Education system based on generative artificial intelligence and knowledge tracking

Abstract

The invention relates to an education system based on generative artificial intelligence and knowledge tracking, and belongs to the technical field of artificial intelligence, and the system comprises a generative AI engine module which is used for receiving a query request of a user and generating an initial generation content based on a preset large language model according to the query request; the knowledge tracking module is used for generating probability distribution of learning actions of the students according to the initial generation content based on a knowledge tracking technology; the retrieval enhancement generation module is used for matching and correcting the initially generated content and a target industry standard library to obtain corrected content, and the target industry standard library is captured through a network capturing tool; and the personalized recommendation engine module is used for generating feedback content based on the probability distribution of the learning action and the corrected content. According to the method, the dynamic content generation capability of GenAI, the personalized analysis capability of knowledge tracking and the compliance verification function of an external knowledge base are effectively integrated, so that the teaching efficiency and the teaching quality are improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an educational system based on generative artificial intelligence and knowledge tracing. Background Technology

[0002] In recent years, the application of artificial intelligence (AI) technology in the field of education has developed rapidly, especially in the areas of generative artificial intelligence (GenAI) and knowledge tracing (KT), where significant progress has been made.

[0003] Despite the potential of AI technology in education, the following core issues remain in practical applications: The disconnect between generated content and industry standards: General GenAI-generated teaching materials (such as security policies and code examples) lack dynamic adaptation to industry compliance standards (such as ISO 27001 and PEP8), requiring manual secondary verification, which is inefficient; Insufficient personalized adaptation: Traditional knowledge tracking models predict learning status based solely on historical answer data, failing to dynamically adjust recommendation strategies based on real-time generated content, resulting in rigid learning resource recommendations; Lack of multimodal interaction: Existing tools are mostly limited to text interaction, making it difficult to support the real-time feedback needs of complex teaching scenarios such as code debugging and strategy visualization; Ethical and compliance risks: AI-generated content may contain errors or non-compliant statements (such as privacy policy loopholes), requiring the construction of a reliable verification mechanism.

[0004] In summary, existing technologies have failed to effectively integrate the dynamic content generation capabilities of generative AI, the personalized analysis capabilities of knowledge tracking, and the compliance verification functions of external knowledge bases, resulting in low teaching efficiency and quality. Summary of the Invention

[0005] In view of this, it is necessary to provide an education system based on generative artificial intelligence and knowledge tracing to solve the problems of low teaching efficiency and quality in existing technologies.

[0006] To address the aforementioned problems, this invention provides an education system based on generative artificial intelligence and knowledge tracing, comprising: The generative AI engine module is used to receive and generate initial content based on the user's query request and a pre-set large language model. The knowledge tracking module is used to generate a probability distribution of students' learning actions based on the initial generated content using knowledge tracking technology. The retrieval enhancement generation module is used to match and correct the initially generated content with the target industry standard library to obtain the corrected content. The target industry standard library is obtained by crawling text from the target industry using web crawling tools. The personalized recommendation engine module is used to generate feedback content based on the probability distribution of learned actions and the corrected content.

[0007] In one possible implementation, the preset large language model includes one or more of GPT-4 and Gemini.

[0008] In one possible implementation, the knowledge tracking module is also used to update the probability distribution of the student's learning actions based on the feedback content.

[0009] In one possible implementation, the generative AI engine module is used for, Receive and generate content metadata and initial content based on the user's query request and a pre-defined large language model; Extract key parameters based on the query request; The key parameters are input into a preset large language model to obtain temporary content; The temporary content is labeled for compliance, resulting in the initial generated content.

[0010] In one possible implementation, the knowledge tracking module is used to generate a probability distribution of students' learning actions based on content metadata in the initially generated content, using knowledge tracking technology.

[0011] In one possible implementation, the content metadata includes: student ID, skill ID, question ID, and response code.

[0012] In one possible implementation, the knowledge tracking module is used to, The student's historical answer sequence is input into an attention-based time series model for calculation to obtain the skill mastery level; Based on a pre-defined clustering algorithm, the accuracy rate of students' answers within a pre-defined time is calculated to obtain an ability profile; The difficulty of the question is obtained by calculating the historical accuracy rate of students' answers to the target question based on the beta distribution; Implicit relationship features are constructed based on the skill mastery level, the ability profile, and the ability profile; The content metadata is acquired and converted into an embedding vector; The implicit relation features are concatenated with the embedding vector to obtain the concatenated vector; The concatenated vector is input into a preset neural network model to obtain the probability distribution of the student's learning actions.

[0013] In one possible implementation, the retrieval enhancement generation module is used to, The target course text is crawled from the webpage using a pre-set web crawler tool, a target industry standard library is constructed based on the text, and the target industry standard library is converted into a vector database. Extract keywords from the initially generated content and convert the keywords into query vectors; Calculate the cosine similarity between the query vector and all clause vectors in the vector database; The clauses with a cosine similarity greater than a preset threshold are used to correct the initially generated content, resulting in corrected content.

[0014] In one possible implementation, the retrieval enhancement generation module is also used to score the compliance of the initially generated content.

[0015] In one possible implementation, the preset web crawler tool includes: One or more of Octoparse, Web Scraper, ParseHub, and Scrapinghub.

[0016] The beneficial effects of this invention are as follows: This invention provides an education system based on generative artificial intelligence and knowledge tracing. The system includes a generative AI engine module for receiving and generating initial content based on a preset large language model, according to user queries; a knowledge tracing module for generating a probability distribution of student learning actions based on content metadata using a preset neural network model, dynamically interacting with the GenAI-generated content in real time; a retrieval-enhanced generation module for matching and correcting the initial content against a target industry standard library, obtaining corrected content, and automatically correcting the generated content by retrieving the industry standard library using retrieval enhancement technology, thereby improving review efficiency; and a personalized recommendation engine module for generating feedback content based on the probability distribution of learning actions and the corrected content, dynamically providing learning content feedback to students, thereby improving learning efficiency. This invention effectively integrates GenAI's dynamic content generation capabilities, knowledge tracing's personalized analysis capabilities, and the compliance verification functions of external knowledge bases, thereby improving teaching efficiency and quality. Attached Figure Description

[0017] Figure 1 This is a system architecture diagram of an embodiment of an education system based on generative artificial intelligence and knowledge tracing provided by the present invention; Figure 2 This is a system architecture diagram of an embodiment of an education system based on generative artificial intelligence and knowledge tracing provided by the present invention; Figure 3 This is a system architecture diagram of an embodiment of an education system based on generative artificial intelligence and knowledge tracing provided by the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0019] In the description of the embodiments of the present invention, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0020] The terms "first," "second," etc., used in the embodiments of this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a technical feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature.

[0021] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0022] Before demonstrating the embodiments, the following terms will be explained.

[0023] Generative Artificial Intelligence (GenAI) is an important branch of artificial intelligence, a technology that generates text, images, audio, video, code, and other content based on algorithms and models. Unlike the analytical functions of traditional AI, generative AI can learn and generate new content with logical structure. Unlike traditional AI, which only processes and analyzes input data, generative AI can learn and simulate the inherent patterns of things, generating new content with logic and coherence based on user input. The core of this technology relies on multimodal models, enabling generative output of heterogeneous data tailored to user needs.

[0024] Knowledge tracing (KT) is a technology used to track changes in students' knowledge status during the learning process and predict their performance in future exercises. KT simulates and predicts students' knowledge mastery by analyzing their interaction data within the learning system, such as answer records and exercise completion status.

[0025] Retrieval-Augmented Generation (RAG) is a model that combines retrieval and generation techniques. It generates answers or content by referencing information from external knowledge bases, offering strong interpretability and customizability. It is suitable for various natural language processing tasks, including question-answering systems, document generation, and intelligent assistants. The advantages of RAG models lie in their versatility, ability to achieve real-time knowledge updates, and the provision of more efficient and accurate information services through end-to-end evaluation methods.

[0026] Figure 1 A system framework diagram of an embodiment of the educational system based on generative artificial intelligence and knowledge tracing provided by the present invention is shown below. Figure 1 As shown, an education system 100 based on generative artificial intelligence and knowledge tracing includes: Generative AI engine module 101 is used to receive and generate initial content based on the user's query request and a preset large language model. The knowledge tracking module 102 is used to generate a probability distribution of students' learning actions based on the initial generated content using knowledge tracking technology. The retrieval enhancement generation module 103 is used to match and correct the initial generated content with the target industry standard library to obtain the corrected content. The target industry standard library is obtained by crawling text from the target industry using a web crawling tool. The personalized recommendation engine module 104 is used to generate feedback content based on the probability distribution of learned actions and the corrected content.

[0027] Compared with existing technologies, this embodiment provides an education system based on generative artificial intelligence and knowledge tracing. The system includes a generative AI engine module for receiving and generating initial content based on a preset large language model, according to user queries; a knowledge tracing module for generating a probability distribution of student learning actions based on content metadata using a preset neural network model, dynamically interacting with the GenAI-generated content in real time; a retrieval-enhanced generation module for matching and correcting the initial content against a target industry standard library, obtaining corrected content, and automatically correcting the generated content by retrieving the industry standard library using retrieval enhancement technology, thereby improving review efficiency; and a personalized recommendation engine module for generating feedback content based on the probability distribution of learning actions and the corrected content, dynamically providing learning content feedback to students, thereby improving learning efficiency. This invention effectively integrates GenAI's dynamic content generation capabilities, knowledge tracing's personalized analysis capabilities, and the compliance verification functions of external knowledge bases, thereby improving teaching efficiency and quality.

[0028] In specific embodiments of the present invention, such as Figure 2 The diagram illustrates the system architecture of an example of an education system based on generative artificial intelligence and knowledge tracing. The user interaction layer provides multimodal input / output interfaces (text, code, visualization strategies) and supports browser plugins, mobile devices, and API access. The generative AI engine (GenAI) generates initial teaching content (such as security policies and code examples) based on large language models (e.g., GPT-4, Gemini). The knowledge tracing module (KT) uses a multi-feature implicit relation model (MLFBK) to analyze students' skill mastery, ability profiles, and problem difficulty in real time. The retrieval-enhanced generation module (RAG) dynamically crawls industry standard libraries (such as the NIST framework and GDPR provisions) and course-related texts (video subtitles, academic papers) to ensure the compliance of generated content. The personalized recommendation engine combines KT analysis results with RAG retrieval content to generate dynamic learning suggestions (such as materials to strengthen weak knowledge points and advanced case studies).

[0029] like Figure 3 The diagram shows a flowchart of an embodiment of an educational system based on generative artificial intelligence and knowledge tracing. Input phase: Students submit requests via browser plugins (e.g., "Generate XYZ Bank's email security policy"). Generation phase: GenAI generates an initial policy draft (including basic measures such as encryption and password protection). RAG searches the NIST framework and GDPR provisions, revising the policy content (e.g., adding multi-factor authentication). Tracking and recommendation phase: The KT module analyzes students' historical data and marks skills not mastered (such as "not covered by the APRA guidelines").

[0030] The recommendation engine pushes targeted learning materials (such as APRA compliance cases). Feedback and optimization phase: Students submit revised strategies, and the system generates a compliance score (e.g., 95% alignment with NIST CSF).

[0031] The KT module updates student ability profiles and iteratively optimizes subsequent recommendation content.

[0032] In some embodiments of the present invention, the preset large language model includes one or more of GPT-4 and Gemini.

[0033] In some embodiments of the present invention, the knowledge tracking module 102 is further configured to update the probability distribution of the student's learning actions based on the feedback content.

[0034] In some embodiments of the present invention, the generative AI engine module 101 is used for, Receive and generate content metadata and initial content based on the user's query request and a pre-defined large language model; Extract key parameters based on the query request; The key parameters are input into a preset large language model to obtain temporary content; The temporary content is labeled for compliance, resulting in the initial generated content.

[0035] In a specific embodiment of the present invention, the generative AI engine module 101 specifically includes the following functions: Input preprocessing: Parse user requests and extract key parameters (such as industry type and regulatory requirements).

[0036] Content generation: Invoke large language models (such as GPT-4) to generate text, code, or strategy drafts.

[0037] Compliance marking: Initial compliance marking of generated content (e.g., marking NIST clauses that are not covered).

[0038] First, the system parses the content output by generative AI (such as security policies or code), extracts key measures (such as "SSL encryption" and "password complexity") using natural language processing technology, and performs static analysis on the code to identify security operations. Next, it performs semantic matching between the extracted measures and pre-stored industry standard libraries (such as NIST and GDPR), converts the text into vectors using a pre-trained model (such as Sentence-BERT), calculates the cosine similarity between the vector and the standard clause, and marks the clause as uncovered if the similarity is below a threshold (such as 0.7). Subsequently, the system automatically injects structured tags to distinguish between compliant items (marked with specific clauses, such as "NIST PR.DS-2") and uncovered items (marked with risk levels, such as high-risk "missing multi-factor authentication"), and supports dynamic updates to the standard library (synchronizing with the latest regulations every 24 hours).

[0039] In some embodiments of the present invention, the knowledge tracking module is used for, This is used to generate a probability distribution of student learning actions based on content metadata in the initially generated content, using knowledge tracing technology.

[0040] In some embodiments of the present invention, the content metadata includes: student ID, skill ID, question ID, and response code.

[0041] In some embodiments of the present invention, the knowledge tracking module 102 is used for, The student's historical answer sequence is input into an attention-based time series model for calculation to obtain the skill mastery level; Based on a pre-defined clustering algorithm, the accuracy rate of students' answers within a pre-defined time is calculated to obtain an ability profile; The difficulty of the question is obtained by calculating the historical accuracy rate of students' answers to the target question based on the beta distribution; Implicit relationship features are constructed based on the skill mastery level, the ability profile, and the ability profile; The content metadata is acquired and converted into an embedding vector; The implicit relation features are concatenated with the embedding vector to obtain the concatenated vector; The concatenated vector is input into a preset neural network model to obtain the probability distribution of the student's learning actions.

[0042] In a specific embodiment of the present invention, the preset neural network model includes: One or more of the following: feedforward neural network model, convolutional neural network model, recurrent neural network model, and generative adversarial neural network model.

[0043] In a specific embodiment of the present invention, the knowledge tracking module 102 predicts the learning status and generates a probability distribution of the student's learning actions based on the student's historical data (answer records, interaction behavior), specifically including the following steps: Multi-feature embedding: Encode student ID, skill ID, question ID, and response ID into feature vectors.

[0044] Implicit Relationship Modeling: Calculating Skill Mastery, Ability Profile, and Problem Difficulty using the MLFBK model.

[0045]

[0046] Predicted output: Generates the probability distribution of the student's next learning action (e.g., "Review the NIST Asset Management Framework").

[0047] Detailed steps: 1. Multi-feature embedding Input: Student ID, Skill ID, Question ID, Response ID (Correct / Incorrect).

[0048] operate: Embedding Layer: Maps discrete IDs to continuous vectors (dimension d=128) to capture latent semantic relationships.

[0049]

[0050] 2. Implicit Relationship Modeling (1) Calculation of Skill Mastery Input: Student's historical answer sequence (skill IDs and response results sorted by time).

[0051] Model: Time series models based on attention mechanisms (such as Transformer Encoder).

[0052] formula:

[0053] By weighting historical answer records using a self-attention mechanism, the current level of skill mastery is dynamically reflected. (2) Ability Profile Construction Input: The student's answer accuracy rate in different time windows (e.g., the most recent 30 days).

[0054] Clustering method: K-means clustering (k=5 classes, such as "beginner", "intermediate" and "expert").

[0055] formula:

[0056] Output cluster labels (e.g., "Advanced Learner") to reflect students' overall ability level. (3) Problem Difficulty Assessment Input: The historical accuracy rate of all students on this question.

[0057] Dynamic calculation: Difficulty coefficient (range 1-10) updated based on Beta distribution. =

[0058] The larger the value, the more difficult the problem. 3. Comprehensive Prediction Model Input integration: concatenating the embedded vectors with implicit relation features:

[0059] Prediction Networks: Fully connected layer (MLP): Maps the input to the action space (e.g., "Review the NIST framework" or "Try advanced questions").

[0060]

[0061]

[0062] Output: The probability distribution of each learned action, for example: P=[0.6,0.3,0.1] 4. Training and Optimization Loss function: Cross-Entropy Loss, which minimizes the difference between the predicted action and the actual action.

[0063]

[0064] Optimizer: Adam optimizer (learning rate lr=0.001), combined with gradient clipping to prevent overfitting.

[0065] In this embodiment, the knowledge tracking module tracks student ability profiles (such as skill mastery and problem difficulty) in real time using a multi-feature implicit relation model (MLFBK), and dynamically interacts with GenAI-generated content to form a closed loop of "generation → evaluation → optimization → regeneration". Experiments show that students' knowledge mastery improves by 15% (compared to the traditional KT model).

[0066] In some embodiments of the present invention, the retrieval enhancement generation module 103 is used for, The target course text is crawled from the webpage using a pre-set web crawler tool, a target industry standard library is constructed based on the text, and the target industry standard library is converted into a vector database. Extract keywords from the initially generated content and convert the keywords into query vectors; Calculate the cosine similarity between the query vector and all clause vectors in the vector database; Clauses with a cosine similarity greater than a preset threshold are used to correct the initially generated content, resulting in corrected content. In some embodiments of the present invention, the retrieval enhancement generation module 103 is also used to score the compliance of the initially generated content.

[0067] In some embodiments of the present invention, the preset web crawler tool includes: One or more of Octoparse, Web Scraper, ParseHub, and Scrapinghub.

[0068] In a specific embodiment of the present invention, the main function of the retrieval enhancement generation module 103 is to dynamically retrieve external knowledge bases and optimize GenAI-generated content, specifically including the following steps: Knowledge base construction: Use Scraper to crawl course webpage text (video subtitles, legal clauses) and build a vectorized knowledge base.

[0069] Search and Enhancement: Retrieve relevant clauses (such as Article 32 of the GDPR) based on keywords in GenAI-generated content (such as "email encryption").

[0070] 1. Location and Implementation of the Search The retrieval operation is performed in a vectorized knowledge base, which is constructed by the following steps: Knowledge sources: Course-related texts (such as video subtitles, legal clauses, and academic papers) are scraped using Scraper.

[0071] Vectorization: Use pre-trained models (such as Sentence-BERT or OpenAI Embeddings) to convert text into high-dimensional vectors.

[0072] Storage: The vectorized content is stored in a vector database (such as FAISS, Pinecone, or Milvus) to support efficient similarity retrieval.

[0073] 2. Search Process After GenAI generates the initial content, the system performs the following steps: Keyword extraction: Extract key entities (such as "email encryption" and "GDPR") from the generated content. Example: If the generated content is "It is recommended to use SSL to encrypt emails", extract the keywords ["SSL encryption", "email security"].

[0074] Vectorized query: Input the keywords into the same pre-trained model to generate query vectors.

[0075] formula:

[0076] Similarity search: Calculate the cosine similarity between the query vector and all clause vectors in the vector database:

[0077] Filter out terms with a similarity exceeding a threshold (e.g., >0.7).

[0078] Return result: Return to the matching terms (such as Article 32 of the GDPR, "Security requirements for data processing").

[0079] The search results are injected into the GenAI generation process to correct content biases.

[0080] Compliance verification: Calculate the alignment between the generated content and industry standards. Compliance score = (Number of matching clauses / Total number of relevant clauses) × 100% In a specific embodiment of the present invention, the personalized recommendation engine module dynamically recommends learning resources based on KT and RAG outputs. Specifically, it includes the following steps: Demand matching: Based on the weak knowledge points predicted by KT (such as "lack of multi-factor certification design"), match the strengthening materials retrieved by RAG (such as the ISO 27001 case).

[0081] Detailed steps: 1. Input: Weak knowledge points and learning resource database Weak knowledge points: Obtained from the Knowledge Trace (KT) module, for example: { "weak_skills": ["Multi-factor authentication design", "Data encryption algorithm"], "ability_profile": "Intermediate learner", "predicted_next_action": "Review the NIST framework" } Learning resource repository: Built by the RAG module, containing structured metadata (such as ISO 27001 case studies, NIST CSF guidelines, GDPR clause explanation videos, etc.), each resource is labeled: { "resource_id": "ISO27001-Case1", Title: Multi-Factor Certification Implementation Case "tags": ["MFA", "ISO 27001", "Access Control"], "difficulty": "intermediate", "related_standards": ["NIST PR.AC-1", "GDPR Article 32"]} } 2. Keyword Expansion and Semantic Mapping Step 1: Extract core keywords Extract core terms from weak knowledge points (e.g., “Multi-Factor Authentication Design” → ["MFA", "Multi-Factor Authentication", "Access Control"]).

[0082] Step 2: Synonym Expansion Expand synonyms using a predefined industry glossary (e.g., “MFA” → ["Two-Factor Authentication", "2FA"]).

[0083] Step 3: Associate Standard Clauses This is mapped to relevant industry standards (such as "multi-factor certification" being associated with NIST PR.AC-1 and ISO 27001A.9.4.2).

[0084] 3. Multi-dimensional search and candidate resource filtering Search dimensions: 1) Tag matching: Filter resources whose tags contain extended keywords (such as MFA).

[0085] 2) Standard association: Match the standard terms for resource association (such as NIST PR.AC-1).

[0086] 3) Difficulty matching: Filter resources that match the student's ability profile (such as "intermediate" difficulty).

[0087] Example query (pseudocode): candidate_resources = Resource.objects.filter( tags__overlap=["MFA", "Access Control"], related_standards__contains="NIST PR.AC-1", difficulty="Intermediate" ) 4. Semantic similarity ranking Step 1: Vectorization Use a pre-trained model (such as Sentence-BERT) to convert candidate resource titles and descriptions into vectors.

[0088]

[0089] Step 2: Calculate similarity The descriptions of weak knowledge points (such as "lack of multi-factor authentication design") are converted into vectors, and the cosine similarity with candidate resources is calculated:

[0090] Step 3: Sorting and Threshold Filtering Resources with a similarity score > 0.7 are retained and sorted in descending order of score.

[0091] 5. Context Enhancement and Priority Adjustment Step 1: Combine with learning scenarios Increase the weight of relevant resources based on students' recent behavior (such as frequent access to "NIST Framework" materials).

[0092] Step 2: Dynamic weight allocation Urgency weight: If the exam is approaching, the weight of high-urgency resources will increase (e.g., formula weight = basic similarity × (1 + urgency factor)).

[0093] Difficulty gradient: It is divided into "easy → medium → difficult" levels, and resources that match the current ability level are recommended first.

[0094] Example of sorting after adjustment: 1. [High Urgency] ISO 27001 Multifactor Certification Case Study (Similarity 0.85) 2. [Difficulty Appropriate] NIST PR.AC-1 Implementation Guidelines (Similarity 0.82) 3. [Basic Consolidation] Introductory Tutorial on Two-Factor Authentication (Similarity 0.78) 6. Output: Structured recommendation list The final recommendation results are generated with priority and metadata: [ { "resource_id": "ISO27001-Case1", Title: Multi-Factor Certification Implementation Case "type": "Text / Case Study", "Matching criteria": "Label: MFA, Standard: NIST PR.AC-1", Priority: 0.92, "Reason for recommendation": "You need to strengthen your knowledge of access control recently. This case is at an intermediate level and is related to the key points of the exam." }, { "resource_id": "NIST-Guide-AC1", "title": "NIST PR.AC-1 Implementation Guide", "type": "video / tutorial", Matching criteria: Standard: NIST PR.AC-1, Difficulty: Intermediate Priority: 0.88 Recommendation Reason: Based on your skill profile, this step-by-step guide is recommended for mastering multi-factor certification design. } ] Prioritization: Recommended content is sorted by urgency (e.g., exam is approaching) and difficulty level (from easy to difficult).

[0095] Multimodal output: Provides text explanations, code examples, and strategy visualization tools (such as compliance comparison charts).

[0096] The technical effects of the present invention are illustrated below through examples. Table 1 shows the parameter configurations and compliance standards for each scenario.

[0097] Table 1: Parameter Configurations and Compliance Standards for Each Scenario

[0098] Performance verification: Through test cases, 50 cybersecurity students used this system to generate security policies. When the RAG retrieval frequency was updated once every 24 hours, the clause coverage was the best (F1=0.91). The prediction accuracy was the highest under the KT time window and the 30-day historical data window (AUC=0.88).

[0099] The test results showed improved efficiency, with the strategy optimization time reduced from an average of 4.2 hours to 2.5 hours (p<0.01), improved compliance, and the final strategy's alignment with NIST CSF increased from 72% to 93%.

[0100] This embodiment dynamically retrieves industry standard libraries (such as the NIST framework and GDPR provisions) using Retrieval Enhanced Generation (RAG) technology, automatically corrects the generated content, and achieves a compliance alignment rate of over 95% (experimental data). Student policy optimization time is reduced by 30%, and manual review costs are reduced by 50%.

[0101] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0102] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. An education system based on generative artificial intelligence and knowledge tracing, characterized in that, include: The generative AI engine module is used to receive and generate initial content based on the user's query request and a pre-set large language model. The knowledge tracking module is used to generate a probability distribution of students' learning actions based on the initial generated content using knowledge tracking technology. The retrieval enhancement generation module is used to match and correct the initially generated content with the target industry standard library to obtain the corrected content. The target industry standard library is obtained by crawling text from the target industry using web crawling tools. The personalized recommendation engine module is used to generate feedback content based on the probability distribution of learned actions and the corrected content.

2. The education system based on generative artificial intelligence and knowledge tracing according to claim 1, characterized in that, The preset large language model includes one or more of GPT-4 and Gemini.

3. The education system based on generative artificial intelligence and knowledge tracing according to claim 1, characterized in that, The knowledge tracking module is also used to update the probability distribution of student learning actions based on the feedback content.

4. The education system based on generative artificial intelligence and knowledge tracing according to claim 1, characterized in that, The generative AI engine module is used for, Receive and generate content metadata and initial content based on the user's query request and a pre-defined large language model; Extract key parameters based on the query request; The key parameters are input into a preset large language model to obtain temporary content; The temporary content is labeled for compliance, resulting in the initial generated content.

5. The education system based on generative artificial intelligence and knowledge tracing according to claim 1, characterized in that, The knowledge tracking module is used for, This is used to generate a probability distribution of student learning actions based on content metadata in the initially generated content, using knowledge tracing technology.

6. The education system based on generative artificial intelligence and knowledge tracing according to claim 5, characterized in that, The content metadata includes: student ID, skill ID, question ID, and response code.

7. The education system based on generative artificial intelligence and knowledge tracing according to claim 6, characterized in that, The knowledge tracking module is used for, The student's historical answer sequence is input into an attention-based time series model for calculation to obtain the skill mastery level; Based on a pre-defined clustering algorithm, the accuracy rate of students' answers within a pre-defined time is calculated to obtain an ability profile; The difficulty of the question is obtained by calculating the historical accuracy rate of students' answers to the target question based on the beta distribution; Implicit relationship features are constructed based on the skill mastery level, the ability profile, and the ability profile; The content metadata is acquired and converted into an embedding vector; The implicit relation features are concatenated with the embedding vector to obtain the concatenated vector; The concatenated vector is input into a preset neural network model to obtain the probability distribution of the student's learning actions.

8. The education system based on generative artificial intelligence and knowledge tracing according to claim 1, characterized in that, The search enhancement generation module is used for, The target course text is crawled from the webpage using a pre-set web crawler tool, a target industry standard library is constructed based on the text, and the target industry standard library is converted into a vector database. Extract keywords from the initially generated content and convert the keywords into query vectors; Calculate the cosine similarity between the query vector and all clause vectors in the vector database; The clauses with a cosine similarity greater than a preset threshold are used to correct the initially generated content, resulting in corrected content.

9. The education system based on generative artificial intelligence and knowledge tracing according to claim 1, characterized in that, The search enhancement generation module is also used to score the compliance of the initially generated content.

10. The education system based on generative artificial intelligence and knowledge tracing according to claim 8, characterized in that, The preset web crawler tools include: One or more of Octoparse, Web Scraper, ParseHub, and Scrapinghub.

Images