Education chat robot system based on large language model and teaching recommendation method

Abstract

The invention discloses an education chat robot system based on a large language model and a teaching recommendation method. According to the education chat robot system, different functional modules are designed based on a large language model, and a great contribution is made to application of artificial intelligence in education. Meanwhile, a teaching recommendation method is designed. According to the method, original questionnaire data is converted into cognitive template data, an original interaction graph structure is constructed through user IDs and knowledge point IDs output by an educational chat robot system, node features are updated through a graph neural network, a dynamic interaction graph structure is formed, and the dynamic interaction graph structure is obtained. And for one user node, finding out a knowledge point node which is farthest from the path length of the user node as a knowledge point which needs teaching recommendation, querying corresponding teaching resources in the teaching resource recommendation library, and finally pushing the teaching resources to the user. According to the method, students can be inspired to think innovation, the ability of the students to independently learn and explore problems is cultivated, and accurate teaching recommendation and intervention strategies are provided for the students.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence and prompt engineering, and in particular relates to an educational chat robot system based on a large language model and a teaching recommendation method. Background Art

[0002] With the rapid development of artificial intelligence technology, AI has been widely used in many fields such as education, medical care, and finance. Especially in the field of education, chatbots based on general language models such as the ChatGPT series are becoming increasingly popular. For example, the Mika platform provided by Carnegie Learning uses its chatbot component to provide personalized teaching specifically for mathematics learning. The robot can analyze students' homework and test results in real time, provide targeted feedback and improvement suggestions, and effectively help students identify and fill knowledge gaps. Similarly, the "JillWatson" AI teaching assistant developed by Georgia Institute of Technology based on IBM Watson is also widely used in online courses to answer students' common questions about course content, homework, and projects.

[0003] However, despite the success of existing AI teaching tools in some aspects, they have obvious limitations when using large models for teaching assistance. First, these tools focus on specific learning stages, such as homework help or exam preparation, and lack a comprehensive teaching strategy that covers the entire learning cycle. Second, existing teaching robots mostly exist in the form of assistants, mainly providing information query and question answering services, rather than acting as a "teacher" who can guide and inspire students to think.

[0004] Therefore, there is an urgent need to develop a new type of educational chatbot system that can not only provide support at all stages of learning, from pre-class preparation to in-class interaction to post-class review, but also more deeply promote students' thinking development and improvement of problem-solving ability through heuristic teaching methods from the perspective of teachers. Summary of the invention

[0005] Most of the existing chatbots in the education field focus on tutoring at specific learning stages, such as homework help or exam preparation, but fail to provide comprehensive support covering the entire learning cycle; they mainly appear as assistants and lack the ability to guide and inspire students as teachers. Although existing robots can provide certain personalized feedback based on students' performance, this personalization is usually based on static data analysis and lacks in-depth understanding and dynamic adaptation of students' personality, learning style and cognitive ability. Therefore, existing technologies are insufficient in providing teaching support that truly meets the needs of individual students.

[0006] The purpose of the present invention is to solve the above-mentioned deficiencies in the prior art and to provide an educational chat robot system and a teaching recommendation method based on a large language model.

[0007] In order to achieve the above-mentioned invention object, the present invention specifically adopts the following technical solutions:

[0008] In a first aspect, the present invention provides an educational chat robot system based on a large language model, comprising:

[0009] The base model acquisition module is used to acquire a general large language model and fine-tune it to make it a dedicated base model for the field of programming teaching;

[0010] A thinking prompt module, used for obtaining a first user input and converting it into a designed first prompt word, inputting the question to be thought-prompted and the first prompt word into the exclusive base model, and outputting the thinking prompt by the exclusive base model and displaying it on the user interface of the educational chat robot system;

[0011] A creative prompt module, used for obtaining a second user input and converting it into a designed second prompt word, inputting the topic to be creatively prompted and the second prompt word into the exclusive base model, and outputting the creative prompt by the exclusive base model and displaying it on the user interface;

[0012] A case prompt module, used to obtain cases related to the problem to be solved and display them on the user interface;

[0013] The work evaluation and feedback module is used to intelligently evaluate the works uploaded by users and use pre-built prompt words to guide the exclusive base model to give feedback and display it on the user interface;

[0014] The auxiliary learning module is used to explain concepts that users do not understand and incorrect strategies adopted by the exclusive base model, and to provide assistance for irrelevant concepts raised by users.

[0015] Based on the above scheme, each step can be implemented in the following preferred specific manner.

[0016] As a preferred embodiment of the above-mentioned first aspect, in the base model acquisition module, the specific process of fine-tuning the large language model is: use the LoRA model to fine-tune the large language model, and use pre-designed restrictive words to constrain the output content and language style of the large language model, obtain a pre-built information technology course knowledge base, and use the information technology course knowledge base and the large language model fine-tuned by LoRA as inputs to the large model knowledge chain framework, obtain the large language model with reduced hallucinations and use it as an exclusive base model.

[0017] As a preferred embodiment of the first aspect, in the thinking prompt module, the first prompt word includes the theme, elements and expression form of the user's work when designing the project, and indicates that a dedicated base model is needed to provide thinking prompts.

[0018] As a preferred embodiment of the first aspect, in the creative prompt module, the second prompt word includes the theme of the user's work, the elements included in the work and the form of expression of the work when the user is designing the project, and indicates that a dedicated base model is needed to provide creative prompts.

[0019] As a preferred embodiment of the first aspect, the work evaluation and feedback module includes a self-checklist module, a work evaluation module, a soul-searching module and a work feedback module;

[0020] The self-checklist module is used to obtain a self-checklist about the user's uploaded works and display it on the user interface, wherein the self-checklist includes dimensions for evaluating the user's uploaded works and specific evaluation criteria;

[0021] The work evaluation module is used to obtain the work uploaded by the user and retrieve the evaluation scale, design the third prompt word according to the work uploaded by the user and the evaluation scale, input the third prompt word into the exclusive base model, and the exclusive base model outputs the evaluation of the work uploaded by the user and displays it on the user interface;

[0022] The soul-questioning module is used to input the problem to be solved, the works uploaded by the user, the historical conversation records and the designed fourth prompt words into the exclusive base model, and the exclusive base model outputs the questions about the works uploaded by the user and displays them on the user interface;

[0023] The work feedback module is used to input the problem to be solved, the work uploaded by the user, the historical conversation record and the designed fifth prompt word into the exclusive base model, and the exclusive base model outputs the feedback on the user uploaded work and displays it on the user interface.

[0024] As a preferred embodiment of the above-mentioned first aspect, in the work evaluation module, the elements included in the evaluation scale are the same as the self-checklist elements in the self-checklist module; in the soul-questioning module, the fourth prompt word includes the identity that the exclusive base model needs to play and the language style of the corresponding identity; in the work feedback module, the fifth prompt word includes the identity, language style and expression method that the exclusive base model needs to play.

[0025] As a preferred embodiment of the first aspect, the auxiliary learning module includes a concept explanation module, a strategy explanation module and a help guidance module;

[0026] The concept explanation module is used to input the concepts that the user does not understand, the historical conversation records and the designed sixth prompt words into the exclusive base model, and the exclusive base model outputs the explanation of the concept and displays it on the user interface;

[0027] The strategy explanation module is used to input the wrong strategy adopted by the user, the historical dialogue record and the designed seventh prompt word into the exclusive base model, and the exclusive base model outputs the explanation of the wrong strategy and displays it on the user interface;

[0028] The help guidance module is used to obtain irrelevant concepts raised by users when facing new knowledge during the interaction process, and input the designed eighth prompt word into the exclusive base model, which is then used to lead the topic and output help for users.

[0029] As a preferred embodiment of the above-mentioned first aspect, in the concept explanation module, the sixth prompt word includes the identity that the exclusive base model needs to play, the concept that needs to be explained, and the pre-designed first thinking chain. The first thinking chain decomposes the concept that needs to be explained into four parts: analyzing the involved fields, explaining the concepts for the first time, combining the fields, and explaining the concepts again; in the strategy explanation module, the seventh prompt word includes the identity that the exclusive base model needs to play, the strategy that needs to be explained, and the pre-designed second thinking chain. The second thinking chain decomposes the strategy that needs to be explained into three parts: analyzing the projects that users hope to produce, providing feasible paths, and combining thinking prompts to improve the paths.

[0030] In a second aspect, the present invention provides a teaching recommendation method for an educational chat robot system based on a large language model according to any one of the solutions in the first aspect, comprising the following steps:

[0031] S1. Obtaining original questionnaire data filled out by the user, and converting the original questionnaire data into cognitive template data by a five-point Likert scoring method; wherein the original questionnaire data is data filled out by the user on the cognitive template questionnaire containing different cognitive dimensions;

[0032] S2. Obtain the user ID of the user to be recommended for teaching and the knowledge point ID corresponding to the user's question to be answered from the educational chat robot system, use the user to be recommended for teaching as the user node, use the cognitive template data as the node feature of the user node, use the knowledge point corresponding to the user's question to be answered as the knowledge point node, and use the interaction relationship between the user and the knowledge point as the edge to obtain the original interaction graph structure;

[0033] S3, input the original interaction graph structure into the trained graph neural network, update the node features in the original interaction graph structure, and form a dynamic interaction graph structure;

[0034] S4. Traverse each user node in the dynamic interactive graph structure. For a user node, take the K knowledge point nodes with the longest path length from the user node as the knowledge points that need to be recommended for teaching. Query the teaching resources corresponding to the knowledge points that need to be recommended for teaching in the pre-built teaching resource recommendation library and push the teaching resources to the user in a preset manner.

[0035] As a preferred embodiment of the second aspect mentioned above, the cognitive template questionnaire contains 10 cognitive dimensions, namely students' academic stage, prior knowledge, learning investment, learning style, cognitive load, learning motivation, high-order thinking, human-computer trust, learning method preference and learning attitude.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] The present invention makes full use of the application potential of large language models in the field of education, significantly optimizes the teaching interaction mode, and effectively solves the problem that students passively accept knowledge and lack opportunities for active exploration in traditional teaching. At the same time, by introducing the comprehensive functions of full-process teaching guidance and teacher roles, dynamic personalized adaptation and enhanced interactivity of teaching content are achieved, which greatly stimulates students' active learning willingness and innovative thinking ability, and is conducive to cultivating students' independent thinking and bold innovation capabilities. In addition, the educational chat robot system of the present invention can also adjust teaching strategies and content according to each student's learning progress and feedback, provide a truly personalized learning experience, thereby optimizing learning effects and improving students' overall learning satisfaction.

[0038] The teaching recommendation method of the present invention uses an educational chat robot system to effectively solve the problem of no inspiration and passive acceptance in teaching. Students can explore problems independently under the guidance of a large language model, which is conducive to the cultivation of students' independent thinking and bold innovation capabilities. At the same time, through the construction of a dynamic interactive graph structure, customized information teaching is provided for students, and teaching recommendations and intervention strategies are accurately provided to students. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a module schematic diagram of the educational chat robot system of the present invention;

[0040] Figure 2 Schematic diagram of the module of the work evaluation and feedback module in this embodiment;

[0041] Figure 3 Schematic diagram of the auxiliary learning module in this embodiment;

[0042] Figure 4 A schematic diagram of the change in the learning attitude of students after the educational chat robot system is applied in an embodiment of the present invention;

[0043] Figure 5 Schematic diagram of the steps of the method of the present invention. DETAILED DESCRIPTION

[0044] In order to make the above-mentioned purpose, features and advantages of the present invention more obvious and easy to understand, the specific implementation mode of the present invention is described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below. The technical features in each embodiment of the present invention can be combined accordingly without conflicting with each other.

[0045] In the description of the present invention, it should be understood that the terms "first" and "second" are only used for the purpose of distinguishing descriptions, and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features.

[0046] like Figure 1 As shown, in a preferred implementation of the present invention, the above-mentioned educational chat robot system based on a large language model includes:

[0047] The base model acquisition module is used to acquire a general large language model and fine-tune it to make it a dedicated base model for the field of programming teaching;

[0048] A thinking prompt module, used for obtaining a first user input and converting it into a designed first prompt word, inputting the question to be thought-prompted and the first prompt word into the exclusive base model, and outputting the thinking prompt by the exclusive base model and displaying it on the user interface of the educational chat robot system;

[0049] A creative prompt module, used for obtaining a second user input and converting it into a designed second prompt word, inputting the topic to be creatively prompted and the second prompt word into the exclusive base model, and outputting the creative prompt by the exclusive base model and displaying it on the user interface;

[0050] A case prompt module, used to obtain cases related to the problem to be solved and display them on the user interface;

[0051] The work evaluation and feedback module is used to intelligently evaluate the works uploaded by users and use pre-built prompt words to guide the exclusive base model to give feedback and display it on the user interface;

[0052] The auxiliary learning module is used to explain concepts that users do not understand and incorrect strategies adopted by the exclusive base model, and to provide assistance for irrelevant concepts raised by users.

[0053] It should be noted that in the base model acquisition module, the specific process of fine-tuning the large language model is: use the LoRA model to fine-tune the large language model, and use pre-designed restrictive words to constrain the output content and language style of the large language model, obtain a pre-built information technology course knowledge base, and use the information technology course knowledge base and the large language model fine-tuned by LoRA as inputs to the large model knowledge chain framework, obtain the large language model with reduced hallucinations and use it as an exclusive base model.

[0054] In this embodiment, the educational chat robot system is mainly developed for information technology courses in Zhejiang Province, so the information technology course teaching materials from elementary school to high school are used to fine-tune the large language model commonly used in China, and many restrictive words are designed to limit the output content and language style of the large language model, so as to achieve the purpose of constraining the large language model. In addition to the information technology course teaching materials, this embodiment also obtains relevant knowledge points supplemented by front-line teachers for the scope of information technology courses, and builds an information technology course knowledge base based on these relevant knowledge points as course resources. Then the designed information technology course knowledge base and the large language model after LoRA fine-tuning will be used as the input of the large model knowledge chain (CoK) framework, and the accuracy of reasoning will be improved by multiple examples to obtain a large language model with reduced hallucinations (i.e., an exclusive base model). In this way, the correctness of the exclusive base model in generating answers is further improved.

[0055] It should be noted that, for the problem that users do not know where to start thinking when designing a project, the present invention designs a thinking prompt module. Specifically, the user input (i.e., the first user input) is first obtained through the question template. The user input here specifically refers to the theme of the user's work when designing the project, the elements contained in the work, and the form of expression of the work. After the first user input is converted into the designed first prompt word, it is input into the exclusive base model together with the topic to be thought about, and the output result of the exclusive base model is displayed in the dialog box of the user interface to guide the user to think. Here, the first prompt word input into the exclusive base model needs to include the theme of the user's work when designing the project, the elements contained in the work, and the form of expression of the work, and indicates that the exclusive base model is needed to provide thinking prompts.

[0056] In this embodiment, the designed first prompt word example is: "Please provide an activity prompt with ${Theme} as the theme for users of ${Grade}. Now I want to make an ${output}, which must include ${element} elements. Which aspects should I consider?"

[0057] It should be noted that, in order to solve the problem that users may not be able to think of innovative points when designing projects, the present invention has designed a creative prompt module. Specifically, the user input (i.e., the second user input) is first obtained through a question template. The user input here specifically refers to the theme of the user's work when designing the project, the elements contained in the work, and the form of expression of the work. After the second user input is converted into the designed second prompt word, it is input into the exclusive base model together with the topic to be thought about, and the output result of the exclusive base model is displayed in the dialog box of the user interface to guide users to innovate. Here, the first prompt word input into the exclusive base model needs to include the theme of the user's work when designing the project, the elements contained in the work, and the form of expression of the work, and indicates that the exclusive base model is needed to provide innovative points.

[0058] In this embodiment, the designed second prompt word example is: "Please provide an activity prompt with ${Theme} as the theme for users of ${Grade}. Now I want to make an ${output}, which must contain ${element} elements. What creativity can I bring out in terms of form, method and theme?"

[0059] It should be noted that in the present invention, the case prompt module is used to provide excellent cases related to the problem to be solved for user reference. In this embodiment, taking students as users as an example, these cases can be provided by teachers, so that the cases provided by the teacher will be directly returned in the dialog box. If the teacher does not provide cases, the dedicated base model will generate relevant cases based on the task and the student's input, and guide the student to solve the problem based on the relevant cases.

[0060] It can be seen that based on the above-mentioned exclusive base model, an auxiliary work design function has been developed at the initial stage of project-based learning. Specifically, through the three functional modules of thinking prompt module, creative prompt module and case prompt module, the prompt project is used to provide users with thinking prompts, creative prompts and case prompts. The educational chat robot system can initially guide users to form a preliminary design plan for the work, what kind of cases the work can refer to, from which aspects to think, and what aspects of innovation have been made.

[0061] It should be noted that in the present invention, based on the exclusive base model, an intelligent work evaluation and feedback system, namely, a work evaluation and feedback module, is developed. It can intelligently evaluate the works uploaded by users and use prompt words to guide the exclusive base model to give feedback, which can be positive encouragement or objective criticism.

[0062] In the present invention, Figure 2 As shown, the work evaluation and feedback module includes a self-checklist module, a work evaluation module, a soul-searching module and a work feedback module;

[0063] The self-checklist module is used to obtain a self-checklist about the user's uploaded works and display it on the user interface, wherein the self-checklist includes dimensions for evaluating the user's uploaded works and specific evaluation criteria;

[0064] The work evaluation module is used to obtain the work uploaded by the user and retrieve the evaluation scale, design the third prompt word according to the work uploaded by the user and the evaluation scale, input the third prompt word into the exclusive base model, and the exclusive base model outputs the evaluation of the work uploaded by the user and displays it on the user interface;

[0065] The soul-questioning module is used to input the problem to be solved, the works uploaded by the user, the historical conversation records and the designed fourth prompt words into the exclusive base model, and the exclusive base model outputs the questions about the works uploaded by the user and displays them on the user interface;

[0066] The work feedback module is used to input the problem to be solved, the work uploaded by the user, the historical conversation record and the designed fifth prompt word into the exclusive base model, and the exclusive base model outputs the feedback on the user uploaded work and displays it on the user interface.

[0067] It should be noted that in the present invention, a self-checklist module is developed. Specifically, a self-checklist for user-uploaded works is obtained in advance, and the self-checklist should include the dimensions from which users can judge the works, what the evaluation criteria are, etc. The educational chatbot system will directly return the self-checklist in the dialog box, and users can refer to the self-checklist to evaluate and improve the works themselves.

[0068] It should be noted that in the present invention, a work evaluation module is developed. After the user uploads the work, the educational chat robot system will download the work uploaded by the user, retrieve the evaluation scale at the same time, and then automatically generate prompt words based on the work and the evaluation scale, and then input the prompt words to the exclusive base model, and return the exclusive base model's evaluation of the user's work in the dialog box. It should be noted that the elements included in the evaluation scale are the same as the self-checklist elements in the self-checklist module.

[0069] In this embodiment, an example of a prompt word in the work evaluation module is: "Please evaluate the work ${submit_content} uploaded by the user according to the provided evaluation scale ${evaluate}, score each dimension from 1 to 10, and calculate the total score. Please evaluate the work uploaded by the user according to the provided evaluation scale. Describe the level of this score in the first sentence of the evaluation." It can be seen that the above design not only saves the teacher's time, but also provides an objective and fair evaluation.

[0070] It should be noted that in the present invention, a soul-searching module is developed to enable users to further reflect on the shortcomings of their works. After the problem to be solved, the work uploaded by the user, the historical conversation record and the designed fourth prompt word are input into the exclusive base model, the exclusive base model's questions about the work are returned in the dialog box, guiding the user to further reflect and cultivate the user's independent problem-solving ability and critical thinking. It should be noted that the fourth prompt word contains the identity that the exclusive base model needs to play and the language style of the corresponding identity. The educational chat robot system hopes to encourage users to think in the form of questions, so the exclusive base model is required to return questions.

[0071] In this embodiment, an example of a prompt word in the soul-searching module is: "You are a strict information technology teacher for junior high school students, and you are good at guiding junior high school students to think about and reflect on their learning problems. Please analyze the following students' questions and then generate 1 to 2 questions that will trigger their thinking. The language should be concise and easy to understand, especially considering the cognitive level of junior high school students."

[0072] It should be noted that in the present invention, a work feedback module is developed to increase the user's confidence and enjoyment in the practice process. The exclusive base model is fed with the questions to be solved, the works uploaded by the user, the historical dialogue records and the designed prompt words, and the encouragement of the exclusive base model is returned in the dialog box. It should be noted that the prompt words of the work feedback module include the identity, language style and expression method that the exclusive base model needs to play.

[0073] In this embodiment, the dedicated base model needs to play the role of a kind and amiable teacher and give feedback with affirmative and warm words. An example of a corresponding prompt word is: "You are a caring and amiable junior high school information technology teacher who is good at giving clear encouragement when students raise learning difficulties with you to promote their learning motivation. The language should be concise and easy to understand, especially considering the cognitive level of junior high school students."

[0074] Through the above functions, the educational chatbot system realizes intelligent evaluation of student works, and provides users with warm comfort while taking users' psychology into consideration, while inspiring users to further improve their works.

[0075] In the present invention, Figure 3 As shown, the auxiliary learning module includes a concept explanation module, a strategy explanation module and a help guidance module;

[0076] The concept explanation module is used to input the concepts that the user does not understand, the historical conversation records and the designed sixth prompt words into the exclusive base model, and the exclusive base model outputs the explanation of the concept and displays it on the user interface;

[0077] The strategy explanation module is used to input the wrong strategy adopted by the user, the historical dialogue record and the designed seventh prompt word into the exclusive base model, and the exclusive base model outputs the explanation of the wrong strategy and displays it on the user interface;

[0078] The help guidance module is used to obtain irrelevant concepts raised by users when facing new knowledge during the interaction process, and input the designed eighth prompt word into the exclusive base model, which is then used to lead the topic and output help for users.

[0079] It should be noted that in the present invention, a concept explanation module is developed to explain concepts that users do not understand. In addition to basic guiding words, such as the role that the exclusive base model needs to play (strict teacher) and the problem that needs to be solved (explaining the concept), the prompt words in the concept explanation module also need to include a pre-designed thinking chain. For the concept explanation problem, it should be combined with the specific application field, so the concept explanation problem should be decomposed into: analyzing the involved fields, explaining the input concepts, and explaining again in combination with the fields and concept explanations. Finally, the exclusive base model will analyze the concepts that the user does not understand from the historical conversation records and return the explanation of the concepts in the dialog box. It should be noted that in order to improve the accuracy of reasoning, the exclusive base model should generate explanations of several fields, and then select the most relevant results from them. Considering the user's cognitive level, the exclusive base model should be explained in a way that is as simple and easy to understand as possible.

[0080] In this embodiment, an example of a prompt word in the concept explanation module is: "You are a strict information technology teacher for junior high school students. You are good at explaining concepts to students in a simple and easy-to-understand way, and then generate 1-2 related concepts for them to understand. The language is required to be concise and easy to understand, especially considering the cognitive level of junior high school students. Answer in a teacher's tone and style. You need six steps to perform this task: 1) Analyze the knowledge concepts involved in the student input, and please give 3 analysis results. 2) Select a more accurate one from the results of step 1). 3) Use 1-2 sentences to explain the field selected in 2). 4) Explain the concept input by the student. 5) Analyze 3) Another concept related to 4), please give 5 analysis results. 6) Select the 2 most relevant ones from 5). "

[0081] It should be noted that in the present invention, a strategy explanation module has been developed to address the problem of unreasonable strategy selection when users conduct project-based learning. The educational chatbot system can formulate a more complete and feasible path based on the historical dialogue records and the user's previous implementation path. In addition to basic guiding words, such as the identity that the exclusive base model needs to play (strict teacher) and the problem that needs to be solved (explanation strategy), the prompt words in the strategy explanation module also need to include a pre-designed thinking chain. For the problem of strategy explanation, it should be improved by the user's own implementation strategy, so the strategy explanation problem should be decomposed into: analyzing the projects that the user hopes to produce, providing feasible paths, and combining thinking prompts to improve the path. Finally, the exclusive base model returns the explanation of the strategy in the dialog box. It should be noted that in order to improve the accuracy of reasoning, the exclusive base model should generate several more projects and implementation paths, and then select the most relevant results from them. At the same time, it is necessary to let the exclusive base model consider the limitations of the user's current academic stage cognitive level and give concise answers.

[0082] In this embodiment, an example of a prompt word in the strategy explanation module is: "You are a strict information technology teacher for junior high school students, and you are good at proposing project-based learning strategies for students. The language should be concise and easy to understand, especially considering the cognitive level and knowledge reserves of junior high school students. Answer in a teacher's tone and style. 1) Analyze what kind of projects can be produced by the student's input and give 3 analysis results. 2) Select a more accurate one from the results of step 1). 3) Use 1-2 sentences to explain the project selected in 2). 4) Provide a feasible path for the project for the student's input. Please give 3 paths. 5) Choose the simplest one from 4). 6) Analyze how to use thinking prompts, creative prompts, case prompts, and self-checklists to assist the feasible path."

[0083] It should be noted that in the present invention, in order to solve the problem that users may gradually deviate from the topic during the interaction process, a help guidance module is developed to input designed prompt words into the exclusive base model. The prompt words should include the identity that the exclusive base model needs to play (strict but kind teacher) and the task that needs to be completed (dominant topic). An example of a corresponding prompt word is: 'You are a strict but kind information technology teacher for junior high school students. You are good at associating irrelevant concepts raised by students with... and shifting the topic to... The language should be concise and easy to understand, especially considering the cognitive level and knowledge reserves of junior high school students. Answer in a teacher's tone and style. The following are irrelevant concepts input by students: 'Provide help to students as an assistant teacher. "

[0084] Through the above three modules, the educational chatbot system solves problems such as the lack of help for users when faced with new knowledge and some unreasonable mistakes that may occur when designing implementation paths during project-based learning.

[0085] In this embodiment, the educational chatbot system adopts a front-end and back-end separation development model, decoupling the front-end project from the back-end project, designing and developing them separately, and realizing the interaction between the front-end and back-end through the RESTful API interface. The front-end development uses Node.js to build and manage dependencies. The back-end development uses Python and adopts the FastAPI framework to build efficient Web API services.

[0086] In order to better demonstrate the specific implementation and technical effects of the present invention, the educational chat robot system based on the large language model shown in the above preferred implementation method is applied to a specific example.

[0087] Example

[0088] The specific modules included in the educational chat robot system based on the large language model used in this embodiment are as described above and will not be repeated here.

[0089] The educational chat robot system in this embodiment has been put into use in many primary and secondary schools, with the number of registered users reaching 560, and a total of 2,236 backend interaction data have been collected.

[0090] In order to demonstrate the application effect of this embodiment, the present invention analyzes the cognitive template data of 60 students in a middle school before and after the semester. Figure 4 shows the changes in these students’ learning attitudes, and Table 1 shows the changes in these students’ computational thinking.

[0091] Table 1. Changes in students' thinking

[0092] abstract break down Algorithmic Thinking evaluate Generalization Mean before semester 3.72 3.75 3.79 3.72 3.70 3.74 After the semester 3.82 3.77 3.86 3.81 3.74 3.79 Change 0.10 0.02 0.05 0.09 0.04 0.05

[0093] from Figure 4 As shown in Table 1, after one semester of use, students' learning attitudes have become more positive and their computational thinking has also improved. It can be seen that the educational chat robot system of the present invention can make learners enjoy the learning process more and exercise students' thinking ability.

[0094] Similarly, based on the same inventive concept, another preferred embodiment of the present invention also provides a teaching recommendation method corresponding to the educational chat robot system based on the large language model provided in the above embodiment, such as Figure 5 As shown, the following steps are included:

[0095] S1. Obtaining original questionnaire data filled out by the user, and converting the original questionnaire data into cognitive template data by a five-point Likert scoring method; wherein the original questionnaire data is data filled out by the user on the cognitive template questionnaire containing different cognitive dimensions;

[0096] S2. Obtain the user ID of the user to be recommended for teaching and the knowledge point ID corresponding to the user's question to be answered from the educational chat robot system, use the user to be recommended for teaching as the user node, use the cognitive template data as the node feature of the user node, use the knowledge point corresponding to the user's question to be answered as the knowledge point node, and use the interaction relationship between the user and the knowledge point as the edge to obtain the original interaction graph structure;

[0097] S3, input the original interaction graph structure into the trained graph neural network, update the node features in the original interaction graph structure, and form a dynamic interaction graph structure;

[0098] S4. Traverse each user node in the dynamic interactive graph structure. For a user node, take the K knowledge point nodes with the longest path length from the user node as the knowledge points that need to be recommended for teaching. Query the teaching resources corresponding to the knowledge points that need to be recommended for teaching in the pre-built teaching resource recommendation library and push the teaching resources to the user in a preset manner.

[0099] It should be noted that in step S1, the cognitive template questionnaire contains 10 cognitive dimensions, namely, student stage, prior knowledge, learning investment, learning style, cognitive load, learning motivation, high-order thinking, human-computer trust, learning method preference, and learning attitude.

[0100] The above-described embodiment is only a preferred solution of the present invention, but it is not intended to limit the present invention. A person skilled in the relevant technical field may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, any technical solution obtained by equivalent replacement or equivalent transformation falls within the protection scope of the present invention.

Claims

1. An educational chat robot system based on a large language model, characterized in that: include: The base model acquisition module is used to acquire a general large language model and fine-tune it to make it a dedicated base model for the field of programming teaching; A thinking prompt module, used for obtaining a first user input and converting it into a designed first prompt word, inputting the question to be thought-prompted and the first prompt word into the exclusive base model, and outputting the thinking prompt by the exclusive base model and displaying it on the user interface of the educational chat robot system; A creative prompt module, used for obtaining a second user input and converting it into a designed second prompt word, inputting the topic to be creatively prompted and the second prompt word into the exclusive base model, and outputting the creative prompt by the exclusive base model and displaying it on the user interface; A case prompt module, used to obtain cases related to the problem to be solved and display them on the user interface; The work evaluation and feedback module is used to intelligently evaluate the works uploaded by users and use pre-built prompt words to guide the exclusive base model to give feedback and display it on the user interface; The auxiliary learning module is used to explain concepts that users do not understand and incorrect strategies adopted by the exclusive base model, and to provide assistance for irrelevant concepts raised by users.

2. The educational chat robot system based on a large language model as described in claim 1, characterized in that: In the base model acquisition module, the specific process of fine-tuning the large language model is: use the LoRA model to fine-tune the large language model, and use pre-designed restrictive words to constrain the output content and language style of the large language model, obtain a pre-built information technology course knowledge base, and use the information technology course knowledge base and the large language model fine-tuned by LoRA as inputs to the large model knowledge chain framework, obtain the large language model with reduced hallucinations and use it as an exclusive base model.

3. The educational chat robot system based on a large language model as claimed in claim 1, characterized in that: In the thinking prompt module, the first prompt word includes the theme, elements and forms of expression of the user's work when designing the project, and points out that a dedicated base model is needed to provide thinking prompts.

4. The educational chat robot system based on a large language model as claimed in claim 1, characterized in that: In the creative prompt module, the second prompt word includes the theme of the user's work, the elements included in the work, and the form of expression of the work when the user is designing the project, and points out that a dedicated base model is needed to provide creative prompts.

5. The educational chat robot system based on a large language model as claimed in claim 1, characterized in that: The work evaluation and feedback module includes a self-checklist module, a work evaluation module, a soul-searching module and a work feedback module; The self-checklist module is used to obtain a self-checklist about the user's uploaded works and display it on the user interface, wherein the self-checklist includes dimensions for evaluating the user's uploaded works and specific evaluation criteria; The work evaluation module is used to obtain the work uploaded by the user and retrieve the evaluation scale, design the third prompt word according to the work uploaded by the user and the evaluation scale, input the third prompt word into the exclusive base model, and the exclusive base model outputs the evaluation of the work uploaded by the user and displays it on the user interface; The soul-questioning module is used to input the problem to be solved, the works uploaded by the user, the historical conversation records and the designed fourth prompt words into the exclusive base model, and the exclusive base model outputs the questions about the works uploaded by the user and displays them on the user interface; The work feedback module is used to input the problem to be solved, the work uploaded by the user, the historical conversation record and the designed fifth prompt word into the exclusive base model, and the exclusive base model outputs the feedback on the user uploaded work and displays it on the user interface.

6. The educational chat robot system based on a large language model as claimed in claim 5, characterized in that: In the work evaluation module, the elements included in the evaluation scale are the same as the self-checklist elements in the self-checklist module; in the soul-questioning module, the fourth prompt includes the identity that the exclusive base model needs to play and the language style of the corresponding identity; in the work feedback module, the fifth prompt includes the identity, language style and expression method that the exclusive base model needs to play.

7. The educational chat robot system based on a large language model as claimed in claim 1, characterized in that: The auxiliary learning module includes a concept explanation module, a strategy explanation module and a help guidance module; The concept explanation module is used to input the concepts that the user does not understand, the historical conversation records and the designed sixth prompt words into the exclusive base model, and the exclusive base model outputs the explanation of the concept and displays it on the user interface; The strategy explanation module is used to input the wrong strategy adopted by the user, the historical dialogue record and the designed seventh prompt word into the exclusive base model, and the exclusive base model outputs the explanation of the wrong strategy and displays it on the user interface; The help guidance module is used to obtain irrelevant concepts raised by users when facing new knowledge during the interaction process, and input the designed eighth prompt word into the exclusive base model, which is then used to lead the topic and output help for users.

8. The educational chat robot system based on a large language model as claimed in claim 7, characterized in that: In the concept explanation module, the sixth prompt word includes the role that the exclusive base model needs to play, the concept that needs to be explained, and the pre-designed first thinking chain. The first thinking chain decomposes the concept that needs to be explained into four parts: analyzing the involved fields, explaining the concepts for the first time, combining the fields, and explaining the concepts again; in the strategy explanation module, the seventh prompt word includes the role that the exclusive base model needs to play, the strategy that needs to be explained, and the pre-designed second thinking chain. The second thinking chain decomposes the strategy that needs to be explained into three parts: analyzing the projects that users hope to produce, providing feasible paths, and combining thinking prompts to improve the paths.

9. A teaching recommendation method based on the educational chat robot system according to any one of claims 1 to 8, characterized in that: The following steps are involved: S1. Obtaining original questionnaire data filled out by the user, and converting the original questionnaire data into cognitive template data by a five-point Likert scoring method; wherein the original questionnaire data is data filled out by the user on the cognitive template questionnaire containing different cognitive dimensions; S2. Obtain the user ID of the user to be recommended for teaching and the knowledge point ID corresponding to the user's question to be answered from the educational chat robot system, use the user to be recommended for teaching as the user node, use the cognitive template data as the node feature of the user node, use the knowledge point corresponding to the user's question to be answered as the knowledge point node, and use the interaction relationship between the user and the knowledge point as the edge to obtain the original interaction graph structure; S3, input the original interaction graph structure into the trained graph neural network, update the node features in the original interaction graph structure, and form a dynamic interaction graph structure; S4. Traverse each user node in the dynamic interactive graph structure. For a user node, take the K knowledge point nodes with the longest path length from the user node as the knowledge points that need to be recommended for teaching. Query the teaching resources corresponding to the knowledge points that need to be recommended for teaching in the pre-built teaching resource recommendation library and push the teaching resources to the user in a preset manner.

10. The teaching recommendation method according to claim 9, characterized in that: The cognitive template questionnaire contains 10 cognitive dimensions, namely, students' academic stage, prior knowledge, learning involvement, learning style, cognitive load, learning motivation, higher-order thinking, human-computer trust, learning method preference, and learning attitude.

Images