatology department resident physician standardized training teaching process management method, which comprises the steps of obtaining a historical case library, randomly generating case cues and teaching questions according to the historical case library, pre-training a preset large language model according to the case cues, and generating a teaching question according to the pre-trained large language model. A virtual patient is obtained; and the target student performs diagnosis and treatment inquiry on the virtual patient according to the teaching test question to obtain student inquiry voice, and performs voice recognition on the student inquiry voice to obtain an inquiry text. By dynamically adjusting the difficulty of the teaching test questions and the case cues, the teaching content can be customized according to the actual performance of the student, specifically, the weak links of the student can be recognized according to the inquiry voice, emotion recognition and the difference between the treatment text and the historical cases of the student, and the difficulty of the teaching content is adjusted accordingly, so that the teaching efficiency is improved. The personalized teaching adjustment is helpful for improving the learning efficiency of the students to the maximum extent.

Description

Technical Field

[0001] The present invention relates to the technical field of teaching management, and particularly relates to a teaching process management method for standardized training of dermatology residents. Background Art

[0002] Traditional teaching methods often adopt a unified teaching mode and test questions, and cannot be dynamically adjusted according to the actual performance of students. This means that each student studies under the same teaching content and difficulty, and cannot adjust the teaching content according to personal weaknesses or progress. Therefore, some students may find the content too simple, while others may find it too difficult, resulting in uneven learning effects. Moreover, in traditional methods, students may only rely on real patients for clinical training, which not only requires a large number of patients to participate, but may also involve patient privacy and ethical issues. In contrast, virtual patients provide a way to conduct simulation training without real patients, greatly avoiding such limitations and risks. And in traditional teaching, the cultivation of doctor-patient communication skills is usually carried out through face-to-face role-playing or actual communication with patients, but this lacks systematic emotion recognition and feedback, and there is no special emotion analysis tool. It is difficult for students to get precise guidance from the perspective of emotional responses. Students may not be aware of their own emotional expression problems during the communication process. Therefore, the improvement of communication skills is relatively slow and the effect is not necessarily ideal. And traditional performance evaluations mainly rely on the scoring of written tests or practical operations, which are usually rather general and single, lacking detailed data analysis based on the specific performance of students. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to overcome the above-mentioned disadvantages of the prior art and provide a teaching process management method for standardized training of dermatology residents.

[0004] The technical solution adopted to solve the above technical problem is: A teaching process management method for standardized training of dermatology residents, comprising:

[0005] Obtain a historical case library, randomly generate case prompt words and teaching test questions according to the historical case library, and pre-train a preset large language model according to the case prompt words to obtain a virtual patient;

[0006] The target student conducts a diagnosis and treatment interview on the virtual patient according to the teaching test questions to obtain the student's interview voice, performs speech recognition on the student's interview voice to obtain an interview text, the virtual patient generates the symptom information of the patient according to the interview text, and the target student generates a student treatment text according to the symptom information;

[0007] Perform voice emotion recognition on the target trainee and the historical case based on the voice of the trainee's medical consultation and the historical case database to obtain the first emotion intensity of the target trainee and the second emotion intensity of the historical case, and determine the first teaching score of the target trainee according to the difference between the first emotion intensity and the second emotion intensity;

[0008] Determine the second teaching score of the target trainee according to the difference between the trainee's treatment text and the historical case;

[0009] Adjust the difficulty of the teaching questions and the case prompt words according to the first teaching score and the second teaching score.

[0010] Preferably, the historical case includes patient disease information and historical diagnosis and treatment information, wherein the patient disease information includes the patient's basic information, diseases and symptoms, the historical diagnosis and treatment information includes expert treatment text and expert medical consultation voice, the case prompt words include patient disease information, and the teaching questions include the patient's basic information and diseases.

[0011] Preferably, randomly generate case prompt words according to the historical case database, including:

[0012] Define a semantic graph, wherein the semantic graph includes a plurality of triples, and the triple includes a disease node, a symptom node and a relationship edge, and the expression of the triple is as follows:

[0013] G i =(d i ,r ij ,s j );

[0014] Wherein, G i represents the i-th triple in the semantic graph, d i represents the i-th disease node in the semantic graph, s j represents the i-th symptom node in the semantic graph, and r ij represents the relationship between the i-th disease node in the semantic graph and the i-th symptom node in the semantic graph;

[0015] Aggregate information of the nodes and adjacent nodes in the semantic graph according to the graph attention network, and adjust the contribution degrees of different nodes according to the attention mechanism, and generate the final case prompt words according to the contribution degrees of different nodes, wherein the generation formula of the case prompt words is as follows:

[0016]

[0017] Wherein, Prompt kDenote the case prompt words generated from the k-th historical case in the historical case library, and GAT(d k ) represents the weighted feature representation based on disease nodes generated by the graph attention network, and TF-IDF(c k ) represents the symptom node-related vocabulary features in the k-th historical case in the historical case library extracted by the TF-IDF algorithm.

[0018] Preferably, perform speech emotion recognition on the target trainee and the historical case according to the trainee's inquiry speech and the historical case library to obtain the first emotion intensity of the target trainee and the second emotion intensity of the historical case, including:

[0019] Extract the first Mel-frequency cepstral coefficient features of the trainee's inquiry speech;

[0020] According to the mapping of quantum states, convert the first Mel-frequency cepstral coefficient features into the first state vector in the quantum space, where the first state vector is the first emotion intensity of the target trainee;

[0021] Extract the second Mel-frequency cepstral coefficient features of the expert inquiry speech in the historical case;

[0022] According to the mapping of quantum states, convert the second Mel-frequency cepstral coefficient features into the second state vector in the quantum space, where the second state vector is the second emotion intensity of the historical case.

[0023] Preferably, the calculation formula of the first state vector is as follows:

[0024]

[0025] where, |ψ audio > represents the first state vector, e i represents the basis vector of the first quantum state, n represents the total number of basis vectors of the first quantum state, α i represents the weight of the basis vector of the first quantum state, and α i = Softmax(W a · MFCC(A)), W a represents the first projection matrix, and MFCC(A) represents the first Mel-frequency cepstral coefficient features of the trainee's inquiry speech;

[0026] The calculation formula of the second state vector is as follows:

[0027]

[0028] where, |ψ exp > represents the second state vector, b jdenote the basis vectors of the second quantum state, m denotes the total number of basis vectors of the second quantum state, θ j denote the weights of the basis vectors of the second quantum state, and θ j = Softmax(W b ·MFCC(B)), W b denotes the second projection matrix, and MFCC(B) denotes the second Mel Frequency Cepstral Coefficient feature of the voice of the expert consultation.

[0029] Preferably, the calculation formula of the first teaching achievement is as follows:

[0030]

[0031] wherein, L1 denotes the first teaching achievement, T denotes the number of time steps, σ denotes the Gaussian kernel width, |ψ audio,t > denotes the first state vector at the t-th time step, |ψ exp,t > denotes the second state vector at the t-th time step.

[0032] Preferably, determining the second teaching achievement of the target student according to the difference between the student treatment text and the historical case includes:

[0033] Define a treatment decision knowledge graph, wherein the treatment decision knowledge graph includes a node set and connection edges, wherein the node set includes symptom nodes, diagnosis nodes and treatment option nodes, and the connection edges are used to represent the relationships between the various nodes in the node set;

[0034] Extract the treatment path from the student treatment text according to the treatment decision knowledge graph to obtain the student treatment path;

[0035] Extract the treatment path from the expert treatment text in the historical case according to the treatment decision knowledge graph to obtain the expert treatment path;

[0036] Calculate the path difference between the student treatment path and the expert treatment path according to the minimum path edit distance to obtain the difference between the student treatment text and the historical case;

[0037] Calculate the case complexity adjustment coefficient of the student treatment text according to the treatment decision knowledge graph;

[0038] Calculate the second teaching achievement of the target student according to the case complexity adjustment coefficient of the student treatment text and the difference between the student treatment text and the historical case.

[0039] Preferably, the calculation formula of the path difference is as follows:

[0040]

[0041] Among them, Δ represents the path difference, |P audio | represents the total number of nodes in the trainee's treatment path, P audio represents the nodes in the trainee's treatment path, P exp represents the nodes in the expert's treatment path, and d(p, q) represents the path edit distance;

[0042] The calculation formula of the case complexity adjustment coefficient is as follows:

[0043]

[0044] Among them, Complex(c) represents the case complexity adjustment coefficient, N c represents the number of symptom nodes, τ c represents the number of diagnosis nodes, M c represents the number of treatment option nodes;

[0045] The calculation formula of the second teaching score is as follows:

[0046]

[0047] Among them, L2 represents the second teaching score, and β represents the preset weight coefficient.

[0048] Preferably, adjusting the difficulty of the teaching questions and the case prompts according to the first teaching score and the second teaching score includes:

[0049] Defining an objective function according to the preset target score, the first teaching score, and the second teaching score;

[0050] Solving the objective function to obtain the change amount of the difficulty parameter;

[0051] Adjusting the difficulty of the teaching questions and the case prompts according to the change amount of the difficulty parameter.

[0052] Preferably, the objective function is as follows:

[0053]

[0054] Among them, L target represents the preset objective function, γ represents the preset penalty coefficient, and Δθ represents the change amount of the difficulty parameter to be solved.

[0055] The beneficial effects of the present invention are as follows: (1) By dynamically adjusting the difficulty of teaching test questions and case prompt words, the teaching content can be customized according to the actual performance of the trainees. Specifically, based on the differences between the trainees' inquiry voices, emotion recognition, and treatment texts and historical cases, the weak links of the trainees can be identified, and the difficulty of the teaching content can be adjusted accordingly. This personalized teaching adjustment helps to maximize the learning efficiency and knowledge mastery level of the trainees. Moreover, by using virtual patients to simulate a real diagnosis and treatment environment, trainees can conduct clinical training without real patients. These virtual patients are generated based on a historical case database and generate corresponding symptom information according to the trainees' diagnosis and treatment inquiries, greatly enhancing the realism of teaching and the cultivation of clinical practice ability; (2) By analyzing the trainees' voice emotion recognition, it can be determined whether the trainees show appropriate emotional responses during the inquiry process. Emotion recognition not only helps to evaluate the professionalism and communication ability of the trainees, but also helps the trainees to be aware of their emotional expressions during the communication process, thereby improving the communication skills between doctors and patients. This emotion analysis can also help the instructors to provide better feedback to the trainees and improve their doctor-patient communication methods. Moreover, based on the trainees' scores obtained from voice recognition and emotion analysis, the teaching process can more accurately quantify the learning achievements of the trainees. This data-driven performance evaluation method is not only objective and accurate, but also can timely feedback the problems of the trainees during the diagnosis and treatment process, providing a scientific basis for subsequent teaching adjustments; (3) By randomly generating case prompt words and teaching test questions, the present invention can ensure that each trainee is exposed to a diverse range of case types, avoiding the monotony and limitations of the teaching content. This diverse training method helps trainees to widely accumulate clinical experience and improve their adaptability in actual work. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a schematic flowchart of the steps of the overall method in an embodiment proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0057] Embodiment 1, as Figure 1 shown, a method for managing the teaching process of standardized training for dermatology residents proposed by the present invention includes:

[0058] S1. Obtain a historical case database, randomly generate case prompt words and teaching test questions according to the historical case database, and pre-train a preset large language model according to the case prompt words to obtain virtual patients;

[0059] S2. The target trainee conducts a diagnosis and treatment inquiry on the virtual patient according to the teaching test questions to obtain the trainee's inquiry voice, perform voice recognition on the trainee's inquiry voice to obtain an inquiry text, the virtual patient generates the patient's symptom information according to the inquiry text, and the target trainee generates a trainee's treatment text according to the symptom information;

[0060] S3. Perform voice emotion recognition on the target trainee and historical cases based on the trainee's consultation voice and the historical case database to obtain the first emotion intensity of the target trainee and the second emotion intensity of the historical cases, and determine the first teaching score of the target trainee according to the difference between the first emotion intensity and the second emotion intensity;

[0061] S4. Determine the second teaching score of the target trainee according to the difference between the trainee's treatment text and the historical cases;

[0062] S5. Adjust the difficulty of the teaching questions and case prompts according to the first teaching score and the second teaching score.

[0063] In the present invention, the target trainee conducts a consultation on a virtual patient through the given teaching questions to generate the trainee's consultation voice; the virtual patient feedback is that the virtual patient generates corresponding symptom feedback according to the trainee's consultation text; the trainee generates corresponding treatment text according to the symptom information provided by the virtual patient; by comparing the difference between the emotion intensities of the trainee and the historical cases, it can be inferred whether the trainee's emotional response conforms to the case characteristics, thereby deriving the first teaching score of the trainee. Whether the trainee can calmly cope with the emotions of the patient may reflect their professional level and clinical ability; by comparing the treatment text generated by the trainee with the actual treatment methods of the historical cases, the second teaching score of the trainee is determined; according to the first teaching score (based on emotion analysis) and the second teaching score (based on the analysis of the difference in treatment text), the system will adjust the difficulty of the subsequent teaching questions and case prompts. Specifically, if the trainee performs well in emotional coping and treatment plan formulation, more challenging cases and questions can be provided; if the performance is poor, questions of moderate difficulty are provided to help the trainee gradually improve.

[0064] Embodiment 2. A method for managing the teaching process of standardized training for dermatology residents proposed by the present invention. Compared with Embodiment 1, this embodiment further includes: the historical cases include patient disease information and historical diagnosis and treatment information, where the patient disease information includes the patient's basic information, diseases, and symptoms, the historical diagnosis and treatment information includes expert treatment text and expert consultation voice, the case prompts include patient disease information, and the teaching questions include the patient's basic information and diseases.

[0065] In an optional embodiment, randomly generate case prompts according to the historical case database, including:

[0066] Define a semantic graph, where the semantic graph includes multiple triples, and a triple includes a disease node, a symptom node, and a relationship edge. The expression of the triple is as follows:

[0067] G i =(d i ,r ij ,sj );

[0068] Among them, G i represents the i-th triple in the semantic graph, d i represents the i-th disease node in the semantic graph, s j represents the i-th symptom node in the semantic graph, r ij represents the relationship between the i-th disease node in the semantic graph and the i-th symptom node in the semantic graph;

[0069] According to the graph attention network, information aggregation is performed on the nodes and adjacent nodes in the semantic graph, and the contribution degrees of different nodes are adjusted according to the attention mechanism. The final case prompt words are generated according to the contribution degrees of different nodes. Among them, the generation formula of the case prompt words is as follows:

[0070]

[0071] Among them, Prompt k represents the case prompt words generated from the k-th historical case in the historical case library, GAT(d k ) represents the weighted feature representation of the disease node generated by the graph attention network, TF-IDF(c k ) represents the symptom node-related vocabulary feature in the k-th historical case in the historical case library extracted by the TF-IDF algorithm.

[0072] It should be noted that the graph attention network is a kind of graph neural network. Through the attention mechanism, weighted aggregation is performed on the nodes in the graph, so as to generate a representation for each node. Here, GAT is used to aggregate the information of the disease node and its neighboring nodes (symptom nodes); the case prompt words are generated by combining the disease node features generated by the graph attention network and the symptom-related vocabulary features extracted by the TF-IDF algorithm; the TF-IDF algorithm can be used to extract the vocabulary features related to the symptom nodes in each historical case. These vocabulary features represent the descriptive words of the symptoms, and the frequency of symptom occurrence and its importance in the historical cases can be reflected through TF-IDF.

[0073] In an alternative embodiment, voice emotion recognition is performed on the target student and the historical cases according to the student's inquiry voice and the historical case library, so as to obtain the first emotion intensity of the target student and the second emotion intensity of the historical case, including:

[0074] Extract the first Mel frequency cepstral coefficient features of the student's inquiry voice;

[0075] According to the mapping of the quantum state, the first Mel frequency cepstral coefficient features are converted into the first state vector in the quantum space, where the first state vector is the first emotion intensity of the target student;

[0076] Extract the second Mel Frequency Cepstral Coefficient (MFCC) features of the expert inquiry voice in historical cases;

[0077] According to the mapping of quantum states, convert the second MFCC features into the second state vector in the quantum space, where the second state vector is the second emotional intensity of the historical case.

[0078] It should be noted that the Mel Frequency Cepstral Coefficient (MFCC) is a commonly used feature extraction method in speech signal processing. It converts the audio signal into a set of numerical values that can reflect the spectral characteristics of the speech; map the MFCC features to the quantum state space through a certain function. The state space in quantum computing is usually a complex vector space with a specific dimension, and the numerical values of MFCC need to be converted into the corresponding quantum states through a suitable quantum mapping method (such as quantum state encoding or quantum machine learning methods).

[0079] In an optional embodiment, the calculation formula of the first state vector is as follows:

[0080]

[0081] where, |ψ audio > represents the first state vector, e i represents the basis vector of the first quantum state, n represents the total number of basis vectors of the first quantum state, α i represents the weight of the basis vector of the first quantum state, and α i = Softmax(W a ·MFCC(A)), W a represents the first projection matrix, and MFCC(A) represents the first MFCC features of the student inquiry voice;

[0082] The calculation formula of the second state vector is as follows:

[0083]

[0084] where, |ψ exp > represents the second state vector, b j represents the basis vector of the second quantum state, m represents the total number of basis vectors of the second quantum state, θ j represents the weight of the basis vector of the second quantum state, and θ j = Softmax(W b ·MFCC(B)), W b represents the second projection matrix, and MFCC(B) represents the second MFCC features of the expert inquiry voice.

[0085] In an optional embodiment, the calculation formula of the first teaching achievement is as follows:

[0086]

[0087] Among them, L1 represents the first teaching achievement, T represents the number of time steps, σ represents the Gaussian kernel width, and |ψ audio,t > represents the first state vector at the t-th time step, and |ψ exp,t > represents the second state vector at the t-th time step.

[0088] In an alternative embodiment, determining the second teaching achievement of the target trainee according to the difference between the trainee's treatment text and the historical case includes:

[0089] Define a treatment decision knowledge graph, where the treatment decision knowledge graph includes a set of nodes and connecting edges. Among them, the set of nodes includes symptom nodes, diagnosis nodes, and treatment option nodes, and the connecting edges are used to represent the relationships between the various nodes in the set of nodes;

[0090] Extract the treatment path from the trainee's treatment text according to the treatment decision knowledge graph to obtain the trainee's treatment path;

[0091] Extract the treatment path from the expert treatment text in the historical case according to the treatment decision knowledge graph to obtain the expert treatment path;

[0092] Calculate the path difference between the trainee's treatment path and the expert treatment path according to the minimum path edit distance to obtain the difference between the trainee's treatment text and the historical case;

[0093] Calculate the case complexity adjustment coefficient of the trainee's treatment text according to the treatment decision knowledge graph;

[0094] Calculate the second teaching achievement of the target trainee according to the case complexity adjustment coefficient of the trainee's treatment text and the difference between the trainee's treatment text and the historical case.

[0095] It should be noted that the trainee's treatment path is the treatment decision-making process made by the trainee based on symptoms, diagnosis, and treatment options. Through the path extraction method based on the treatment decision knowledge graph, the treatment decision-making process of the trainee when dealing with a certain case can be identified; this path includes the treatment process formed by the trainee starting from the symptom node, passing through the diagnosis node and the treatment option node; the expert treatment path is the path extracted from the expert treatment text in the historical case, which usually represents the best treatment process verified by practice; this path reflects the handling method of the expert when facing a certain case, including the process from symptoms to diagnosis and then to treatment options; the path edit distance is a standard for measuring the difference between two paths, and the minimum path edit distance measures the similarity between the two by calculating the minimum number of editing operations (such as adding, deleting, or modifying nodes) required to transform the trainee's treatment path into the expert treatment path.

[0096] In an alternative embodiment, the calculation formula for the path difference is as follows:

[0097]

[0098] where Δ represents the path difference, |P audio | represents the total number of nodes in the trainee's treatment path, P audio represents a node in the trainee's treatment path, P exp represents a node in the expert's treatment path, and d(p,q) represents the path edit distance;

[0099] The calculation formula for the case complexity adjustment coefficient is as follows:

[0100]

[0101] where Complex(c) represents the case complexity adjustment coefficient, N c represents the number of symptom nodes, τ c represents the number of diagnosis nodes, M c represents the number of treatment option nodes;

[0102] The calculation formula for the second teaching score is as follows:

[0103]

[0104] where L2 represents the second teaching score, and β represents a preset weight coefficient.

[0105] In an alternative embodiment, the difficulty of teaching test questions and case prompt words is adjusted according to the first teaching score and the second teaching score, including:

[0106] Defining an objective function according to the preset target score, the first teaching score, and the second teaching score;

[0107] Solving the objective function to obtain the change in difficulty parameters;

[0108] Adjusting the difficulty of teaching test questions and case prompt words according to the change in difficulty parameters.

[0109] It should be noted that the optimization of the objective function needs to be achieved by adjusting the "difficulty parameters". The difficulty parameters are the key factors affecting the difficulty of teaching test questions or case prompt words, and they examine the change in the trainee's performance between the first teaching score and the second teaching score. This change reflects the suitability of the current teaching difficulty.

[0110] In an alternative embodiment, the objective function is as follows:

[0111]

[0112] Among them, L target represents a preset objective function, γ represents a preset penalty coefficient, and Δθ represents a change in the difficulty parameter to be solved.

[0113] The embodiments of the present invention have been described in detail above in conjunction with the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made without departing from the spirit of the present invention within the scope of knowledge possessed by those skilled in the art.

Claims

1. A teaching process management method for standardized training of dermatology residents, characterized in that, Including: Obtain a historical case database, randomly generate case prompt words and teaching test questions according to the historical case database, and pre-train a preset large language model according to the case prompt words to obtain a virtual patient; The target trainee conducts a diagnosis and treatment consultation on the virtual patient according to the teaching test questions to obtain the trainee's consultation speech, performs speech recognition on the trainee's consultation speech to obtain a consultation text, the virtual patient generates symptom information of the patient according to the consultation text, and the target trainee generates a trainee treatment text according to the symptom information; Perform speech emotion recognition on the target trainee and the historical case according to the trainee's consultation speech and the historical case database to obtain the first emotion intensity of the target trainee and the second emotion intensity of the historical case, and determine the first teaching score of the target trainee according to the difference between the first emotion intensity and the second emotion intensity; Determine the second teaching score of the target trainee according to the difference between the trainee treatment text and the historical case; Adjust the difficulty of the teaching test questions and the case prompt words according to the first teaching score and the second teaching score.

2. The teaching process management method for standardized training of dermatology residents according to claim 1, characterized in that The historical case includes patient disease information and historical diagnosis and treatment information. Among them, the patient disease information includes the patient's basic information, diseases and symptoms, the historical diagnosis and treatment information includes expert treatment text and expert consultation speech, the case prompt words include patient disease information, and the teaching test questions include the patient's basic information and diseases.

3. The teaching process management method for standardized training of dermatology residents according to claim 2, characterized in that, Randomly generating case prompt words according to the historical case database includes: Define a semantic graph, where the semantic graph includes multiple triples, and the triple includes a disease node, a symptom node and a relationship edge, and the expression of the triple is as follows: G i =(d i , r ij , s j ); Among them, G i represents the i-th triple in the semantic graph, d i represents the i-th disease node in the semantic graph, s j represents the i-th symptom node in the semantic graph, r ij represents the relationship between the i-th disease node and the i-th symptom node in the semantic graph; Aggregate information of the nodes and adjacent nodes in the semantic graph according to the graph attention network, adjust the contribution degrees of different nodes according to the attention mechanism, and generate the final case prompt words according to the contribution degrees of different nodes. The generation formula of the case prompt words is as follows: Prompt k = GAT(d k ) ⊕ TF-IDF(c k ); Among them, Prompt k represents the case prompt word generated from the k-th historical case in the historical case library, and GAT(d k ) represents the weighted feature representation based on disease nodes generated by the graph attention network. TF-IDF(c k ) represents the symptom node-related vocabulary features in the k-th historical case in the historical case library extracted by the TF-IDF algorithm.

4. A method for managing the teaching process of standardized training for dermatology residents as described in claim 3, characterized in that, Performing speech emotion recognition on the target trainee and the historical case according to the trainee's consultation speech and the historical case database to obtain the first emotion intensity of the target trainee and the second emotion intensity of the historical case, including: Extract the first Mel frequency cepstral coefficient feature of the trainee's consultation speech; According to the mapping of the quantum state, convert the first Mel frequency cepstral coefficient feature into a first state vector in the quantum space, where the first state vector is the first emotion intensity of the target trainee; Extract the second Mel frequency cepstral coefficient feature of the expert consultation speech in the historical case; According to the mapping of the quantum state, convert the second Mel frequency cepstral coefficient feature into a second state vector in the quantum space, where the second state vector is the second emotion intensity of the historical case.

5. The teaching process management method for standardized training of dermatology residents according to claim 4, characterized in that, The calculation formula of the first state vector is as follows: Among them, |ψ audio > represents the first state vector, e i represents the basis vector of the first quantum state, n represents the total number of basis vectors of the first quantum state, α i represents the weight of the basis vector of the first quantum state, and α i = Softmax(W a ·MFCC(A)), W a represents the first projection matrix, and MFCC(A) represents the first Mel-frequency cepstral coefficient feature of the student's consultation voice; The calculation formula of the second state vector is as follows: Among them, |ψ exp > represents the second state vector, b j represents the basis vector of the second quantum state, m represents the total number of basis vectors of the second quantum state, θ j represents the weight of the basis vector of the second quantum state, and θ j = Softmax(W b ·MFCC(B)), W b represents the second projection matrix, and MFCC(B) represents the second Mel-frequency cepstral coefficient feature of the voice of the expert consultation.

6. A teaching process management method for standardized training of dermatology residents according to claim 5, characterized in that The calculation formula of the first teaching score is as follows: Among them, L1 represents the first teaching achievement, T represents the number of time steps, σ represents the Gaussian kernel width, |ψ audio,t > represents the first state vector at the t-th time step, |ψ exp,t > represents the second state vector at the t-th time step.

7. A teaching process management method for standardized training of dermatology residents according to claim 6, characterized in that, Determining the second teaching score of the target trainee according to the difference between the trainee treatment text and the historical case includes: Define a treatment decision knowledge graph, where the treatment decision knowledge graph includes a node set and connecting edges. Among them, the node set includes symptom nodes, diagnosis nodes, and treatment option nodes, and the connecting edges are used to represent the relationships between the various nodes in the node set; Extract the treatment path from the trainee's treatment text according to the treatment decision knowledge graph to obtain the trainee's treatment path; Extract the treatment path from the expert treatment text in the historical case according to the treatment decision knowledge graph to obtain the expert treatment path; Calculate the path difference between the trainee's treatment path and the expert treatment path according to the minimum path edit distance to obtain the difference between the trainee's treatment text and the historical case; Calculate the case complexity adjustment coefficient of the trainee's treatment text according to the treatment decision knowledge graph; Calculate the second teaching score of the target trainee according to the case complexity adjustment coefficient of the trainee's treatment text and the difference between the trainee's treatment text and the historical case.

8. A teaching process management method for standardized training of dermatology residents as described in claim 7, characterized in that The calculation formula for the path difference is as follows: where, Δ represents the path difference, |P audio | represents the total number of nodes in the trainee's treatment path, P audio represents the nodes in the trainee's treatment path, P exp represents the nodes in the expert's treatment path, and d(p, q) represents the path edit distance; The calculation formula for the case complexity adjustment coefficient is as follows: Among them, Complex(c) represents the case complexity adjustment coefficient, N c represents the number of symptom nodes, τ c represents the number of diagnosis nodes, M c represents the number of treatment option nodes; The calculation formula for the second teaching score is as follows: Among them, L2 represents the second teaching score, and β represents a preset weight coefficient.

9. A teaching process management method for standardized training of dermatology residents as claimed in claim 8, characterized in that Adjust the difficulty of the teaching questions and the case prompt words according to the first teaching score and the second teaching score, including: Define an objective function according to the preset target score, the first teaching score, and the second teaching score; Solve the objective function to obtain the change amount of the difficulty parameter; Adjust the difficulty of the teaching questions and the case prompt words according to the change amount of the difficulty parameter.

10. The teaching process management method for standardized training of dermatology residents according to claim 9, characterized in that, The objective function is as follows: Among them, L target represents a preset objective function, γ represents a preset penalty coefficient, and Δθ represents the change in the difficulty parameter to be solved.

Images