Group-type dialogue generation and tutoring learning method and system based on large language model

Abstract

The invention discloses a team-type dialogue generation and tutoring learning method and system based on a large language model, and the method comprises the steps: constructing a rule-based expert course knowledge base according to the course demands and a course knowledge structure; secondly, entering an observation type learning situation, generating personalized knowledge conversations between different teachers and fixed students according to the generation type teacher prompt instruction set, and generating personalized knowledge conversations between the fixed teachers and different students according to the generation type student prompt instruction set; then, entering an interactive learning situation, generating teaching statements of experts and teachers according to a generative interaction initiating prompt instruction set, and generating personalized feedback statements of the experts and teachers according to feedback of students and a generative interaction feedback prompt instruction set; and finally, according to the use experience feedback of the students, knowledge learning effect analysis is carried out, an accurate team-type learning situation is optimized and generated, and personalized tutoring of the students is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of large language models, and specifically relates to a team-based dialogue generation and personalized tutoring learning method and system based on a large language model, in order to use the prompt word strategy of the large language model to design generative teachers and generative students, optimize dialogue generation, initiation and feedback technology, help students to carry out intelligent learning, and develop intelligent tutoring systems in the direction of personalization and intelligence. Background Art

[0002] The Intelligent Tutoring System (ITS) is a new form of traditional teaching in accordance with students' aptitude. It simulates the role of professional tutors and uses intelligent technology to provide students with customized content and feedback to achieve highly personalized learning. Although the intelligent tutoring system has significant advantages in improving learning outcomes, the traditional intelligent guidance system has problems such as isolated learning, mechanical interaction, and a single mode, which makes it difficult to meet the requirements of intelligent teaching or support higher-level thinking activities. With the "Generative Pre-Training" (GPT) large language models represented by Sora, GPT-4o, and iFlytek Spark releasing huge potential in language learning, math problem solving, psychological counseling, etc., GPT-based intelligent tutoring applications have gradually attracted attention. So, how to use GPT to empower the future development of intelligent tutoring systems has attracted widespread attention.

[0003] GPT's anthropomorphic interactive advantages are expected to strengthen expert teachers and interactive dialogue components, enhance the naturalness of human-computer interaction in intelligent tutoring systems, and accelerate their integration into classroom teaching applications. The "Plan for the Development of New Generation Artificial Intelligence" and other documents propose that we should make full use of technologies such as artificial intelligence to build a new education system of intelligent learning and interactive learning. At present, there are large language model education applications such as Spark Language Partner, EmoGPT, and MathGPT. These applications follow the design methods of traditional intelligent tutoring systems, emphasize the personalized generation of knowledge, and are easier for students to accept and use. However, these applications still face the following problems: (1) The quality of dialogue is not high and there is a lack of domain knowledge of expert teachers; (2) There is a lack of educational theory guidance, which may make it difficult to achieve ideal tutoring results; (3) The prompt technology is single and students are still in a one-to-one isolated learning situation.

[0004] In summary, although the large language model has the function of enhancing the practicality of the intelligent tutoring system, the generative dialogue without prompting and verification may not necessarily improve students' learning effects. At present, it is still necessary to explore a teaching model that is compatible with intelligent guidance so that it can produce deeper interactions in the classroom. To this end, it is necessary to reconstruct the intelligent tutoring system in combination with design, theory and practice to provide students with an effective learning experience. Therefore, based on the current development status of the intelligent tutoring system, the present invention designs a team-based dialogue generation and personalized tutoring learning method and system based on a large language model to realize the integration of student observation learning and interactive learning, and provide technical support for students' autonomous learning and personalized learning. Summary of the invention

[0005] In order to solve the problems of the current intelligent tutoring system, such as the rigid dialogue mode, mechanical interaction and single prompt technology, the present invention starts from the large language model to generate personalized knowledge dialogues, teaching sentences and personalized feedback sentences, so as to achieve the teaching goals of accurately tutoring students, accompanying students throughout the process and actively guiding students. The present invention also provides a tutoring learning system based on the large language model to provide technical support for accurately tutoring students in course learning.

[0006] The present invention provides a team-based dialogue generation and tutoring learning method based on a large language model, comprising the following steps:

[0007] Step 1: Build a rule-based expert course knowledge base according to course needs and course knowledge structure;

[0008] Step 2, enter the observational learning scenario, generate personalized knowledge dialogues A between different teachers and fixed students according to the generative teacher prompt instruction set, and generate personalized knowledge dialogues B between fixed teachers and different students according to the generative student prompt instruction set, wherein the personalized knowledge dialogues A and personalized knowledge dialogues B are used to assist students in understanding knowledge and accompany students in learning knowledge respectively;

[0009] Step 3, entering the interactive learning situation, initiating a prompt instruction set according to the generative interaction, generating the teaching sentence C of the expert teacher, wherein the teaching sentence C is used to induce students to participate in knowledge interaction;

[0010] Step 4, enter the interactive learning situation, generate the expert teacher's personalized feedback sentence D based on the student's feedback and the generated interactive feedback prompt instruction set, and use it to respond to the student's knowledge needs;

[0011] Step 5, repeat steps 2, 3 and 4, optimize personalized knowledge dialogue A, personalized knowledge dialogue B, teaching sentence C and personalized feedback sentence D, until the student no longer repeatedly enters steps 2, 3 and 4, it is considered that the student has mastered the course knowledge, and intelligent learning guidance for the student is achieved.

[0012] Furthermore, in step 1, building a rule-based expert course knowledge base includes:

[0013] Organize course knowledge concepts, key points and cases according to the overall knowledge representation rules;

[0014] According to the rule of gradual deepening of knowledge from simple to complex, organize the connection between the key and difficult points of knowledge, including sequential relationship and complementary relationship;

[0015] The course knowledge concepts, key points, cases and the connections between key points are collected to obtain the expert course knowledge base.

[0016] Furthermore, the steps for generating personalized knowledge dialogue A and personalized knowledge dialogue B in step 2 are as follows:

[0017] (21) calling the large language model, configuring the generative teacher prompt instruction set environment and the generative student prompt instruction set environment;

[0018] (22) Obtain the knowledge content that needs assistance selected by the student and obtain the corresponding knowledge content in the expert course knowledge base;

[0019] (23) determining a generative teacher prompt instruction set and a generative student prompt instruction set;

[0020] (24) The generative teacher prompt instruction set environment and the generative student prompt instruction set environment, knowledge content, the generative teacher prompt instruction set and the generative student prompt instruction set are input into the large language model for natural language processing to generate personalized knowledge dialogue A and personalized knowledge dialogue B.

[0021] Furthermore, the set of generative teacher prompt instructions includes:

[0022] (a1) Construct a generative teacher T = {t1, t2, t3} and a randomly generated student s, where t1, t2 and t3 are active teachers, constructive teachers and interactive teachers respectively;

[0023] (a2) The prompting instructions of the active teacher t1 are to directly tell the randomly generated students what the knowledge point K is, what the principle of the knowledge point is, and how to apply the knowledge point, so as to stimulate the feedback of the randomly generated students; the prompting instructions of the constructive teacher t2 are to tell the randomly generated students the knowledge points, principles and applications through analogies and metaphors, so as to stimulate the randomly generated students to connect the new information with the original knowledge; the prompting instructions of the interactive teacher t3 are to guide the randomly generated students through cases and questions, so as to encourage the randomly generated students to construct their understanding of the knowledge points and participate in classroom activities;

[0024] (a3) Input the constructed generative teacher T into the large language model to obtain the classroom speech of the generative teacher T;

[0025] (a4) Determine the prompt instruction P(s) of the random generative student s, where P(s) is to answer the generative teacher T in the role of an ordinary student based on the classroom speech of the generative teacher T;

[0026] (a5) The constructed generative teacher T = {t1, t2, t3} and the prompt instructions of the randomly generated student s are the generative teacher prompt instruction set.

[0027] Furthermore, the generative student prompt instruction set includes:

[0028] (b1) construct a random generative teacher t and a generative student S = {s1, s2}, where s1 and s2 are an expert student and a novice student respectively;

[0029] (b2) Determine the randomly generated teacher t’s prompt instruction P(t), where P(t) is a classroom dialogue around the knowledge point “pointer” in the role of an expert teacher;

[0030] (b3) Input the constructed random generative teacher t into the large language model to obtain the classroom speech of the generative teacher T;

[0031] (b4) Determine the prompt instructions for the generative student S. The prompt instructions for the expert student s1 are to correctly answer and supplement the knowledge points taught by the random generative teacher t in the role of an expert student based on the classroom speech of the random generative teacher t, and to ask questions to the random generative teacher t and apply the knowledge by analogy; the prompt instructions for the novice student s2 are to incorrectly answer and refuse to answer the knowledge points taught by the random generative teacher t in the role of a novice student based on the classroom speech of the random generative teacher t, and to ask questions to the random generative teacher t and seek help;

[0032] (b5) The prompt instructions of the constructed random generative teacher t and generative student S = {s1, s2} are the generative student prompt instruction set.

[0033] Furthermore, the steps for generating teaching sentence C in step 3 are as follows:

[0034] (31) Calling a large language model and configuring a generative interaction initiation prompt instruction set environment;

[0035] (32) Obtain the knowledge content that needs assistance selected by the student and obtain the corresponding knowledge content in the expert course knowledge base;

[0036] (33) Determine a generative interaction initiation prompt instruction set, wherein the prompt instruction set includes {P 1 (T Q ), P2 (T Q )};P 1 (T Q ) is when students first enter the interactive learning situation, the teacher, in the role of an expert teacher, asks students whether they need learning help; P 2 (T Q ) is when students are not in the interactive learning situation for the first time, the teacher, in the role of an expert, asks students application questions and sets cases based on the knowledge content selected by the students, the corresponding knowledge content in the expert course knowledge base, the previous round of teaching sentences C and the students' responses. The questions are in the form of closed questions and open questions, and the difficulty of the questions ranges from simple to difficult;

[0037] (34) The generative interaction initiation prompt instruction set environment, knowledge content, and the generative interaction initiation prompt instruction set are input into the large language model for natural language processing to generate teaching sentences C.

[0038] Furthermore, the steps for generating the personalized feedback sentence D in step 4 are as follows:

[0039] (41) Calling a large language model and configuring a generative interactive feedback prompt instruction set environment;

[0040] (42) Obtain the knowledge content that needs assistance, teaching sentences C, and students’ responses Sr selected by students;

[0041] (43) Determine the expert teacher T who can answer questions QA The generative interactive feedback prompt instruction set comprises {P 1 (T A ), P 2 (T A ), P 3 (T A ), P 4 (T A )},P 1 (T A ) is when the student's reply Sr is not empty and is related to the knowledge content selected by the student that needs assistance, the teacher, in the role of an expert teacher, gives feedback to the student around the student's reply Sr. The feedback includes judging the correctness of the answer, giving encouragement or praise, and providing detailed explanations and prompts; P 2 (T A ) is when the student reply Sr is not empty and has nothing to do with the knowledge content that needs assistance selected by the student, remind the student to answer around the knowledge content that needs assistance selected by the student, or modify the selected knowledge content that needs assistance; P 3 (T A) is when the student does not respond within a short period of time, the teacher, in the role of an expert teacher, asks the student if he needs help; P 4 (T A ) is when students do not respond for a long time, they exit the interactive learning situation;

[0042] (44) inputting the generative interactive feedback prompt instruction set environment, knowledge content, and the generative interactive feedback prompt instruction set into the large language model for natural language processing to generate a personalized feedback sentence D;

[0043] (45) When the generative interactive feedback prompts P in the instruction set 2 (T A ) is activated, repeat steps (2) to (4) until the student exits the interactive learning situation.

[0044] Furthermore, the large language model includes Chatglm-4 and Chatglm-3-turbo.

[0045] The present invention also provides a team-based dialogue generation and tutoring learning system based on a large language model, comprising the following modules:

[0046] Login and registration module, used to record student information so that the system can track students' learning progress and needs;

[0047] Expert course knowledge base module, which is used to store knowledge points, teaching resources and learning materials of various subjects so that students can consult them at any time, and is constructed according to course needs and course knowledge structure;

[0048] An autonomous observational learning module is used to generate personalized knowledge dialogue A and personalized knowledge dialogue B according to a generative teacher prompt instruction set and a generative student prompt instruction set, so as to help students understand and master knowledge through observation;

[0049] The autonomous interactive learning module is used to provide teaching sentences C based on the generative interaction initiation prompt instruction set to induce students to participate in knowledge interaction; based on students' feedback and the generative interaction feedback prompt instruction set, it generates personalized feedback sentences D from expert teachers to respond to students' knowledge needs. Students deepen their understanding and application of knowledge through interactive learning.

[0050] Compared with existing research and technology, the present invention has the following beneficial effects:

[0051] 1. The present invention proposes a course knowledge base construction technology based on a large language model, which establishes connections between course knowledge points through knowledge representation rules and progressive deepening rules, reduces the error rate of knowledge generation in the intelligent tutoring process, and improves the understanding and call of the course knowledge by the large language model.

[0052] 2. The present invention combines pedagogy, psychology theories and big language model technology to establish a personalized knowledge dialogue driven by a big language model, realize the entire process of personalized knowledge tutoring, meet students' observation and learning needs for course knowledge through team-based learning scenarios, and lay the foundation for students' efficient and seamless learning.

[0053] 3. The present invention designs an interactive dialogue mode that integrates teaching sentences and personalized feedback sentences, adaptively adjusts different generation modes, and supports students' personalized intelligent learning needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A flowchart for the team-based dialogue generation and personalized tutoring learning method based on a large language model;

[0055] Figure 2 The generation and learning guidance diagram of personalized knowledge dialogue A;

[0056] Figure 3 The generation and learning guidance diagram of personalized knowledge dialogue B;

[0057] Figure 4 The diagram for the generation and guidance of teaching sentences C and personalized feedback sentences D;

[0058] Figure 5 This is a diagram showing the effect of personalized tutoring on student knowledge learning based on a large language model;

[0059] Figure 6 A learning feedback graph for personalized tutoring based on a large language model;

[0060] Figure 7 Diagram of the personalized tutoring learning system based on a large language model. DETAILED DESCRIPTION

[0061] The technical solution of the present invention is further described in detail below with reference to the accompanying drawings.

[0062] The embodiment of the present invention provides a team-based dialogue generation and personalized tutoring learning method and system based on a large language model. Figure 1 The problems and core ideas solved by these steps are shown. First, the large language model considered in the embodiment of the present invention will be configured as an observational learning scenario and an interactive learning scenario, and constructed as a generative teacher T = {t1, t2, t3}, a randomly generated student s, a randomly generated teacher t, a generative student S = {s1, s2} and an expert question-answering teacher T QA It is hoped that the personalized knowledge dialogue A, personalized knowledge dialogue B, teaching sentences C and personalized feedback sentences D it provides to students can be consistent with the teaching content and basic teaching rules, so as to prompt it to pay attention to the details of learning guidance when outputting. The specific steps include the following:

[0063] Step 1: Build a rule-based expert course knowledge base according to course needs and course knowledge structure;

[0064] Step 2, enter the observational learning scenario, generate personalized knowledge dialogues A between different teachers and fixed students according to the generative teacher prompt instruction set, and generate personalized knowledge dialogues B between fixed teachers and different students according to the generative student prompt instruction set, wherein the personalized knowledge dialogues A and personalized knowledge dialogues B are used to assist students in understanding knowledge and accompany students in learning knowledge respectively;

[0065] Step 3, entering the interactive learning situation, initiating a prompt instruction set according to the generative interaction, generating the teaching sentence C of the expert teacher, wherein the teaching sentence C is used to induce students to participate in knowledge interaction;

[0066] Step 4, enter the interactive learning situation, generate the expert teacher's personalized feedback sentence D based on the student's feedback and the generated interactive feedback prompt instruction set, and use it to respond to the student's knowledge needs;

[0067] Step 5, repeat steps 2, 3 and 4, optimize personalized knowledge dialogue A, personalized knowledge dialogue B, teaching sentence C and personalized feedback sentence D, until the student no longer repeatedly enters steps 2, 3 and 4, it is considered that the student has mastered the course knowledge, and intelligent learning guidance for the student is achieved.

[0068] In addition to the above method, we will test the knowledge points that students have learned. When students think they have mastered the knowledge, they will no longer repeat steps 2, 3, and 4. Therefore, the way to judge whether students have mastered the knowledge is to observe whether students repeat steps 2, 3, and 4. If they repeat, they have not mastered it and are still learning; if they do not repeat, they have mastered the knowledge and students will quit learning.

[0069] To achieve the above purpose, taking the university course "C Language Programming" as an example, the specific steps of building an expert course knowledge base include:

[0070] (1) Based on the understanding of the course knowledge structure, and taking Tan Haoqiang's "C Programming (Fifth Edition)" as the standard, C language knowledge is divided into nine major units: algorithm introduction, sequence structure, selection structure, loop structure, array, function, pointer, structure and file. The key points and cases in each major unit are determined in turn to achieve an overall through-and-through representation of the course knowledge. Table 1 shows an overall through-and-through representation using the "pointer" knowledge point as an example.

[0071] Table 1 Overall representation of “pointer” knowledge points

[0072]

[0073]

[0074] (2) In each large unit, the connection relationship between the key and difficult knowledge points is formulated according to the rule of from simple to complex. For example, the key and difficult knowledge points of "pointer" include: the basic concept of pointer, pointer declaration, pointer operation (address operator, dereference operator, pointer arithmetic operation), pointer and array, pointer and function; the connection relationship includes order relationship and complementary relationship, etc. In the "pointer" example, the order relationship is that the key and difficult knowledge points are from left to right, and the learning difficulty increases successively. The complementary relationship exists between pointer operation and pointer and array, because the arithmetic operation of pointer can be used to traverse arrays, and the concepts of pointer and array are closely related and can be alternated when learning;

[0075] Furthermore, taking the knowledge point "pointer" in the university course "C language programming" as an example, we enter the observational learning situation and generate personalized knowledge dialogue A, such as Figure 2 As shown, the specific steps are as follows:

[0076] (1) Call a large language model (such as Chatglm-4, Chatglm-3-turbo, etc.), configure the generative teacher prompt instruction set environment, and set the system information to "You are a setting assistant in an intelligent tutoring system, and your task is to implement intelligent learning tutoring for users";

[0077] (2) Obtain the knowledge content that students select and need assistance from the expert course knowledge base;

[0078] (3) determine the set of generative teacher prompt instructions;

[0079] (4) inputting the generative teacher prompt instruction set environment, the knowledge content, and the generative teacher prompt instruction set into the large language model for natural language processing to generate a personalized knowledge dialogue A;

[0080] (5) If the content does not meet the students’ learning needs, repeat step (4) until a satisfactory personalized knowledge dialogue result is generated.

[0081] Furthermore, the set of generative teacher prompt instructions includes:

[0082] (1) Construct a generative teacher T = {t1, t2, t3} and a randomly generated student s, where t1, t2 and t3 are active teachers, constructive teachers and interactive teachers respectively;

[0083] (2) The prompting instructions of the active teacher t1 are to directly tell the randomly generated students what the knowledge point K is, what the principle of the knowledge point "pointer" is, and how to apply the knowledge point "pointer", so as to stimulate the feedback of the randomly generated students; the prompting instructions of the constructive teacher t2 are to tell the randomly generated students the knowledge point "pointer", principle and application through analogy and metaphor, so as to stimulate the randomly generated students to connect the new information with the original knowledge in a meaningful way; the prompting instructions of the interactive teacher t3 are to guide the randomly generated students through cases and questions, so as to encourage the randomly generated students to construct their own understanding of the knowledge point "pointer" and participate in classroom activities;

[0084] (3) Input the constructed generative teacher T into the large language model to obtain the classroom speech of the generative teacher T;

[0085] (4) Determine the prompt instruction P(s) of the random generative student s, P(s) is to answer the generative teacher T in the role of an ordinary student according to the classroom speech of the generative teacher T;

[0086] (5) The constructed generative teacher T = {t1, t2, t3} and the prompt instructions of the randomly generated student s are the generative teacher prompt instruction set.

[0087] Furthermore, taking the knowledge point "pointer" in the university course "C language programming" as an example, we enter the observational learning situation and generate personalized knowledge dialogue B, such as Figure 3 As shown, the specific steps are as follows:

[0088] (1) Call a large language model (such as Chatglm-4, Chatglm-3-turbo, etc.), configure the generative student prompt instruction set environment, and set the system information to "You are a setting assistant in an intelligent tutoring system, and your task is to implement intelligent learning tutoring for users";

[0089] (2) Obtain the knowledge content that students select and need assistance from the expert course knowledge base;

[0090] (3) determine the set of generative student prompt instructions;

[0091] (4) inputting the generative student prompt instruction set environment, the knowledge content, and the generative student prompt instruction set into the large language model for natural language processing to generate a personalized knowledge dialogue B;

[0092] (5) If the content does not meet the students’ learning needs, repeat step (4) until a satisfactory personalized knowledge dialogue result is generated.

[0093] Furthermore, the generative student prompt instruction set includes:

[0094] (1) Construct a random generative teacher t and a generative student S = {s1, s2}, where s1 and s2 are an expert student and a novice student respectively;

[0095] (2) Determine the randomly generated teacher t’s prompt instruction P(t), where P(t) is a classroom dialogue based on the knowledge point “pointer” in the role of an expert teacher;

[0096] (3) Input the constructed random generative teacher t into the large language model to obtain the classroom speech of the generative teacher T;

[0097] (4) Determine the prompt instructions for the generative student S. The prompt instructions for the expert student s1 are to correctly answer and supplement the knowledge points taught by the random generative teacher t in the role of an expert student based on the classroom speech of the random generative teacher t, and to ask questions to the random generative teacher t and apply knowledge by analogy; the prompt instructions for the novice student s2 are to incorrectly answer and refuse to answer the knowledge points taught by the random generative teacher t in the role of a novice student based on the classroom speech of the random generative teacher t, and to like to ask questions and seek help from the random generative teacher t;

[0098] (5) The prompt instructions of the randomly generated teacher t and the generative student S = {s1, s2} constructed as described above are the generative student prompt instruction set.

[0099] Furthermore, taking the knowledge point "pointer" of the university course "C language programming" as an example, enter the interactive learning situation and generate teaching sentences C, such as Figure 4 As shown, the specific steps are as follows:

[0100] (1) Call a large language model (such as Chatglm-4, Chatglm-3-turbo, etc.), configure the generative interaction initiation prompt instruction set environment, and set the system information as "You are a setting assistant in an intelligent tutoring system. Your task is to provide interactive tutoring for users. Now you need to construct an expert Q&A teacher T around the knowledge point 'pointer'. QA , ask students questions or show them knowledge cases to promote students’ in-depth cognition”;

[0101] (2) Obtain the knowledge content that students select and need assistance from the expert course knowledge base;

[0102] (3) Determine a generative interaction initiation prompt instruction set, wherein the prompt instruction set includes {P 1 (T Q ), P 2 (T Q )};P 1 (T Q) is when students first enter the interactive learning situation, it is necessary to play the role of an expert teacher and ask students whether they need learning help; P 2 (T Q ) is when students are not in the interactive learning situation for the first time, the role of the expert teacher is to ask students application questions and set cases based on the knowledge content selected by the students, the corresponding knowledge content in the expert course knowledge base, the previous round of teaching sentences C and the students' responses. The questions are in the form of closed questions and open questions, and the difficulty of the questions ranges from simple to difficult;

[0103] (4) inputting the generative interaction initiation prompt instruction set environment, the knowledge content, and the generative interaction initiation prompt instruction set into the large language model for natural language processing to generate teaching sentences C;

[0104] (5) If the content does not meet the students’ learning needs, repeat step (4) until a satisfactory teaching sentence result is generated.

[0105] Furthermore, taking the knowledge point "pointer" of the university course "C language programming" as an example, we enter the interactive learning situation and generate personalized feedback sentence D, such as Figure 4 As shown, the specific steps are as follows:

[0106] (1) Call a large language model (such as Chatglm-4, Chatglm-3-turbo, etc.), configure the generative interactive feedback prompt instruction set environment, and set the system information as "You are a setting assistant in an intelligent tutoring system. Your task is to provide interactive tutoring for users. Now you need to construct an expert Q&A teacher T around the knowledge point 'pointer'. QA , answer students’ confusion and promote students’ in-depth cognition”;

[0107] (2) Obtain the knowledge content that needs assistance, teaching sentences C, and students’ responses Sr selected by students;

[0108] (3) Determine the expert teacher T who can answer questions QA The generative interactive feedback prompt instruction set comprises {P 1 (T A ), P 2 (T A ), P 3 (T A ), P 4 (T A )},P 1 (T A) is when the student's reply Sr is not empty and is related to the knowledge content that needs assistance selected by the student, it is necessary to give feedback to the student in the role of an expert teacher around the student's reply Sr. The feedback includes judging the correctness of the answer, giving encouragement or praise, and providing detailed explanations and prompts; P 2 (T A ) is when the student reply Sr is not empty and has nothing to do with the knowledge content that needs assistance selected by the student, remind the student to answer around the knowledge content that needs assistance selected by the student, or modify the selected knowledge content that needs assistance; P 3 (T A ) is when the student does not respond within a short period of time (e.g., stay time >= 5 minutes), it is necessary to ask the student whether he needs help in the role of an expert teacher; P 4 (T A ) is when students do not respond for a long time (e.g., stay time >= 10 minutes), they exit the interactive learning situation;

[0109] (4) inputting the generative interactive feedback prompt instruction set environment, the knowledge content, and the generative interactive feedback prompt instruction set into the large language model for natural language processing to generate a personalized feedback sentence D;

[0110] (5) When the generative interactive feedback prompts P in the instruction set 2 (T A ) is activated, repeat steps (2) to (4) until the student exits the interactive learning situation.

[0111] Furthermore, in order to verify the effectiveness of the method of the present invention, the method was applied to a group of college students who were tutoring the course "C language programming" and their learning performance and interview results after using the method were collected. Figure 5 As shown, the horizontal axis represents the knowledge point. At the request of the instructor, the sequential structure and the selection structure are combined into "data type" for testing. The test corresponding to the loop structure is "program and process". The vertical axis represents the answer rate and the average score of the students. For the convenience of display, the average score is presented in the form of percentage. It is found that after students use the method of the present invention for intelligent learning on different knowledge points, they can achieve relatively ideal results. When the knowledge point is more complex (for example, function, pointer), the student's learning enthusiasm is higher and the learning effect is better. The interview results are as follows Figure 6 As shown, the larger the font size, the higher the frequency of keyword occurrence. Most students believe that this method is helpful for learning and is suitable for learning knowledge points. This type of tool provides them with a new way of guiding learning and is easy to use and understand. It has the value of continued use and promotion.

[0112] The embodiment of the present invention also provides a team-based dialogue generation and tutoring learning system based on a large language model, such as Figure 7 As shown, it includes the following modules:

[0113] Login and registration module, used to record student information so that the system can track students' learning progress and needs;

[0114] Expert course knowledge base module, which is used to store knowledge points, teaching resources and learning materials of various subjects so that students can consult them at any time, and is constructed according to course needs and course knowledge structure;

[0115] An autonomous observational learning module is used to generate personalized knowledge dialogue A and personalized knowledge dialogue B according to a generative teacher prompt instruction set and a generative student prompt instruction set, so as to help students understand and master knowledge through observation;

[0116] The autonomous interactive learning module is used to provide teaching sentences C based on the generative interaction initiation prompt instruction set to induce students to participate in knowledge interaction; based on students' feedback and the generative interaction feedback prompt instruction set, it generates personalized feedback sentences D from expert teachers to respond to students' knowledge needs. Students deepen their understanding and application of knowledge through interactive learning.

[0117] The specific implementation methods of the expert course knowledge base module, the autonomous observational learning module and the autonomous interactive learning module are the same as steps 1 to 4, and are not described in the present invention.

[0118] The specific embodiments described herein are merely examples of the spirit of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in similar ways, but they will not deviate from the spirit of the present invention or exceed the scope defined by the appended claims.

Claims

1. A team-based dialogue generation and tutoring learning method based on a large language model, characterized in that: The steps include: Step 1: Build a rule-based expert course knowledge base according to course needs and course knowledge structure; Step 2, enter the observational learning scenario, generate personalized knowledge dialogues A between different teachers and fixed students according to the generative teacher prompt instruction set, and generate personalized knowledge dialogues B between fixed teachers and different students according to the generative student prompt instruction set, wherein the personalized knowledge dialogues A and personalized knowledge dialogues B are used to assist students in understanding knowledge and accompany students in learning knowledge respectively; Step 3, entering the interactive learning situation, initiating a prompt instruction set according to the generative interaction, generating the teaching sentence C of the expert teacher, wherein the teaching sentence C is used to induce students to participate in knowledge interaction; Step 4, enter the interactive learning situation, generate the expert teacher's personalized feedback sentence D based on the student's feedback and the generated interactive feedback prompt instruction set, and use it to respond to the student's knowledge needs; Step 5, repeat steps 2, 3 and 4, optimize personalized knowledge dialogue A, personalized knowledge dialogue B, teaching sentence C and personalized feedback sentence D, until the student no longer repeatedly enters steps 2, 3 and 4, it is considered that the student has mastered the course knowledge, and intelligent learning guidance for the student is achieved.

2. A team-based dialogue generation and tutoring learning method based on a large language model as claimed in claim 1, characterized in that: In step 1, building a rule-based expert course knowledge base includes: Organize course knowledge concepts, key points and cases according to the overall knowledge representation rules; According to the rule of gradual deepening of knowledge from simple to complex, organize the connection between the key and difficult points of knowledge, including sequential relationship and complementary relationship; The course knowledge concepts, key points, cases and the connections between key points are collected to obtain the expert course knowledge base.

3. The team-based dialogue generation and tutoring learning method based on a large language model as claimed in claim 1, characterized in that: The steps for generating personalized knowledge dialogue A and personalized knowledge dialogue B in step 2 are as follows: (21) calling the large language model, configuring the generative teacher prompt instruction set environment and the generative student prompt instruction set environment; (22) Obtain the knowledge content that needs assistance selected by the student and obtain the corresponding knowledge content in the expert course knowledge base; (23) determining a generative teacher prompt instruction set and a generative student prompt instruction set; (24) The generative teacher prompt instruction set environment and the generative student prompt instruction set environment, knowledge content, the generative teacher prompt instruction set and the generative student prompt instruction set are input into the large language model for natural language processing to generate personalized knowledge dialogue A and personalized knowledge dialogue B.

4. The team-based dialogue generation and tutoring learning method based on a large language model as claimed in claim 3, characterized in that: The set of generative teacher prompt instructions includes: (a1) Construct a generative teacher T = {t1, t2, t3} and a randomly generated student s, where t1, t2 and t3 are active teachers, constructive teachers and interactive teachers respectively; (a2) The prompting instructions of the active teacher t1 are to directly tell the randomly generated students what the knowledge point K is, what the principle of the knowledge point is, and how to apply the knowledge point, so as to stimulate the feedback of the randomly generated students; the prompting instructions of the constructive teacher t2 are to tell the randomly generated students the knowledge points, principles and applications through analogies and metaphors, so as to stimulate the randomly generated students to connect the new information with the original knowledge; the prompting instructions of the interactive teacher t3 are to guide the randomly generated students through cases and questions, so as to encourage the randomly generated students to construct their understanding of the knowledge points and participate in classroom activities; (a3) Input the constructed generative teacher T into the large language model to obtain the classroom speech of the generative teacher T; (a4) Determine the prompt instruction P(s) of the random generative student s, where P(s) is to answer the generative teacher T in the role of an ordinary student based on the classroom speech of the generative teacher T; (a5) The constructed generative teacher T = {t1, t2, t3} and the prompt instructions of the randomly generated student s are the generative teacher prompt instruction set.

5. The team-based dialogue generation and tutoring learning method based on a large language model as claimed in claim 3, characterized in that: The set of generative student prompt instructions includes: (b1) construct a random generative teacher t and a generative student S = {s1, s2}, where s1 and s2 are an expert student and a novice student respectively; (b2) Determine the randomly generated teacher t’s prompt instruction P(t), where P(t) is a classroom dialogue based on the knowledge point “pointer” in the role of an expert teacher; (b3) Input the constructed random generative teacher t into the large language model to obtain the classroom speech of the generative teacher T; (b4) Determine the prompt instructions for the generative student S. The prompt instructions for the expert student s1 are to correctly answer and supplement the knowledge points taught by the random generative teacher t in the role of an expert student based on the classroom speech of the random generative teacher t, and to ask questions to the random generative teacher t and apply the knowledge by analogy; the prompt instructions for the novice student s2 are to incorrectly answer and refuse to answer the knowledge points taught by the random generative teacher t in the role of a novice student based on the classroom speech of the random generative teacher t, and to ask questions to the random generative teacher t and seek help; (b5) The prompt instructions of the constructed random generative teacher t and generative student S = {s1, s2} are the generative student prompt instruction set.

6. The team-based dialogue generation and tutoring learning method based on a large language model as claimed in claim 1, characterized in that: The steps for generating teaching sentence C in step 3 are as follows: (31) Calling a large language model and configuring a generative interaction initiation prompt instruction set environment; (32) Obtain the knowledge content that needs assistance selected by the student and obtain the corresponding knowledge content in the expert course knowledge base; (33) Determine a generative interaction initiation prompt instruction set, wherein the prompt instruction set includes {P1(T Q ), P2(T Q )};P1(T Q ) is when students first enter the interactive learning situation, the teacher, in the role of an expert teacher, asks students whether they need learning help; P2 (T Q ) is when students are not in the interactive learning situation for the first time, the teacher, in the role of an expert, asks students application questions and sets cases based on the knowledge content selected by the students, the corresponding knowledge content in the expert course knowledge base, the previous round of teaching sentences C and the students' responses. The questions are in the form of closed questions and open questions, and the difficulty of the questions ranges from simple to difficult; (34) The generative interaction initiation prompt instruction set environment, knowledge content, and the generative interaction initiation prompt instruction set are input into the large language model for natural language processing to generate teaching sentences C.

7. The team-based dialogue generation and tutoring learning method based on a large language model as claimed in claim 1, characterized in that: The steps for generating personalized feedback sentence D in step 4 are as follows: (41) Calling a large language model and configuring a generative interactive feedback prompt instruction set environment; (42) Obtain the knowledge content that needs assistance, teaching sentences C, and students’ responses Sr selected by students; (43) Determine the expert teacher T who can answer questions QA The generative interactive feedback prompt instruction set includes {P1(T A ), P2(T A ), P3(T A ), P4(T A )},P1(T A ) is when the student's reply Sr is non-empty and is related to the knowledge content selected by the student that needs assistance, the teacher, in the role of an expert teacher, gives feedback to the student around the student's reply Sr. The feedback includes judging the correctness of the answer, giving encouragement or praise, and providing detailed explanations and prompts; P2(T A ) is when the student's reply Sr is not empty and has nothing to do with the knowledge content that needs assistance selected by the student, remind the student to answer around the knowledge content that needs assistance selected by the student, or modify the selected knowledge content that needs assistance; P3(T A ) is when the student does not respond within a short period of time, the teacher, in the role of an expert teacher, asks the student if he needs help; P4 (T A ) is when students do not respond for a long time, they exit the interactive learning situation; (44) inputting the generative interactive feedback prompt instruction set environment, knowledge content, and the generative interactive feedback prompt instruction set into the large language model for natural language processing to generate a personalized feedback sentence D; (45) When the generative interactive feedback prompts P2(T A ) is activated, repeat steps (2) to (4) until the student exits the interactive learning situation.

8. A team-based dialogue generation and tutoring learning method based on a large language model as claimed in any one of claims 3 to 7, characterized in that: The large language models include Chatglm-4 and Chatglm-3-turbo.

9. A team-based dialogue generation and tutoring learning system based on a large language model, characterized in that: Includes the following modules: Login and registration module, used to record student information so that the system can track students' learning progress and needs; Expert course knowledge base module, which is used to store knowledge points, teaching resources and learning materials of various subjects so that students can consult them at any time, and is constructed according to course needs and course knowledge structure; An autonomous observational learning module is used to generate personalized knowledge dialogue A and personalized knowledge dialogue B according to a generative teacher prompt instruction set and a generative student prompt instruction set, so as to help students understand and master knowledge through observation; The autonomous interactive learning module is used to provide teaching sentences C based on the generative interaction initiation prompt instruction set to induce students to participate in knowledge interaction; based on students' feedback and the generative interaction feedback prompt instruction set, it generates personalized feedback sentences D from expert teachers to respond to students' knowledge needs. Students deepen their understanding and application of knowledge through interactive learning.

Images