Multi-round dialogue question generation method for computer personalized education

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a multi-round dialogue question generation method for computer personalized education. Comprising the following steps: acquiring textbook and examination outline data, and storing the data in a database; specifying a graph normal form to construct a knowledge graph, extracting knowledge point data in the database data, and constructing a knowledge tree; generating question-answer pairs for the knowledge points by utilizing a Self-QA strategy and combining a Bloom classification method; obtaining a current knowledge point range, generating a question of a current learning progress, performing weight incremental processing according to answer data, and generating a guiding question; aiming at the dialogue data, updating a learning progress; by means of the mode, the next related question or guiding question is generated according to the knowledge point mastering situation, and deep learning and ability improvement of students are facilitated.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method for generating multi-round dialogue questions for computer personalized education. Background Art

[0002] With the rapid development of artificial intelligence technology, personalized education has become an important trend in the field of education, especially in computer education.

[0003] Traditional education models often adopt a one-size-fits-all teaching method, which is difficult to meet the unique learning needs of each student. Multi-round dialogue technology, by simulating human dialogue, can more accurately understand students' learning status and needs, thereby providing customized teaching content.

[0004] However, the existing multi-round dialogue system still has shortcomings in question generation. When generating educational questions, students are usually required to manually input the question range, and the generated questions lack the distinction between students with different levels of understanding, which is not conducive to students' in-depth learning and ability improvement. Summary of the invention

[0005] The purpose of the present invention is to provide a method for generating multi-round dialogue questions for computer personalized education, aiming to solve the technical problems in the prior art that students are required to manually input the question range, and the generated questions lack the distinction between students with different levels of cognition, which is not conducive to students' in-depth learning and ability improvement.

[0006] To achieve the above object, the present invention adopts a method for generating multi-round dialogue questions for computer personalized education, comprising the following steps:

[0007] Obtain teaching materials and examination syllabus data and store them in the database;

[0008] Specify the graph paradigm to construct a knowledge graph, extract knowledge point data from the database data, and construct a knowledge tree;

[0009] Using the Self-QA strategy and combining it with Bloom's taxonomy, we can generate question-answer pairs for knowledge points.

[0010] Get the current knowledge point range, generate questions about the current learning progress, perform weight increment processing based on the answer data, and generate guiding questions;

[0011] Update the learning progress based on the conversation data.

[0012] Among them, in the step of obtaining teaching materials and examination syllabus data and storing them in the database:

[0013] Collect computer professional course textbooks and professional examination syllabus data, and upload the materials to the database for backup.

[0014] Among them, in the steps of constructing a knowledge graph by specifying a graph paradigm, extracting knowledge point data from database data, and constructing a knowledge tree:

[0015] Specify the form of computer professional knowledge graph and divide the knowledge graph nodes into multiple levels;

[0016] Collect knowledge point data in the database data in the form of computer professional knowledge graph, and pre-process the knowledge point data;

[0017] According to the hierarchical relationship between knowledge graph nodes, a knowledge tree is constructed, and the knowledge tree is cross-validated and evaluated.

[0018] Among them, in the step of using the Self-QA strategy and combining Bloom's taxonomy to generate question-answer pairs for knowledge points:

[0019] Design question frames and prompts based on Bloom’s taxonomy;

[0020] Design a small number of examples;

[0021] Based on the designed prompts and a small number of examples, the large model generates questions at multiple cognitive levels corresponding to each knowledge point;

[0022] The knowledge points and questions are input into the big model at the same time. The big model performs the reading comprehension task and generates relevant answers to the questions based on the content of the knowledge points to form question-answer pairs.

[0023] Among them, in the steps of obtaining the current knowledge point range, generating questions about the current learning progress, performing weight increment processing according to the answer data, and generating guiding questions:

[0024] The weight increment process is calculated using the following formula:

[0025]

[0026] Among them, in the steps of obtaining the current knowledge point range, generating questions about the current learning progress, performing weight increment processing according to the answer data, and generating guiding questions:

[0027] Single dialogue answer score i The score range is 0 to 3;

[0028] Weight

[0029] Weight w i Increases as the number of conversation turns increases.

[0030] The present invention provides a method for generating multi-round dialogue questions for personalized computer education, which obtains teaching materials and examination syllabus data and stores them in a database; specifies a graph paradigm to construct a knowledge graph, extracts knowledge point data from database data, and constructs a knowledge tree; utilizes a Self-QA strategy and combines it with Bloom's taxonomy to generate question-answer pairs for knowledge points; obtains the current knowledge point range, generates questions for the current learning progress, performs weight-increasing processing based on answer data, and generates guiding questions; updates the learning progress based on the dialogue data; through the above-mentioned method, the next related question or guiding question is generated based on the knowledge point mastery, which is conducive to students' in-depth learning and ability improvement. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0032] Figure 1 It is a flowchart of the steps of the method for generating multi-round dialogue questions for computer personalized education of the present invention.

[0033] Figure 2 It is a step flow chart of S200 of the present invention.

[0034] Figure 3 It is a step flow chart of S300 of the present invention.

[0035] Figure 4 It is a schematic diagram of the process of S300 of the present invention.

[0036] Figure 5 It is a flow chart of the large model data knowledge enhancement process of the present invention.

[0037] Figure 6 It is a flowchart of the task of generating guiding questions of the present invention.

[0038] Figure 7 It is a structural principle diagram of the multi-round dialogue question generation system for computer personalized education of the present invention.

[0039] Figure 8 It is a structural principle diagram of the electronic equipment of the present invention.

[0040] 601-data acquisition module, 602-knowledge tree construction module, 603-question-answer pair generation module, 604-dialogue generation module, 605-progress update module. DETAILED DESCRIPTION

[0041] Here, exemplary embodiments are described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application.

[0042] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0043] It should be understood that although the terms first, second, third, etc. may be used in the present application to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0044] See also Figure 1 to Figure 6 The present invention provides a method for generating multi-round dialogue questions for computer personalized education, comprising the following steps:

[0045] S100: obtaining teaching materials and examination syllabus data, and storing them in a database;

[0046] In this embodiment, computer professional course teaching materials and professional examination syllabus data are collected and uploaded to a database for backup.

[0047] S200: Specify the graph paradigm to construct a knowledge graph, extract knowledge point data from the database data, and construct a knowledge tree;

[0048] In this embodiment, after the data in the database is collected, a graph paradigm is specified to construct a knowledge graph, extract knowledge point data from the database data, and construct a knowledge tree; the specific process is as follows:

[0049] S201: Specify the form of computer professional knowledge graph and divide the knowledge graph nodes into multiple levels; the levels include subject, chapter, knowledge point range, knowledge point, knowledge point content, and question and answer pairs;

[0050] S202: collecting knowledge point data in the database data in the form of a computer professional knowledge graph, and preprocessing the knowledge point data;

[0051] S203: Construct a knowledge tree based on the hierarchical relationship between knowledge graph nodes, and cross-validate and evaluate the knowledge tree.

[0052] In the above process, the form of computer professional knowledge graph is first specified, and the knowledge graph nodes are divided into multiple levels; the levels include subjects, chapters, knowledge point ranges, knowledge points, knowledge point content, and question-answer pairs; regarding the relationship between nodes, a series of relationship types are defined to construct the semantic connection between these nodes: there is an inclusion relationship between subjects and chapters, chapters and knowledge point ranges are connected through range associations, knowledge point ranges are further associated with detailed knowledge points, and each knowledge point is associated with specific knowledge point content. Then, the knowledge point data in the database data is collected in the form of computer professional knowledge graph, and the knowledge point data is preprocessed. Then, according to the hierarchical relationship between the nodes of the knowledge graph, a knowledge tree is constructed, and the knowledge tree is cross-validated and evaluated; in the subject level to the knowledge point content level of the computer professional knowledge graph, a manual filling method is adopted, and multiple professional courses of the computer major are handed over to several students to summarize all the knowledge points of the corresponding subjects according to the computer major course materials and professional examination outlines, and according to the hierarchical relationship of the nodes, it is constructed into a computer knowledge tree, and cross-validation and course professional teacher evaluation are carried out to ensure the correctness and completeness of the knowledge.

[0053] S300: Using the Self-QA strategy and combining it with Bloom’s taxonomy, we generate question-answer pairs for knowledge points.

[0054] In this implementation, the Self-QA strategy is used in combination with Bloom's taxonomy to generate question-answer pairs for knowledge points. The specific process is as follows:

[0055] S301: Design question frameworks and prompts based on Bloom’s taxonomy;

[0056] S302: Design a small number of examples;

[0057] S303: Based on the designed prompts and a small number of examples, let the large model generate questions of multiple cognitive levels corresponding to each knowledge point;

[0058] S304: The knowledge points and questions are simultaneously input into the big model, and the big model performs the task of reading comprehension and generates relevant answers to the questions according to the content of the knowledge points to form question-answer pairs.

[0059] First, based on Bloom's taxonomy, a question framework is designed; the cognitive process of computer knowledge is divided into six levels: knowledge, understanding, application, analysis, synthesis and evaluation. Each level corresponds to different types of questions to promote learners to think and learn at different cognitive levels. Then, prompts are designed to guide the big model to generate questions; the prompts include the key elements of the question framework template and specific guidance on how to generate questions based on the content of the knowledge point. Then, a small number of examples are used to input the generated questions into the big model; this method enables the big model to learn and imitate these examples to generate more similar questions. Then, based on the designed prompts and a small number of examples, the big model is used to generate questions of multiple cognitive levels for each knowledge point; these questions not only cover the surface information of the knowledge point, but also go deep into deeper understanding and application. Finally, the knowledge point and the question are input into the big model at the same time. The big model performs the task of reading comprehension and generates relevant answers to the questions based on the content of the knowledge point to form question-answer pairs; the question-answer pair data is stored in the graph database after undergoing a strict data verification process to evaluate its logic, relevance and educational value, forming a question-answer pair layer in the knowledge graph.

[0060] Next, the selected base model Qwen2-7b needs to be fine-tuned. The fine-tuning process is divided into three tasks to be completed, namely generating question-answer pairs, evaluating student answers, and generating guiding questions.

[0061] First, for the task of generating question-answer pairs, the function of the large model is to generate questions that match the user's cognitive level based on the user's current level of knowledge. In order to ensure that the generated questions and answers are of high quality and educational value, the knowledge point content text content and question-answer pair data in the knowledge graph are organized in the data format required by the model fine-tuning, and converted into JSON format for storage.

[0062] For the task of evaluating students' answers, the function of the large model is to evaluate students' answers based on questions, reference answers, and scoring criteria, and to provide a basis for scoring. Self-Instruct technology is used to generate fine-tuning data for this function for the model. With the help of the instruction generation capability of the large language model (GPT-4oAPI), the instruction generation requirements are clearly described and some relevant examples are listed to generate instruction data. In the process of specifying instructions, the role-playing method is used to divide the large model into two roles: the teacher model and the student model: the teacher model is responsible for formulating the scoring criteria for each question based on the knowledge point content, questions, and reference answers, while the student model is responsible for generating various possible answers, including correct, missed answers, and wrong answers. Subsequently, the teacher model will classify and score each answer according to these preset scoring criteria, generate corresponding scoring results, and provide a basis for scoring. In order to improve the objectivity and accuracy of scoring, a multiple scoring method is adopted: the same question and student's answer will be scored independently by the model multiple times, and by comparing these scoring results, the score with the highest frequency will be selected as the final score of the student's answer. This method can reduce the impact of accidental errors and improve the stability of scoring. Finally, the corresponding scoring basis is generated based on the determined score to ensure that students can understand the logic behind the scoring results. When the user expresses doubts / dissatisfaction with the current scoring result, he can give feedback on the current scoring result. The model will then backtrack the scoring operation and splice the user feedback content into the input of the evaluation task.

[0063] For the task of generating guiding questions, the function of the large model is to generate guiding questions based on the content of the knowledge points and the scoring basis of the students' answers in the previous round, combined with the Socratic teaching principle. The above-mentioned Self-Instruct technology is used to generate fine-tuning data for this function for the model. With the help of the instruction generation capability of the large language model (GPT-4o API), the instruction generation requirements are clearly described and some relevant examples are listed to generate instruction data.

[0064] In order to achieve efficient interactive learning, Figure 5As shown in the figure, the text knowledge of the knowledge graph data is enhanced, and a multi-task fine-tuning strategy is adopted for the base model. In the generation of question-answer pairs, student answer evaluation, and generation of guiding questions, they are all trained separately to achieve the different stages of their corresponding teaching links. In order to further improve the performance of the model and realize knowledge sharing between tasks, the LORA (Low-Rank Adaptation) fine-tuning technology is used. By inserting a low-rank matrix into the parameters, not only personalized adjustments are provided for each module, but also the weight sharing of the underlying model is maintained, so that the large model can quickly adapt to specific tasks without significantly increasing the number of parameters. After multi-task fine-tuning, a set of optimized weight files are solidified in the generation of question-answer pairs, student answer evaluation, and generation of guiding questions. These files are the results of the large model learning and adaptation on specific tasks. In the actual dialogue process, the weight files of these three modules will be loaded in sequence to perform their respective functions, ensuring the coherence and effectiveness of the entire teaching interaction.

[0065] like Figure 6 As shown in Figure 2, the specific steps for generating guiding question tasks are as follows:

[0066] Step 1: LoRA fine-tuning;

[0067] Step 2: Train the reward model;

[0068] First, the data is sent to the LM model fine-tuned by LoRA. After obtaining its output, human experts are invited to score each output and collect (output sample-score) tuples as data for fine-tuning the reward model. Then, a small-scale end-to-end language model, such as t5-small (60 million parameters), is selected as the reward model, and the (output sample-score) tuple data just collected is used as the training data to fine-tune the reward model with all parameters. The reward model finally trained can give the corresponding score based on the input data-output data.

[0069] Step 3: Reward model fine-tuning.

[0070] Use two LM models fine-tuned by LoRA, one of which has its parameters frozen. Input prompt to the LM model with frozen parameters and the current fine-tuned LM model to obtain output texts y1 and y2, respectively. Pass the output text y2 from the current fine-tuned LM to the reward model to obtain the reward r. Concatenate the KL divergence with the obtained reward r as the loss, and use the PPO algorithm to optimize the gradient. Then calculate the KL divergence between the output texts y1 and y2, denoted as KL(y1||y2). The KL divergence measures the difference between the two probability distributions, so that the output of the fine-tuned model LM does not deviate significantly from the initial model. Combine the reward r and the KL divergence KL(y1||y2) into a composite loss function:

[0071] Loss = r - λ KL(y1||y2);

[0072] Among them, λ is a hyperparameter used to control the weight of the KL divergence term. Finally, the Proximal Policy Optimization (PPO) algorithm is used to optimize and fine-tune the parameters of the LM model. During the optimization process, the calculated loss function is used to guide the gradient update, and the update rules of the PPO algorithm are followed, including importance sampling and policy clipping, to stabilize the training process and improve the efficiency of learning.

[0073] S400: Obtain the current knowledge point range, generate questions about the current learning progress, perform weight increment processing according to the answer data, and generate guiding questions;

[0074] In this implementation, in order to accurately measure the learning progress of students, a weighted scoring mechanism is adopted, in which the scoring weight of each dialogue will gradually increase with the increase of dialogue rounds. Such a scoring strategy ensures that the performance of students in the most recent dialogue stage has a greater impact on the final score. The method calculates the overall performance of students through the following formula. The weighted scoring process is calculated using the following formula:

[0075]

[0076] Among them, the single dialogue answer score is s i The score range is 0 to 3;

[0077] Weight

[0078] Weight w i As the number of conversation turns increases, the importance of the most recent conversation is emphasized. In addition, a mechanism for terminating the conversation early can be used to end the conversation early when a student's response is seriously insufficient, such as significant deviation from the topic, banned word replies, etc. Such a strategy not only dynamically measures the quality of the conversation, but also ensures that only high-quality and meaningful interactions are included in the scoring considerations, providing strong data support for personalized teaching.

[0079] Corresponding to the aforementioned embodiment of the method for generating multi-round dialogue questions for computer personalized education, the present application also provides an embodiment of a system for generating multi-round dialogue questions for computer personalized education.

[0080] Figure 7 1 is a block diagram of a multi-round dialogue question generation system for computer personalized education according to an exemplary embodiment. Figure 7The system may include: a data acquisition module 601, a knowledge tree construction module 602, a question-answer pair generation module 603, a dialogue generation module 604, and a progress update module 605, wherein:

[0081] The data acquisition module 601 is used to acquire teaching materials and examination syllabus data and store them in a database;

[0082] The knowledge tree construction module 602 is used to construct a knowledge graph by specifying a graph paradigm, extract knowledge point data from the database data, and construct a knowledge tree;

[0083] The question-answer pair generation module 603 is used to establish a large model for knowledge point data and generate question-answer pairs;

[0084] The dialogue generation module 604 is used to obtain the current knowledge point range, generate questions about the current learning progress, perform weight increment processing according to the answer data, and generate guiding questions;

[0085] The progress updating module 605 is used to update the learning progress according to the dialogue data.

[0086] In this embodiment, the data acquisition module 601 acquires teaching materials and examination syllabus data, and stores them in the database; the knowledge tree construction module 602 specifies the graph paradigm to construct a knowledge graph, extracts knowledge point data from the database data, and constructs a knowledge tree; the question-answer pair generation module 603 establishes a large model for the knowledge point data and generates question-answer pairs; the dialogue generation module 604 acquires the current knowledge point range, generates questions for the current learning progress, performs weight increasing processing based on the answer data, and generates guiding questions; the progress update module 605 updates the learning progress based on the dialogue data; through the above method, the next related question or guiding question can be generated according to the mastery of the knowledge point, which is conducive to students' in-depth learning and ability improvement.

[0087] Regarding the system in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0088] For the system embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present application. A person of ordinary skill in the art can understand and implement it without creative work.

[0089] Accordingly, the present application also provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned method for generating multi-round dialogue questions for computer personalized education. Figure 8 As shown in FIG. 1 , a hardware structure diagram of a multi-round dialogue question generation system for computer personalized education provided by an embodiment of the present invention is provided in any device with data processing capability, except Figure 8 In addition to the processor, memory and network interface shown, any device with data processing capability in which the apparatus in the embodiment is located may also include other hardware, generally based on the actual functions of the device with data processing capability, which will not be described in detail.

[0090] Accordingly, the present application also provides a computer-readable storage medium on which computer instructions are stored, and when the instructions are executed by a processor, the multi-round dialogue question generation method for computer personalized education as described above is implemented. The computer-readable storage medium can be an internal storage unit of any device with data processing capabilities described in any of the aforementioned embodiments, such as a hard disk or a memory. The computer-readable storage medium can also be an external storage device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), an SD card, a flash card (Flash Card), etc. equipped on the device. Furthermore, the computer-readable storage medium can also include both an internal storage unit of any device with data processing capabilities and an external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capabilities, and can also be used to temporarily store data that has been output or is to be output.

[0091] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the contents disclosed herein. The present application is intended to cover any variations, uses or adaptations of the present application, which follow the general principles of the present application and include common knowledge or customary technical means in the art that are not disclosed in the present application.

[0092] It should be understood that the present application is not limited to the exact construction that has been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof.

Claims

1. A method for generating multi-round dialogue questions for computer personalized education, characterized in that: The steps include: Obtain teaching materials and examination syllabus data and store them in the database; Specify the graph paradigm to construct a knowledge graph, extract knowledge point data from the database data, and construct a knowledge tree; Using the Self-QA strategy and combining it with Bloom's taxonomy, we can generate question-answer pairs for knowledge points. Get the current knowledge point range, generate questions about the current learning progress, perform weight increment processing based on the answer data, and generate guiding questions; Update the learning progress based on the conversation data.

2. The method for generating multi-round dialogue questions for computer personalized education as claimed in claim 1, characterized in that: In the step of obtaining teaching materials and examination syllabus data and storing them in the database: Collect computer professional course textbooks and professional examination syllabus data, and upload the materials to the database for backup.

3. The method for generating multi-round dialogue questions for computer personalized education as claimed in claim 2, characterized in that: In the steps of constructing a knowledge graph by specifying a graph paradigm, extracting knowledge point data from database data, and constructing a knowledge tree: Specify the form of computer professional knowledge graph and divide the knowledge graph nodes into multiple levels; Collect knowledge point data in the database data in the form of computer professional knowledge graph, and pre-process the knowledge point data; According to the hierarchical relationship between knowledge graph nodes, a knowledge tree is constructed, and the knowledge tree is cross-validated and evaluated.

4. The method for generating multi-round dialogue questions for computer personalized education as claimed in claim 3, characterized in that: In the step of generating question-answer pairs for knowledge points using the Self-QA strategy and combining Bloom's taxonomy: Design question frames and prompts based on Bloom’s taxonomy; Design a small number of examples; Based on the designed prompts and a small number of examples, the large model generates questions at multiple cognitive levels corresponding to each knowledge point; The knowledge points and questions are input into the big model at the same time. The big model performs the reading comprehension task and generates relevant answers to the questions based on the content of the knowledge points to form question-answer pairs.

5. The method for generating multi-round dialogue questions for computer personalized education as claimed in claim 4, characterized in that: In the steps of obtaining the current knowledge point range, generating questions about the current learning progress, performing weight increment processing based on the answer data, and generating guiding questions: The weight increment process is calculated using the following formula:

6. The method for generating multi-round dialogue questions for computer personalized education as claimed in claim 5, characterized in that: In the steps of obtaining the current knowledge point range, generating questions about the current learning progress, performing weight increment processing based on the answer data, and generating guiding questions: Single dialogue answer score i The score range is 0 to 3; Weight Weight w i Increases as the number of conversation turns increases.

Images