Learning condition insight and student growth analysis method and system based on large language model

Abstract

The invention relates to the field of education technology and artificial intelligence cross technology, in particular to a learning condition insight and student growth analysis method and system based on a large language model. The learning condition insight method comprises the following specific steps: acquiring original learning condition data of students, and converting the original learning condition data into standardized numerical data; the method comprises the following steps: receiving a teaching document uploaded by a user, performing semantic analysis to extract keywords and generate abstracts, and constructing an intelligent teaching resource library; teaching business problems are proposed, teaching documents corresponding to the teaching business problems are retrieved, and structured cue words are constructed; and inputting the structured cue word into a large language model, and obtaining an evaluation report of learning condition insight and operable teaching suggestions based on the evaluation report. By adopting the method, automatic intelligent learning condition insight and analysis can be carried out for multi-dimensional and quantifiable learning condition data, and the method has the characteristic of supporting full-process teaching closed-loop evaluation.

Description

Technical Field

[0001] This application relates to the interdisciplinary fields of educational technology and artificial intelligence, and in particular to methods and systems for learning insights and student growth analysis based on large language models. Background Technology

[0002] With the rapid development of educational informatization and artificial intelligence technologies, precision teaching and personalized education have become important directions for higher education reform.

[0003] However, existing learning analysis and teaching support methods typically suffer from the following significant technical bottlenecks and shortcomings: The ability to quantify and visualize learning data in multiple dimensions is insufficient: Traditional learning analysis relies heavily on teachers' classroom observations, individual conversations, or simple two-dimensional charts. These methods struggle to comprehensively quantify and assess multiple key dimensions such as "knowledge base," "learning interest," and "learning habits" simultaneously. When faced with massive amounts of class data, teachers often find it difficult to intuitively "see at a glance" the overall distribution characteristics of the class and individual differences.

[0004] The application of large language models in vertical education scenarios lacks context awareness: Although general generative artificial intelligence has powerful text generation capabilities, when it is directly applied to teaching preparation or Q&A, it often lacks context information specific to the scenario.

[0005] Existing lesson preparation tools struggle to automatically perform semantic analysis, tagging, and automatic classification of these unstructured resources, resulting in these resources being unable to be effectively accessed and activated by intelligent systems during lesson preparation and the generation of teaching strategies.

[0006] Lack of data-driven, full-process teaching closed-loop assessment: Current teaching evaluations are often lagging behind and focus more on outcome-based assessments, making it difficult for teachers to form a scientific teaching closed loop of "diagnosis-intervention-assessment-improvement".

[0007] Therefore, how to construct an automatic and intelligent method and system for learning insight and student growth analysis based on multi-dimensional and quantifiable learning data is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0008] Therefore, it is necessary to provide a method and system for learning insight and student growth analysis based on a large language model to address the above-mentioned technical problems. This system has the feature of supporting closed-loop evaluation of the entire teaching process.

[0009] To achieve the above-mentioned objectives of this invention, the technical solution adopted is as follows: A learning insight method based on a large language model includes the following specific steps: Obtain students' raw learning data and convert it into multidimensional standardized numerical data through a pre-set scoring model; It receives teaching documents uploaded by users, performs semantic analysis to extract keywords and generate summaries, and builds an intelligent teaching resource library with semantic retrieval capabilities. The system identifies teaching business issues, retrieves corresponding teaching documents from the intelligent teaching resource database, and constructs structured prompts. Call the large language model interface, input the structured prompt words into the large language model, and obtain an assessment report on learning insights and actionable teaching suggestions based on it.

[0010] Preferably, the structured prompts are composed of text instructions that include teaching business questions, teaching resource summaries, standardized numerical data, lesson preparation notes, and case studies, which are logically concatenated.

[0011] Furthermore, the scoring model is specifically a large language model limited by preset scoring rule prompts. The standardized numerical data includes three dimensions: learning interest, knowledge base, and learning habits. After being converted into multidimensional standardized numerical data, the standardized numerical data is also mapped into scatter points in multidimensional space to generate an interactive multidimensional learning distribution map, where each scatter point represents a student, allowing users to intuitively identify teaching business problems. The specific steps for generating an interactive multidimensional learning distribution map are as follows: constructing a three-dimensional coordinate system, where the X-axis, Y-axis, and Z-axis correspond to learning interests, knowledge base, and learning habits, respectively; the color of the scatter points corresponds to the numerical value of the learning interests through a color mapping algorithm; the interactive multidimensional learning distribution map is configured such that when the cursor hovers over a scatter point, it displays the student's identity information and specific dimension score corresponding to that scatter point.

[0012] Furthermore, the method for constructing the teaching resource database is as follows: Receive teaching documents uploaded by users and extract the text content; The large language model is invoked to perform semantic analysis on the text content, generating a keyword set; Set up a classification rule base to automatically categorize teaching documents into at least one of the following categories: case library, discussion library, exercise library, feedback library, or gain library, based on the keyword set. Generate a content summary of the teaching documents; combine the content summary, keywords, and document classification to construct an intelligent teaching resource library with semantic retrieval capabilities.

[0013] A student growth analysis method based on the aforementioned learning insight method includes the following specific steps: The learning insight method described above generates an assessment report on the learning insights of the target student group and actionable teaching suggestions based on it. Based on the aforementioned assessment report and its actionable teaching recommendations, teaching interventions were conducted for the target student group. An assessment report on the learning situation of the target student group after teaching intervention is generated using the aforementioned learning insight method. By comparing the data in the pre-test assessment reports of the target student group before the teaching intervention with the data in the post-test assessment reports after the teaching intervention, the large language model is used to generate a teaching effectiveness evaluation report including attribution analysis and further actionable teaching suggestions.

[0014] Preferably, by comparing the data in the pre-intervention assessment reports of the target student group before the teaching intervention with the data in the post-intervention assessment reports after the teaching intervention, a large language model is used to generate a teaching effectiveness evaluation report including attribution analysis and further actionable teaching suggestions. The specific steps are as follows: Calculate the difference between the mean values ​​of pre-test and post-test data on multidimensional standardized numerical data; Draw a multi-dimensional comparison chart showing the spatial changes of the data before and after the test; Input the average difference, multidimensional positional variation characteristics, and preset attribution analysis instructions into the large language model; Obtain the textual descriptions of the reasons for the rise and fall of indicators and improvement strategies from the output of the large language model.

[0015] A learning insight system based on a large language model includes: The core analysis engine includes a data loading and verification submodule and an AI-powered intelligent question answering submodule. The data loading and verification submodule is configured to acquire students' original learning data and convert it into multidimensional standardized numerical data through a preset scoring model. The AI ​​intelligent question-answering submodule is configured to call the large language model interface, input the structured prompt words into the large language model, and obtain an assessment report on learning insights and actionable teaching suggestions based on it; The teaching resource management module includes a teaching resource database construction submodule and a prompt word project submodule; The teaching resource library construction submodule is configured to receive teaching documents uploaded by users, perform semantic analysis to extract keywords and generate summaries, and build an intelligent teaching resource library with semantic retrieval capabilities; and receive teaching business questions input by users. The prompt word engineering submodule is configured to raise teaching business questions, retrieve teaching documents corresponding to the teaching business questions from the intelligent teaching resource library, and construct structured prompt words.

[0016] Preferably, the core analysis engine further includes a visualization generation submodule; the visualization generation submodule is used to map the standardized numerical data into scatter points in a multidimensional space to generate an interactive multidimensional student learning distribution map, where each scatter point represents a student, allowing users to intuitively identify teaching business problems.

[0017] Furthermore, it also includes a unified data manager, session management, and key scheduler; The data unification manager is responsible for the persistence of all data, maintaining user, chat history, lesson preparation notes and cases, teaching resource library and API key data table, ensuring data consistency and security; The aforementioned session management and key scheduler are responsible for the lifecycle management of user sessions and load balancing of multiple API keys, ensuring the stable operation of the system in a multi-user environment.

[0018] Furthermore, the system adopts a Python-based B / S architecture, with the front end using the Grado framework to build an interactive web interface, the back end using MySQL for data persistence, and intelligent services provided through the integration of a large language model API.

[0019] The beneficial effects of this invention are as follows: This invention aims to overcome the above-mentioned defects and solve the following core technical problems: This invention discloses a learning insight method based on a large language model. By acquiring and standardizing students' raw learning data, it achieves a precise characterization of teaching-related issues concerning students' ideological and political literacy and learning status. This invention constructs structured prompts by matching teaching-related issues, teaching documents, lesson preparation notes, and case studies, and calls the large language model interface, deeply integrating the powerful analytical and generative capabilities of the large language model with specific educational business scenarios. The method organically integrates learning analysis, resource management, and lesson preparation data, while simultaneously outputting evaluation opinions and improvement methods, achieving seamless connection between data flow and business flow, forming a closed-loop teaching system. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating a learning insight method based on a large language model in one embodiment; Figure 2 This is a flowchart illustrating the growth analysis method in one embodiment. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0022] Example 1 like Figure 1 As shown, a learning insight method based on a large language model includes the following specific steps: Obtain students' raw learning data and convert it into multidimensional standardized numerical data through a pre-set scoring model; It receives teaching documents uploaded by users, performs semantic analysis to extract keywords and generate summaries, and builds an intelligent teaching resource library with semantic retrieval capabilities. The system proposes a teaching business question, retrieves corresponding teaching documents from the intelligent teaching resource library, and constructs structured prompt words. The structured prompt words are composed of text instructions that include the teaching business question, the matching teaching resource summary, standardized numerical data, and lesson preparation notes and case studies, which are logically concatenated. Call the large language model interface, input the structured prompt words into the large language model, and obtain an assessment report on learning insights and actionable teaching suggestions based on it; Obtain pre-test data and post-test data of the same student group before and after the teaching intervention, compare and calculate the changes in each dimension of indicators, and call the large language model to generate a teaching effectiveness evaluation report including attribution analysis.

[0023] In this embodiment, obtaining students' original learning data specifically involves: distributing questionnaires based on a 3D model (3 questions per dimension, as described in the test instructions) through the Questionnaire Star platform to collect student data.

[0024] In this embodiment, the conversion of the original answer sheet data into multidimensional standardized numerical data using a preset scoring model is specifically as follows: a large language model limited by preset scoring rule prompts is used to assist in determining the scoring coefficients, and the original answer sheet data is converted into a CSV file containing three standardized scores (0-10 points) for "learning interest", "knowledge base" and "learning habits".

[0025] In one specific embodiment, the standardized numerical data includes three dimensions: learning interest, knowledge base, and learning habits.

[0026] In one specific embodiment, after being converted into multidimensional standardized numerical data, the standardized numerical data is also mapped into scatter points in multidimensional space to generate an interactive multidimensional learning distribution map, where each scatter point represents a student, allowing users to intuitively identify teaching business problems. The specific steps for generating an interactive multidimensional learning distribution map are as follows: constructing a three-dimensional coordinate system, where the X-axis, Y-axis, and Z-axis correspond to learning interests, knowledge base, and learning habits, respectively; the color of the scatter points corresponds to the numerical value of the learning interests through a color mapping algorithm; the interactive multidimensional learning distribution map is configured such that when the cursor hovers over a scatter point, it displays the student's identity information and specific dimension score corresponding to that scatter point.

[0027] In this embodiment, ScoreAnalyzer is called to generate a 3D learning distribution map.

[0028] In one specific embodiment, the method for constructing the teaching resource database is as follows: Receive teaching documents uploaded by users and extract the text content; The large language model is invoked to perform semantic analysis on the text content, generating a keyword set; Set up a classification rule base to automatically categorize teaching documents into at least one of the following categories: case library, discussion library, exercise library, feedback library, or gain library, based on the keyword set. Generate a content summary of the teaching documents; combine the content summary, keywords, and document classification to construct an intelligent teaching resource library with semantic retrieval capabilities.

[0029] In this embodiment, the teaching business problem raised by the user is "designing teaching activities to enhance 'cultural confidence' for e-commerce classes"; the ScoreAnalyzer.query_with_libraries method is used to retrieve relevant cases and the teacher's own lesson preparation notes and cases from the intelligent teaching resource library corresponding to the teaching business problem; When constructing structured prompts, the learning data, including standardized numerical data on students' preferences for "visual tools" and "film and television documentaries", searched resources, and lesson preparation needs, are combined into structured prompts. The large language model is then invoked to return personalized and operable teaching design schemes.

[0030] In this embodiment, the scoring range for the three core dimensions "learning interest," "knowledge base," and "learning habits" is 0-10. When calling the large language model API, temperature=0.3 is set to ensure the stability of the responses, and max_tokens=2000 to ensure the comprehensiveness of the analysis report. In the 3D scatter plot, the data point size is fixed at 8, the transparency is 0.8, and the color mapping uses the Viridis scheme for easy differentiation. An optimized category keyword library is used; for example, content containing words such as "case" or "example" will be prioritized for inclusion in the "case library."

[0031] Example 2 like Figure 2As shown, a student growth analysis method based on the aforementioned learning insight method includes the following specific steps: The learning insight method described above generates an assessment report on the learning insights of the target student group and actionable teaching suggestions based on it. Based on the aforementioned assessment report and its actionable teaching recommendations, teaching interventions were conducted for the target student group. An assessment report on the learning situation of the target student group after teaching intervention is generated using the aforementioned learning insight method. By comparing the data in the pre-test assessment reports of the target student group before the teaching intervention with the data in the post-test assessment reports after the teaching intervention, the large language model is used to generate a teaching effectiveness evaluation report including attribution analysis and further actionable teaching suggestions.

[0032] In one specific embodiment, by comparing the data from the pre-intervention assessment reports of the target student group before the teaching intervention with the data from the post-intervention assessment reports after the teaching intervention, a large language model is invoked to generate a teaching effectiveness evaluation report containing attribution analysis and further actionable teaching suggestions. The specific steps are as follows: Calculate the difference between the mean values ​​of pre-test and post-test data on multidimensional standardized numerical data; Draw a multi-dimensional comparison chart showing the spatial changes of the data before and after the test; Input the average difference, multidimensional positional variation characteristics, and preset attribution analysis instructions into the large language model; Obtain the textual descriptions of the reasons for the rise and fall of indicators and improvement strategies from the output of the large language model.

[0033] In this embodiment, teachers upload pre-test and post-test data files before and after the teaching intervention, respectively.

[0034] Generates 3D comparison charts of pre-test and post-test data based on pre-test and post-test data files, supporting individual or overlay display, intuitively showing the "movement trajectory" of the student group.

[0035] Call the ScoreAnalyzer.analyze_teaching_effect method to generate a detailed quantitative analysis report (such as the report in the test description), clearly pointing out the changes in each indicator, possible causes, and improvement suggestions.

[0036] Example 3 A learning insight system based on a large language model includes: The core analysis engine includes a data loading and verification submodule and an AI-powered intelligent question answering submodule. The data loading and verification submodule is configured to acquire students' original learning data and convert it into multidimensional standardized numerical data through a preset scoring model. In this embodiment, the loading and verification submodule loads a CSV file containing three columns of data: “Learning Interests”, “Knowledge Base”, and “Learning Habits” (such as the “Third Topic Questionnaire” data you provided) using the load_csv and load_test_data methods, and performs integrity verification.

[0037] The AI ​​intelligent question-answering submodule is configured to call the large language model interface, input the structured prompt words into the large language model, and obtain an assessment report on learning insights and actionable teaching suggestions based on it; In this embodiment, the AI ​​intelligent question-answering submodule achieves a core technological breakthrough through the `query_with_libraries` method. The workflow of this method is as follows: a. Context building: Retrieve the current user's lesson preparation notes from the database (via DatabaseManager) and search for resources related to the question from the teaching resource library (via TeachingLibraryManager).

[0038] b. Prompt Engineering: Dynamically combine student data summaries, lesson preparation notes, relevant teaching resources, and the user's original question into a structured prompt. For example, for the question "How to improve students' learning interest," the system will automatically incorporate relevant film and television resources from the case library and the teacher's previous successful experiences.

[0039] c. Call and return: Call the large language model API to generate a comprehensive analysis report and actionable teaching suggestions that combine data, resources and expertise (such as the detailed teaching strategy report generated by AI in the test instructions).

[0040] In this embodiment, the core analysis engine also includes a visualization generation submodule. This submodule uses the `generate_3d_plot` and `generate_effect_plot` methods, leveraging the Plotly library, to generate interactive 3D scatter plots. Each point in the plot represents a student, with its X, Y, and Z coordinates corresponding to "learning interest," "knowledge base," and "learning habits," respectively. Colors are mapped to the interest dimension, and hovering over a point displays detailed information (such as student ID and class). This allows teachers to "see at a glance" the overall class distribution and individual differences.

[0041] In this embodiment, the core analysis engine also includes a teaching effectiveness quantitative analysis submodule; it uses the analyze_teaching_effect method to compare pre-test and post-test data, calculate the average changes of various indicators, and automatically call LLM to generate a teaching effectiveness analysis report with graphics and text (such as the conclusion and attribution of "knowledge base improved but interest declined" as pointed out in the test instructions). The teaching resource management module includes a teaching resource database construction submodule and a prompt word project submodule; The teaching resource repository construction submodule is configured to receive teaching documents uploaded by users, parse the uploaded Excel and Word teaching resources using the `parse_excel_file` and `parse_word_file` methods, extract keywords with LLM to generate high-quality keywords for the content using the `extract_keywords_with_ai` method, and automatically classify the resources into six categories—"learning information repository, feedback repository, case repository, discussion repository, practice repository, and learning summary repository"—based on a preset keyword library using the `auto_classify_content` method. Finally, it generates a concise summary of the resource content using the `generate_summary` method, thereby constructing an intelligent teaching resource repository with semantic retrieval capabilities. It also receives teaching business questions input by users. The prompt word engineering submodule is configured to raise teaching business questions, retrieve teaching documents corresponding to the teaching business questions from the intelligent teaching resource library, and construct structured prompt words.

[0042] In one specific embodiment, it also includes a data unification manager and a session manager and a key scheduler; The data unification manager is responsible for the persistence of all data, maintaining user, chat history, lesson preparation notes and cases, teaching resource library and API key data table, ensuring data consistency and security; The aforementioned session management and key scheduler are responsible for the lifecycle management of user sessions and load balancing of multiple API keys, ensuring the stable operation of the system in a multi-user environment.

[0043] In one specific embodiment, the system adopts a Python-based B / S architecture, with the front end using the Grado framework to build an interactive web interface, the back end using MySQL for data persistence, and intelligent services provided through the integration of a large language model API.

[0044] Thus, this invention achieves a leap from "vague experience" to "precise data" in learning analysis: through quantitative assessment using a three-dimensional model and 3D visualization, teachers can quickly and intuitively grasp the overall learning situation of the class and individual differences, accurately identifying problem students (such as those with "high foundation but low interest"), providing unprecedented data support for precision teaching. (Actual test data proves that the system accurately diagnosed the key contradiction of "improved knowledge base but decreased learning interest.") This represents a significant leap in AI's role, transforming it from a "tool" to a "partner": Through a unique `query_with_libraries` method, LLM is deeply integrated with teaching resource repositories and lesson preparation notes, resulting in highly personalized, context-relevant, and actionable analyses and suggestions—truly becoming an intelligent teaching research partner for teachers. (This is evidenced by the AI-generated teaching strategy report, which includes specific cases and tiered tasks, as described in the test specifications.) It provides an unprecedented intuitive insight tool: the interactive 3D learning map is the landmark innovation of this invention. It transforms abstract data into a visualized spatial distribution, which greatly reduces the difficulty of data interpretation and improves teachers' decision-making efficiency and teaching insight.

[0045] A complete "diagnosis-intervention-assessment" teaching closed loop has been constructed: the system seamlessly connects pre-class diagnosis, in-class lesson preparation support, and post-class effectiveness assessment, and generates evidence-based assessment reports through AI to drive teaching reflection and continuous improvement, thus achieving a scientific and refined teaching process. (This is demonstrated by the application path described in the paper and the actual closed-loop operation in the tests.) It enhances the effectiveness of guiding teaching values: by accurately matching teaching strategies with students' interests and preferences (such as a preference for film and television cases), it systematically and indirectly enhances the affinity and pertinence of teaching, making value guidance "subtle and imperceptible," and effectively responding to the need for improvement in teaching "formula," "process," and "packaging" from a technical perspective.

[0046] Example 4 In this embodiment, the learning insight method and growth analysis method based on the large language model were used to conduct precise teaching and effect evaluation; The learning situation insight method described above is used: Data Collection: Before class, teachers distribute a pre-test questionnaire to "Class xx" using Wenjuanxing (a survey platform) to collect data on students' cognition, interests, and habits regarding specific content. Data Import and Insights: Teachers uploaded the exported pre-test CSV file (e.g., 335708416_by serial number_Third Topic Questionnaire (Pre-test)_72_72.csv) to the TeachVision system. The system generated a 3D learning map, and teachers found that the class's overall knowledge base was acceptable (mean 6.6), but the distribution of interests was uneven.

[0047] AI-assisted lesson preparation: Teachers raise teaching-related questions: "For this class, how can I design lessons to enhance students' interest in learning the 'xx' theory?" The system retrieved records related to "case teaching" from the teacher's lesson preparation notes using the query_with_libraries method, as well as film and television cases from the resource library. Combined with the high preference of students for "film and television documentaries" (91.7%) in the pre-test data, a detailed teaching strategy report was generated (as stated in the test instructions), which suggested "using a clip from 'xx' as an introduction, combining mind maps to organize core concepts, and designing tiered tasks."

[0048] The student growth analysis method described above is used: Post-test data collection: After the course ends, the teacher immediately distributes the post-test questionnaire and collects the data (e.g., 337413252_by serial number_third topic questionnaire (post-test)_68_68.csv).

[0049] Effectiveness Quantification Analysis: Teachers upload pre- and post-test data simultaneously. The system automatically generates a 3D comparison chart and calls the `analyze_teaching_effect` method to generate an analysis report.

[0050] The report reveals a deeper problem: it clearly points out that while the knowledge base score significantly improved from 6.6 to 9.4 (+2.8 points), the learning interest score dropped from 7.5 to 6.3 (-1.2 points). The report analyzes the reasons as follows: "The theoretical teaching was overemphasized, failing to fully meet students' preferences for case-based teaching and visualization tools."

[0051] Driving reflective teaching: Based on this data evidence, teachers reflected deeply and found a clear direction for the next round of teaching improvement (such as adding more science and technology and current events cases that students are interested in).

[0052] Example 5 In this embodiment, the learning insight system based on the large language model is used for the intelligent construction and application of the teaching resource database: The teacher uploaded a Word document example about "xx".

[0053] The system automatically calls TeachingLibraryManager to parse the data and extracts keywords using AI, then automatically categorizes them into "case libraries" according to rules.

[0054] The system generates a content summary and stores all information in the database. When a teacher asks again, "How to cultivate students' xx spirit," the system will automatically provide the newly added case as context to the LLM during the query_with_libraries process.

[0055] The responses generated by LLM will include: "Using the case of xx, explain how the research team achieved a technological breakthrough through xx...", thus making the suggestions more timely and targeted.

Claims

1. A learning insight method based on a large language model, characterized in that, The specific steps include the following: Obtain students' raw learning data and convert it into multidimensional standardized numerical data through a pre-set scoring model; It receives teaching documents uploaded by users, performs semantic analysis to extract keywords and generate summaries, and builds an intelligent teaching resource library with semantic retrieval capabilities. The system identifies teaching business issues, retrieves corresponding teaching documents from the intelligent teaching resource database, and constructs structured prompts. Call the large language model interface, input the structured prompt words into the large language model, and obtain an assessment report on learning insights and actionable teaching suggestions based on it.

2. The method according to claim 1, characterized in that, The structured prompts are composed of text instructions that include teaching business questions, summaries of teaching resources, standardized numerical data, and lesson preparation notes and case studies, which are logically concatenated.

3. The method according to claim 2, characterized in that, The scoring model is specifically a large language model limited by preset scoring rule prompts. The standardized numerical data includes three dimensions: learning interest, knowledge base, and learning habits. After being converted into multidimensional standardized numerical data, the standardized numerical data is also mapped into scatter points in multidimensional space to generate an interactive multidimensional learning distribution map, where each scatter point represents a student, allowing users to intuitively identify teaching business problems. The specific steps for generating an interactive multidimensional learning distribution map are as follows: constructing a three-dimensional coordinate system, where the X-axis, Y-axis, and Z-axis correspond to learning interests, knowledge base, and learning habits, respectively; the color of the scatter points corresponds to the numerical value of the learning interests through a color mapping algorithm; the interactive multidimensional learning distribution map is configured such that when the cursor hovers over a scatter point, it displays the student's identity information and specific dimension score corresponding to that scatter point.

4. The method according to claim 1, characterized in that, The specific method for constructing the teaching resource database is as follows: Receive teaching documents uploaded by users and extract the text content; The large language model is invoked to perform semantic analysis on the text content, generating a keyword set; Set up a classification rule base to automatically categorize teaching documents into at least one of the following categories: case library, discussion library, exercise library, feedback library, or gain library, based on the keyword set. Generate a content summary of the teaching documents; combine the content summary, keywords, and document classification to construct an intelligent teaching resource library with semantic retrieval capabilities.

5. A student growth analysis method based on the aforementioned learning situation insight method, characterized in that, The specific steps include the following: The learning insight method described above generates an assessment report on the learning insights of the target student group and actionable teaching suggestions based on it. Based on the aforementioned assessment report and its actionable teaching recommendations, teaching interventions were conducted for the target student group. An assessment report on the learning situation of the target student group after teaching intervention is generated using the aforementioned learning insight method. By comparing the data in the pre-test assessment reports of the target student group before the teaching intervention with the data in the post-test assessment reports after the teaching intervention, the large language model is used to generate a teaching effectiveness evaluation report including attribution analysis and further actionable teaching suggestions.

6. The method according to claim 5, characterized in that, By comparing the data from the pre-intervention assessment reports of the target student group before and after the intervention, a large language model is used to generate a teaching effectiveness evaluation report including attribution analysis and further actionable teaching suggestions. The specific steps are as follows: Calculate the difference between the mean values ​​of pre-test and post-test data on multidimensional standardized numerical data; Draw a multi-dimensional comparison chart showing the spatial changes of the data before and after the test; Input the average difference, multidimensional positional variation characteristics, and preset attribution analysis instructions into the large language model; Obtain the textual descriptions of the reasons for the rise and fall of indicators and improvement strategies from the output of the large language model.

7. A learning insight system based on a large language model, characterized in that, include: The core analysis engine includes a data loading and verification submodule and an AI-powered intelligent question answering submodule. The data loading and verification submodule is configured to acquire students' original learning data and convert it into multidimensional standardized numerical data through a preset scoring model. The AI ​​intelligent question-answering submodule is configured to call the large language model interface, input the structured prompt words into the large language model, and obtain an assessment report on learning insights and actionable teaching suggestions based on it; The teaching resource management module includes a teaching resource database construction submodule and a prompt word project submodule; The teaching resource library construction submodule is configured to receive teaching documents uploaded by users, perform semantic analysis to extract keywords and generate summaries, and build an intelligent teaching resource library with semantic retrieval capabilities; and receive teaching business questions input by users. The prompt word engineering submodule is configured to raise teaching business questions, retrieve teaching documents corresponding to the teaching business questions from the intelligent teaching resource library, and construct structured prompt words.

8. The system according to claim 7, characterized in that, The core analysis engine also includes a visualization generation submodule; the visualization generation submodule is used to map the standardized numerical data into scatter points in a multidimensional space to generate an interactive multidimensional student learning distribution map, where each scatter point represents a student, allowing users to intuitively identify teaching business problems.

9. The system according to claim 7, characterized in that, It also includes a unified data manager, session management, and key scheduler; The data unification manager is responsible for the persistence of all data, maintaining user, chat history, lesson preparation notes and cases, teaching resource library and API key data table, ensuring data consistency and security; The aforementioned session management and key scheduler are responsible for the lifecycle management of user sessions and load balancing of multiple API keys, ensuring the stable operation of the system in a multi-user environment.

10. The system according to claim 7, characterized in that, The system adopts a Python-based B / S architecture. The front end uses the Grado framework to build an interactive web interface, the back end uses MySQL for data persistence, and provides intelligent services by integrating a large language model API.

Images