Circuit principle course customized exercise recommendation and auxiliary guidance method based on deep reinforcement learning and large model

Abstract

The invention discloses a circuit principle course customized exercise recommendation and auxiliary guidance method based on deep reinforcement learning and a large model, and the method comprises the steps: building a Transform model to predict the mastering condition of a learner for circuit principle course knowledge, and building a Mama model to predict knowledge points which should be learned by the learner in the next step; establishing a credible exercise difficulty evaluation model to calculate a difficulty coefficient of each exercise, wherein a five-point attribution method is used to assign and classify different types of circuit principle exercises; establishing a Markov decision process model for circuit principle exercise recommendation, generating a recommended optimal circuit principle exercise set by using a deep reinforcement learning algorithm, customizing exercise solving guidance cues according to learning conditions of learners, and calling a large model interface to generate solving guidance for recommended exercises. By adopting the method provided by the invention, the problem solving guidance outline of the circuit principle exercises can be effectively recommended in a customized manner according to the learning condition of the learner.

Description

Technical Field

[0001] The present invention relates to customized auxiliary teaching of circuit principle course, and in particular to a circuit principle course exercise recommendation strategy based on deep reinforcement learning and large model. Background Art

[0002] Circuit principle is one of the core courses in engineering disciplines, but due to its complex content and tedious mathematical derivation, students often face challenges in the learning process. Therefore, it is crucial for students to master circuit principles through exercises in the course for them to master complex concepts and solve practical problems. Existing exercise recommendation models use predefined rules or simple logical judgments to recommend exercises. For example, questions of appropriate difficulty are selected based on students' historical answering situation (such as correct rate). This type of method is simple and direct, but it has poor flexibility and is difficult to adapt to the learning progress and knowledge structure of individual students. Or based on the collaborative filtering technology widely used in recommendation systems, the answer data between similar students is used to recommend exercises. However, it relies on a large amount of data. When the data is sparse or the diversity of the student group is large, the recommendation effect will be significantly reduced. Some systems use shallow machine learning algorithms, such as decision trees, random forests, or support vector machines, to predict the most suitable exercise type based on students' answering performance. Although these algorithms have improved the accuracy of recommendations to a certain extent, they often rely on manual feature engineering and are difficult to capture students' complex learning behaviors and habits.

[0003] Due to the uniqueness of the circuit principle course and the complexity of evaluating the ability to master the principles, the existing recommendation model cannot complete the exercise recommendation task well. The main problems include: the existing exercise recommendation system usually relies on preset rules or simple algorithms, and cannot effectively adapt to students' individual differences and dynamic learning needs; the difficulty division of the circuit principle knowledge involved in the exercises is less reasonable; the existing system has limited ability to analyze and quantify the advanced concepts and knowledge points involved in the circuit principle; the existing system has limited performance in exercise generation and analysis for the advanced concepts and multi-step derivations involved in the circuit principle, and the auxiliary teaching effect is not obvious. Summary of the invention

[0004] The purpose of this invention is to provide a customized auxiliary teaching method for circuit principle courses based on deep reinforcement learning and large models, so as to help circuit principle teaching improve teaching and practice effects through exercise recommendation.

[0005] To achieve the above object, the specific scheme of the present invention is as follows:

[0006] Step S101, establishing a Transformer model to predict the mastery of circuit course knowledge points by circuit course learners;

[0007] The input of the Transformer model is the learner ID, the sequence of circuit principle exercises that have been done, and the binary vector b of the answer status of the question t ∈[0,1], b t The value of x={id,e j (k),b t ,P t}, where all k knowledge points contained in the circuit principle course are represented by K = {k1, k2, ..., k k}, the exercise is represented by e j (k) = [e j (k1),e j (k2),e j (k3),…,e j (k k )], j represents the jth exercise in the exercise bank, e j (k i ) takes a value of 0 or 1 (0 means that the question does not contain the concept of the i-th knowledge point, and 1 means that the question contains it), and embeds the positional encoding P in the input of the Transformer model. t The output of the model is a k-dimensional vector O containing the degree of mastery of each knowledge point in the circuit principle course. t , the optimization loss function for a single learner is expressed as

[0008]

[0009] Here, t represents the t-th position in the sequence and · represents the dot product.

[0010] Step S102, establish a Mamba model to predict the knowledge points that the learner should learn next.

[0011] Use the Mamba model to predict the knowledge points that learners should learn next. The input is the sequence E of exercises a student has done in a certain period of time and the knowledge points K that have been mastered. t , the output is Represents the k-dimensional probability vector of all knowledge points appearing in the next exercise, and uses the binary cross entropy loss function to train the model; the training loss function is

[0012]

[0013] Where t represents the tth position in the sequence and · represents the dot product;

[0014] The teacher gives the weight vector of circuit principle knowledge ω(k) = [ω(k1), ω(k2),…, ω(k k)], use the predicted k-dimensional probability vector and the weight vector to obtain the final knowledge point coverage prediction P(k), which is calculated as follows:

[0015] P(k i )=p(k i )ω(k i ) (10)

[0016] Among them, P(k i )∈P(k);

[0017] Step S103: Establish a reliable exercise difficulty evaluation model to calculate the difficulty coefficient of each exercise.

[0018] Step (3.1): First, use the five-point attribution method to assign values ​​and classify different types of circuit principle exercises, and obtain the difficulty and number of each type of questions: In terms of question type, the test of the understanding of basic circuit knowledge is divided into single, complex and comprehensive integration of multiple concepts, which are assigned values ​​of 0, 0.5 and 1 respectively. The content corresponds to a single circuit element or simple circuit analysis, comprehensive integration of multiple concepts and open knowledge point examination, and the corresponding assignments are 0, 0.5 and 1 respectively; in terms of knowledge content, one knowledge point is assigned 0, two knowledge points are assigned 0.5, and more than two knowledge points are assigned 1; in terms of knowledge type, it is divided into formula type and circuit diagram, which are assigned values ​​of 0.5 and 1 respectively; in terms of reasoning ability, according to personal cognition, one-step reasoning to get the answer is assigned 0, medium reasoning, that is, two-step reasoning to get the answer, is assigned 0.5, and complex reasoning, multiple-step reasoning, is assigned 1; in terms of cognitive level, memorizing circuit knowledge theorems is assigned 0, understanding circuit laws and understanding circuit diagrams is assigned 0.5, and applying and integrating is assigned 1;

[0019] Step (3.2): Calculate the quantitative value of each influencing factor using the weighted average method according to the situation of the question group. The weighted average formula of each factor is as follows:

[0020]

[0021] where p i (i=1,2,3,4,5) is the difficulty value of the i-th difficulty factor among the five difficulty factors. ij represents the number of questions at the jth (j=1, 2, 3) level in the i-th difficulty factor in a group of questions, l ij represents the value of the jth level in the i-th difficulty factor, n is the total number of questions in the group, and after calculating the values ​​of the above five difficulty factors, the difficulty value P in the range of [0,1] is obtained;

[0022] Step (3.3): Calculate the error rate of each question in a natural class of a grade, and adjust the initial difficulty coefficient of the question according to the error rate;

[0023] Step (3.4): Calculate P and the initial difficulty coefficient of the question and multiply them to calculate the weighted average to obtain the final difficulty coefficient d. The weight is given by the teacher or student.

[0024] Step S104, establishing a Markov decision process model for circuit principle exercises recommendation,

[0025] Establish a Markov decision process model for circuit principle exercises. The state includes the current exercise e t , target T, the current learner's mastery of all k knowledge points contained in the circuit course h t , the knowledge point coverage of the next question and the difficulty coefficient d of the exercise, where the target T is represented by the knowledge point vector that the teacher or student needs to master and learn; the action space is to select the candidate exercise set A, and the candidate exercises are matched from the exercise set by predicting the knowledge points that need to be learned in the next step. The reward value at each moment is the average mastery of the knowledge points p t The weighted average of the difficulty difference |θ-d|, θ is the expected difficulty coefficient, and the weight is given by the student or teacher, where p t h t The average value of each element in is calculated at each moment as follows;

[0026] R t =αp t +β|θ-d| (12)

[0027] Among them, a and β are custom weights, p t h t The average value of each element in .

[0028] Step S105, using a deep reinforcement learning algorithm to generate a recommended optimal circuit principle exercise set,

[0029] Step (5.1): Use a policy network π θ and a value network V φ , using the policy network π θ Select action a, obtain the reward value through reward value calculation, and use the value network to estimate the long-term value after taking this action to obtain the target value y t

[0030] y t =R t +γV φ (S t+1 ) (13)

[0031] Where γ represents the discount factor;

[0032] Step (5.2): The policy network optimizes the policy by maximizing the objective function, which is

[0033]

[0034] where r t (θ) represents the ratio between the new and old strategies, S t+1 represents the state at time t+1, S t represents the state at time t, represents the advantage function, which measures the quality of actions, and ε is a hyperparameter that controls the update range between strategies.

[0035] The advantage function of step (5.3) is:

[0036]

[0037] Where Q(S t ,a t ) is defined as the learner’s current knowledge state S t Next, select Exercise a t The subsequent long-term cumulative return is obtained by predicting the learner's knowledge mastery after time t through the model in claim 2 and taking the weighted average; V(S t ) The value network is used to estimate and calculate the degree of mastery of each knowledge, and the weighted average is used to obtain the unfinished action a t The mastery of knowledge points at that time.

[0038] Step S106: Customize prompt words and use the big model interface to provide personalized problem-solving guidance.

[0039] Step (6.1): The prompt words include the learner's mastery of the circuit course knowledge points, the learner's answer history, and splicing all circuit knowledge points;

[0040] Step (6.2): ​​In the prompt words, explain that the principle of generating problem-solving guidance is "to focus on the knowledge points that learners are weak in and the knowledge points that are less covered in the problem-solving guidance";

[0041] Step (6.3): Based on the above, the solution guidance words of the recommended exercises are given, and the large model interface is called in a zero shot manner to generate the solution guidance of the recommended exercises.

[0042] Compared with the existing exercise recommendation method and auxiliary guidance method, the present invention has the following advantages:

[0043] 1. The present invention uses the Transformer and Mamba models to recommend the next learning content for each learner based on the learner's historical answer data and knowledge point mastery, optimize the learning path, and improve learning efficiency.

[0044] 2. The present invention establishes a reliable exercise difficulty evaluation model that comprehensively considers the characteristics of the exercises and the error rate within the class, accurately calculates the exercise difficulty coefficient, and ensures that the recommended questions are of moderate difficulty, neither too simple nor too difficult, to help learners improve effectively.

[0045] 3. The present invention uses the Markov decision process model to track the learner's current learning status in real time, dynamically adjust the recommended exercises based on the mastery of knowledge points and the difficulty of exercises, and ensure that the learning process progresses steadily.

[0046] 4. The present invention combines deep reinforcement learning algorithms and large model interfaces to generate customized problem-solving guidance. By highlighting learners' weak knowledge points and less covered knowledge points, it helps learners improve their shortcomings in a more targeted manner and enhances learning effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A flow chart of a method for recommending customized exercises and assisting guidance for a circuit principle course based on deep reinforcement learning and large models provided in an embodiment of the present invention.

[0048] Figure 2 A diagram of the Transformer structure provided in an embodiment of the present invention.

[0049] Figure 3 A diagram of the Mamba structure provided by an embodiment of the present invention.

[0050] Figure 4 A diagram of a question difficulty classification model for the five-point attribution method provided in an embodiment of the present invention.

[0051] Figure 5 Provided for an embodiment of the present invention is a customized problem-solving guidance flow chart. DETAILED DESCRIPTION

[0052] In order to make the features and advantages of the present invention more obvious and understandable, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0053] Figure 1 This is a flow chart of a method for recommending customized exercises and assisting guidance for a circuit principle course based on deep reinforcement learning and a large model in an embodiment of the present invention. Figure 1 As shown, a method for customizing exercises recommendation and auxiliary guidance for circuit principle courses based on deep reinforcement learning and large models includes:

[0054] Step S101: Establishing a Transformer model to predict the mastery of circuit course knowledge points by circuit course learners;

[0055] Step S102: Establishing a Mamba model to predict the knowledge points that the learner should learn next;

[0056] Step S103: Establish a reliable exercise difficulty evaluation model to calculate the difficulty coefficient of each exercise;

[0057] Step S104: Establishing a Markov decision process model for circuit principle exercises recommendation;

[0058] Step S105: using a deep reinforcement learning algorithm to solve a recommended optimal circuit exercise set;

[0059] Step S106: Customize prompt words and use the large model interface to provide personalized problem-solving guidance;

[0060] Step S101 specifically includes:

[0061] Transformer model Figure 2 As shown, the input is the learner id, the sequence of circuit principle exercises that have been done, and the binary vector b of the answer status of the question t ∈[0,1], b t The value of x={id,e j (k),b t ,P t}, where all k knowledge points contained in the circuit principle course are represented by K = {k1, k2, ..., k k}, the exercise is represented by e j (k) = [e j (k1),e j (k2),e j (k3),…,e j (k k )], j represents the jth exercise in the exercise bank, e j (k i ) takes a value of 0 or 1 (0 means that the question does not contain the concept of the i-th knowledge point, and 1 means that the question contains it), and embeds the positional encoding P in the input of the Transformer model. t The output of the model is a k-dimensional vector O containing the degree of mastery of each knowledge point in the circuit principle course. t , the optimization loss function for a single learner is expressed as

[0062]

[0063] Here, t represents the t-th position in the sequence and · represents the dot product.

[0064] Step S102 specifically includes:

[0065] Use the Mamba model to predict the knowledge points that learners should learn next. The Mamba model unit is as follows: Figure 3 As shown, the input is a sequence E of exercises that a student has done in a certain period of time, and the knowledge points K that have been mastered t , the output is Represents the k-dimensional probability vector of all knowledge points appearing in the next exercise, and uses the binary cross entropy loss function to train the model; the training loss function is

[0066]

[0067] Where t represents the tth position in the sequence and · represents the dot product;

[0068] The teacher gives the weight vector of circuit principle knowledge ω(k) = [ω(k1), ω(k2),…, ω(k k )], use the predicted k-dimensional probability vector and the weight vector to obtain the final knowledge point coverage prediction P(k), which is calculated as follows:

[0069] P(k i )=p(k i )ω(k i )(19)

[0070] Among them, P(k i )∈P(k);

[0071] Step S103 specifically includes:

[0072] Step (3.1): First, use the five-point attribution method to assign values ​​and classify different types of circuit principle exercises, such as Figure 4 As shown, the difficulty and number of each type of questions are obtained: in terms of question type, the understanding of basic circuit knowledge is divided into single, complex and comprehensive integration of multiple concepts, which are assigned values ​​of 0, 0.5 and 1 respectively; in terms of knowledge content, one knowledge point is assigned a value of 0, two knowledge points are assigned a value of 0.5, and more than two knowledge points are assigned a value of 1; in terms of knowledge type, it is divided into formula type and circuit diagram, which are assigned values ​​of 0.5 and 1 respectively; in terms of reasoning ability, one-step reasoning based on personal cognition to get the answer is assigned a value of 0, medium reasoning, that is, two-step reasoning to get the answer, is assigned a value of 0.5, and complex reasoning, multi-step reasoning, is assigned a value of 1; in terms of cognitive level, memorizing circuit knowledge theorems is assigned a value of 0, understanding circuit laws and circuit diagrams is assigned a value of 0.5, and applying and integrating them is assigned a value of 1;

[0073] Step (3.2): Calculate the quantitative value of each influencing factor using the weighted average method according to the situation of the question group. The weighted average formula of each factor is as follows:

[0074]

[0075] where pi (i=1,2,3,4,5) is the difficulty value of the i-th difficulty factor among the five difficulty factors. ij represents the number of questions at the jth (j=1, 2, 3) level in the i-th difficulty factor in a group of questions, l ij represents the value of the jth level in the i-th difficulty factor, n is the total number of questions in the group, and after calculating the values ​​of the above five difficulty factors, the difficulty value P in the range of [0,1] is obtained;

[0076] Step (3.3): At the same time, the initial difficulty coefficient of each question in a natural class of a grade is set to 0.5, and then the error rate of each question in the class is calculated, and the initial difficulty coefficient of the question is adjusted according to the error rate of each question;

[0077] Step (3.4): Calculate P and multiply it by the initial difficulty coefficient of the question to calculate the average value to obtain the final difficulty coefficient d.

[0078] Step S104 specifically includes:

[0079] Establish a Markov decision process model for circuit principle exercises. The state includes the current exercise e t , target T, the current learner's mastery of all k knowledge points contained in the circuit course h t , the knowledge point coverage of the next question and the difficulty coefficient d of the exercise, where the target T is represented by the knowledge point vector that the teacher or student needs to master and learn; the action space is to select the candidate exercise set A, and the candidate exercises are matched from the exercise set by predicting the knowledge points that need to be learned in the next step. The reward value at each moment is the average mastery of the knowledge points p t The weighted average of the difficulty difference |θ-d|, θ is the expected difficulty coefficient, and the weight is given by the student or teacher, where p t h t The average value of each element in is calculated at each moment as follows;

[0080] R t =αp t +β|θ-d| (21)

[0081] Among them, α and β are custom weights, p t h t The average value of each element in .

[0082] Step S105 specifically includes:

[0083] Step (5.1): Use a policy network π θ and a value network V φ , using the policy network π θSelect action a, obtain the reward value through reward value calculation, and use the value network to estimate the long-term value after taking this action to obtain the target value y t

[0084] y t =R t +γV φ (S t+1 ) (twenty two)

[0085] Where γ represents the discount factor;

[0086] Step (5.2): The policy network optimizes the policy by maximizing the objective function, which is

[0087]

[0088] where r t (θ) represents the ratio between the new and old strategies, represents the advantage function, which measures the quality of actions, and ε is a hyperparameter that controls the update range between strategies.

[0089] The advantage function of step (5.3) is:

[0090]

[0091] Where Q(S t ,a t ) is defined as the learner’s current knowledge state S t Next, select Exercise a t The subsequent long-term cumulative return is obtained by predicting the learner's knowledge mastery after time t through the model in claim 2 and taking the weighted average; V(S t ) The value network is used to estimate and calculate the degree of mastery of each knowledge, and the weighted average is used to obtain the unfinished action a t The mastery of knowledge points at that time.

[0092] Step S106 specifically includes:

[0093] Customize prompt words and use the big model interface to give personalized problem-solving guidance outlines. The process is as follows Figure 5 As shown in the figure, the prompt words include the learner's mastery of the circuit course knowledge points, the learner's answer history, and all the circuit knowledge points are spliced ​​together. The prompt words explain the principle of generating problem-solving guidance as "focusing on the learner's weak knowledge points and less covered knowledge points in the problem-solving guidance". In summary, the problem-solving guidance prompt words of the recommended exercises are given, and the large model interface is called in a zero shot manner to generate the problem-solving guidance outline of the recommended exercises.

Claims

1. A customized exercise recommendation and auxiliary guidance method for circuit principle courses based on deep reinforcement learning and large models, characterized by: The following steps are involved: Step S101: Establish a Transformer model to predict the learner's mastery of the knowledge points of the circuit principle course. The input data is a concatenated vector of embedded position coding, which contains a binary vector of the learner ID, the circuit principle exercises done, and the answer status of the questions. The output of the model is a k-dimensional vector containing the mastery level of each knowledge point. The binary cross entropy is used as the loss function for model training. Step S102: Establish a Mamba model to predict the knowledge points that the learner should learn next. The input is the sequence of exercises done by a student in a certain period of time and the knowledge points that have been mastered. The output of the model is a k-dimensional probability vector of all knowledge points appearing in the next exercise. At the same time, the teacher gives a weight vector for each knowledge point. The two vectors are multiplied to obtain the coverage probability of each knowledge point in the next exercise. Step S103: Establish a reliable exercise difficulty evaluation model to calculate the difficulty coefficient of each exercise. First, use the five-point attribution method to assign values ​​and classify different types of circuit principle exercises, obtain the difficulty and number of each type of questions, and calculate the weighted average of the question in the entire question set. At the same time, first set the initial difficulty coefficient of each question in a natural class of a grade to 0.5, and then adjust its initial difficulty coefficient according to the error rate of past answerers of this question. The average of the two is the final difficulty coefficient; Step S104: Establishing a Markov decision process model for circuit principle exercises recommendation, the state includes the current exercises being practiced, the goal, the current learner's mastery of circuit course knowledge points, the knowledge point coverage of the next question, and the difficulty coefficient of the exercise, the action space is to select candidate exercises, and the reward value is the weighted average of the difference between the average mastery of knowledge points and the expected difficulty; Step S105: Use a deep reinforcement learning algorithm to generate a recommended optimal circuit principle exercise set, adopt a PPO algorithm, define a policy network and a value network, select an action through the policy network, obtain the reward value through the reward value calculation defined in S104, and use the value network to estimate the long-term value after taking the action, so as to obtain the target value; then define the advantage function and design the target function training model; Step S106: Customize prompt words and use the big model interface to give a personalized problem-solving guidance outline. The prompt words include the learner's mastery of circuit course knowledge points, the learner's answer history, and splice all circuit knowledge points. The prompt words explain that the principle of generating problem-solving guidance is "focusing on the learner's weak knowledge points and less covered knowledge points in the problem-solving guidance". Based on the above, problem-solving guidance prompt words for recommended exercises are given, and the big model interface is called in a zero shot manner to generate a problem-solving guidance outline for the recommended exercises.

2. The method for customizing exercises and assisting guidance for circuit principle courses based on deep reinforcement learning and large models according to claim 1 is characterized in that The establishment of the Transformer model in step S101 to predict the learner's mastery of the knowledge points of the circuit principle course specifically includes: The input of the Transformer model is the learner ID, the sequence of circuit principle exercises that have been done, and the binary vector b of the answer status of the question t ∈[0,1], b t The value of x={id,e j (k),b t ,P t }, where all k knowledge points contained in the circuit principle course are represented by K = {k1, k2, ..., k k }, the exercise is represented by e j (k) = [e j (k1),e j (k2),e j (k3),…,e j (k k )], j represents the jth exercise in the exercise bank, e j (k i ) takes a value of 0 or 1 (0 means that the question does not contain the concept of the i-th knowledge point, and 1 means that the question contains it), P t Encode the embedded position P t The output of the model is a k-dimensional vector O containing the degree of mastery of each knowledge point in the circuit principle course. t .

3. The method for customizing exercises and assisting guidance for circuit principle courses based on deep reinforcement learning and large models according to claim 1 is characterized in that The knowledge points that the learner should learn next in establishing the Mamba model in step S102 specifically include: Use the Mamba model to predict the knowledge points that learners should learn next. The input is the sequence E of exercises a student has done in a certain period of time and the knowledge points K that have been mastered. t , the output is Represents the k-dimensional probability vector of all knowledge points appearing in the next exercise, and uses the binary cross entropy loss function to train the model; The teacher gives the weight vector of circuit principle knowledge ω(k) = [ω(k1), ω(k2),…, ω(k k )], use the predicted k-dimensional probability vector and the weight vector to obtain the final knowledge point coverage prediction P(k), which is calculated as follows: A.D i )=p(k i )ω(k i )(1) among others,P(k i )∈P(k).

4. The method for customizing exercises recommendation and auxiliary guidance for circuit principle courses based on deep reinforcement learning and large models according to claim 1 is characterized in that The circuit principle question difficulty classification model based on the five-point attribution method described in step S103 specifically includes: A difficulty classification model for circuit principle questions is constructed based on the five-point attribution method. The examination levels are divided from difficulty factors, levels, explanations and assignments, involving multiple knowledge points, understanding, application and practice. The model is summarized as follows: in terms of question types, the understanding of basic circuit knowledge is tested into single, compound and comprehensive integration of multiple concepts, which are assigned values ​​of 0, 0.5 and 1 respectively; in terms of knowledge content, one knowledge point is assigned a value of 0, two knowledge points are assigned a value of 0.5, and more than two knowledge points are assigned a value of 1; in terms of knowledge type, it is divided into formulas and circuit diagrams, which are assigned values ​​of 0.5 and 1 respectively; in terms of reasoning ability, the answer is assigned a value of 0 based on personal cognition, the answer is assigned a value of 0 based on one-step reasoning, 0.5 based on intermediate reasoning, that is, two-step reasoning, and complex reasoning with multiple steps; in terms of cognitive level, memorizing circuit knowledge theorems is assigned a value of 0, understanding circuit laws and circuit diagrams is assigned a value of 0.5, and applying and integrating them is assigned a value of 1.

5. The method for customizing exercises recommendation and auxiliary guidance for circuit principle courses based on deep reinforcement learning and large models according to claim 1 is characterized in that The method for calculating the trustworthy difficulty coefficient described in step S103 specifically includes: According to the five-point attribution method described in claim 4, different types of circuit principle exercises are assigned and classified to obtain the difficulty and number of each type of questions, and then the quantitative value of each influencing factor is calculated by the weighted average method according to the situation of the question group. The weighted average formula of each factor is as follows: where p i (i=1,2,3,4,5) is the difficulty value of the i-th difficulty factor among the five difficulty factors. ij represents the number of questions at the jth (j=1, 2, 3) level in the i-th difficulty factor in a group of questions, l ij represents the assignment of the jth level in the ith difficulty factor, n is the total number of questions in this group, and the difficulty value P in the range of [0,1] is obtained after calculating the values ​​of the above five difficulty factors; the error rate of each question in a natural class of the same grade is counted, and the weighted average value of P multiplied by the error rate of the question is calculated to obtain the final difficulty coefficient d, and the weight is given by the teacher or student.

6. The method for customizing exercises recommendation and auxiliary guidance for circuit principle courses based on deep reinforcement learning and large models according to claim 1 is characterized in that The step S104 of establishing a Markov decision process model for circuit principle exercises recommendation specifically includes: Establish a Markov decision process model for circuit principle exercises. The state includes the current exercise e t , target T, the current learner's mastery of all k knowledge points contained in the circuit course O t , the knowledge point coverage of the next question and the difficulty coefficient d of the exercise, where the target T is represented by the knowledge point vector that the teacher or student needs to master and learn; the action space is to select the candidate exercise set A, and the candidate exercises are matched from the exercise set by predicting the knowledge points that need to be learned in the next step. The reward value at each moment is the average mastery of the knowledge points at the current moment p t The weighted average of the difficulty difference |θ-d|, where θ is the expected difficulty coefficient and the weight is given by the student or teacher.

7. The method for customizing exercises and assisting guidance for circuit principle courses based on deep reinforcement learning and large models according to claim 1 is characterized in that The calculation of the reward value of the Markov decision process model recommended by the circuit principle exercise in step S104 specifically includes: The return value is the average mastery of knowledge points p t The weighted average of the difficulty difference |θ-d|, θ is the expected difficulty coefficient, and the reward value at each moment is calculated as follows; R t =αp t +β|θ-d| (3) where α and β are custom weights, the weights are given by students or teachers, and p t O t The average value of each element in .

8. The method for customizing exercises recommendation and auxiliary guidance for circuit principle courses based on deep reinforcement learning and large models according to claim 1 is characterized in that The step S105 described in which the deep reinforcement learning algorithm is used to generate a recommended optimal circuit exercise set: Using a policy network π θ and a value network V φ , through the policy network π θ Select action a, calculate the reward value of the current recommended exercise, and use the value network to estimate the long-term value after selecting the current recommended exercise, thereby obtaining the target value y t y t =R t +γV φ (S t+1 ) (4) Where R t represents the return value at time t, γ represents the discount factor, S t+1 Indicates the state at time t+1; The policy network optimizes the policy by maximizing the objective function, which is where r t (θ) represents the ratio between the new and old strategies, represents the advantage function, which measures the quality of actions, and ε is a hyperparameter that controls the update range between strategies.

9. The method for customizing exercises recommendation and auxiliary guidance for circuit principle courses based on deep reinforcement learning and large models according to claim 1 is characterized in that Advantage function described in step S105: The advantage function is: Where Q(S t ,a t ) is defined as the learner’s current knowledge state S t Next, select Exercise a t The subsequent long-term cumulative return is obtained by predicting the learner's knowledge mastery after time t through the model in claim 2 and taking the weighted average; V(S t ) The value network is used to estimate and calculate the current learner's knowledge mastery level, and the weighted average is used to obtain the unfinished action a t The mastery of previous knowledge points.

10. The method for customizing exercises recommendation and auxiliary guidance for circuit principle courses based on deep reinforcement learning and large models according to claim 1, characterized in that The step S106 includes the following steps: The prompt words include the learner's mastery of circuit course knowledge points, the learner's answer history, and splicing all circuit knowledge points. The prompt words explain the principle of generating problem-solving guidance as "focusing on the learner's weak knowledge points and less covered knowledge points in the problem-solving guidance". Based on the above, the problem-solving guidance prompt words of the recommended exercises are given, and the large model interface is called in a zero shot manner to generate the problem-solving guidance outline of the recommended exercises.

Images