Personalized scientific education system based on AI agent driving

Abstract

The invention relates to a personalized scientific education system based on AI agent driving, and belongs to the technical field of informatics. The education system runs a main agent, a user twinborn agent, a learning planning platform and a graph structure database. The main agent is used for collecting knowledge states and learning behaviors of users, constructing a user model, and training one or more user twin agents based on the user model. And the user twinborn agent is used for simulating learning behaviors and results of the user under different learning paths. And the learning planning platform combines the user model and the knowledge graph stored in the graph structure database to generate a plurality of candidate learning paths for the twin intelligent agent of the user to perform analogue simulation. And the main agent selects an optimal path according to a simulation result and implements the optimal path to the personalized teaching process of the user. According to the technical scheme, high-adaptability learning path recommendation and dynamic adjustment are realized, and the intelligent level of individualized teaching is improved.

Description

Technical Field

[0001] The present invention belongs to the field of informatics technology, and specifically relates to a personalized science education system driven by an AI agent. Background Art

[0002] With the rapid development of artificial intelligence technology, personalized education has gradually become an important research direction in intelligent education systems. Traditional teaching systems often rely on unified course schedules and fixed learning paths, making it difficult to dynamically adjust teaching strategies based on individual differences, resulting in limited learning efficiency and experience. In recent years, intelligent education platforms based on big data analysis, knowledge graphs, and natural language processing have initially achieved user behavior analysis and content recommendation, but they still rely primarily on static rules or simple user profiles and lack dynamic simulation and adaptive optimization capabilities. At the same time, existing systems lack predictable modeling of the "user learning process" itself, making it impossible to effectively evaluate the adaptability and learning effect of a learning path for a specific user before implementing teaching. Therefore, an education system that can simulate user learning behavior, dynamically optimize learning paths, and possess high-level intelligent decision-making capabilities is needed to achieve truly personalized teaching.

[0003] After consulting relevant public technologies, the technical solution with announcement number TWM431397U proposes a teaching platform system, which reduces the learning pressure in the learning process and improves the teaching effect by setting teaching animation content that is dynamically generated according to the students' immediate status feedback; the technical solution with publication number US20140295398A1 proposes an interactive teaching system, which enables teachers to understand the students' learning status in real time during remote online teaching and adjust the teaching methods by setting multiple interactive components in the system; the technical solution with announcement number CN117830044B proposes a cloud computing-based interactive teaching data management system and method, which can analyze the students' mastery of calligraphy skills through a series of students' performances in the calligraphy teaching process.

[0004] The above technical solutions all propose various types of digital technologies to assist teaching planning or teaching processes; however, there are currently few references to relevant teaching technology solutions that currently apply artificial intelligence technology.

[0005] The foregoing discussion of the background art is intended only to facilitate an understanding of the present invention. This discussion does not acknowledge or admit that any of the material referred to is part of the common general knowledge. Summary of the Invention

[0006] The purpose of the present invention is to provide a personalized science education system driven by an AI agent; the education system runs a main agent, a user twin agent, a learning planning platform and a graph structure database. The main agent is used to collect the user's knowledge status and learning behavior, build a user model, and train one or more user twin agents based on the user model. The user twin agent is used to simulate the user's learning behavior and results under different learning paths. The learning planning platform combines the user model with the knowledge graph stored in the graph structure database to generate multiple candidate learning paths for the user twin agent to simulate. The main agent selects the optimal path based on the simulation results and implements it in the user's personalized teaching process. This technical solution realizes the recommendation and dynamic adjustment of learning paths with high adaptability, and improves the intelligence level of individualized teaching.

[0007] The present invention adopts the following technical solution: a personalized science education system driven by an AI agent, the education system comprising: The main agent is configured to interact with the user, collect the user's knowledge status, behavioral characteristics and learning preferences, and build a user model; One or more user twin agents, generated by training the master agent based on the user model, to simulate the user's learning behavior and learning outcomes under a specified learning path; A learning planning platform, configured to receive input from the master agent and / or the user twin agent and output a multi-stage personalized learning path and learning resources; Among them, the learning planning platform includes a database running a graph structure; the database provides content data support for the learning path; the main intelligent agent obtains the content data support for the learning path from the learning planning platform based on the user model, formulates multiple learning paths, and drives one or more user twin intelligent agents to perform simulation learning, and based on the results of the simulation learning, selects one or more learning paths to implement personalized teaching for the user.

[0008] Preferably, the main agent includes the following functional modules: The large language model module is configured to perform natural language understanding and generation to enable language interaction and content expression with users; The context modeling and memory module is configured to record the user's historical interaction information, intent context, and behavior preferences, and maintain the user state; A graph interface module is configured to connect to the database and perform graph data retrieval, semantic query and learning path related information extraction; The reasoning and evaluation module is configured to analyze, judge, and optimize the current learning state based on the user model, database information, and task objectives; The interaction management module is used to control the dialogue process, information confirmation, feedback triggering and prompt delivery between the main intelligent agent and the user.

[0009] Preferably, the user model is constructed by the master agent, and the construction steps include: Collecting user behavior data during the learning process, including the user's learning time, answer accuracy, reaction time, error type, and operation path; Based on the behavioral data, analyze the user's mastery level and cognitive preferences on different knowledge points; Combining the user's historical learning records with the current learning task, a multi-dimensional vector representation is constructed to characterize the user's knowledge status, behavioral characteristics, and learning tendencies. The multi-dimensional vectorized representation is stored as a user model to drive the initialization of the user twin agent and the planning of personalized learning paths.

[0010] Preferably, the master agent determines the number of user twin agents to be generated based on multiple considerations, and the steps in this process include: S100: Analyze the behavioral characteristics of the user model to identify the user's fluctuations in various learning behaviors and learning results; S200: Calculate the uncertainty score based on the volatility of the performance, which indicates the consistency of the user's learning behavior; S300: Determine the number of generated user twin agents based on the uncertainty score; By taking the above considerations into account, the generated user twin intelligent agent can cover the different possible behavior patterns of the user.

[0011] Preferably, the graph-based database includes one or more knowledge graphs; the knowledge graphs are used to represent structured knowledge relationships of learning content, and the composition of the knowledge graphs includes: Multiple nodes, each node is used to represent an entity, including knowledge points, learning resources, learning goals, user status or skill modules; a plurality of edges, each edge being used to represent a semantic association relationship between two nodes, wherein the association relationship includes one of a prerequisite relationship, a dependency relationship, a similarity relationship, a feedback relationship, or a teaching sequence relationship; Each node and edge is attached with attribute information, including labels, contextual metadata, importance weights or vector embeddings, to support semantic retrieval and path planning.

[0012] Preferably, the master agent is further configured to formulate a plurality of candidate learning paths based on the user model and the database, which includes the following steps: Identifying in the database the node corresponding to the user's current knowledge state and the node corresponding to the target knowledge point; Based on the prerequisite relationship, semantic relevance and teaching sequence between each knowledge point in the database, multiple feasible paths from the current node to the target node are constructed.

[0013] The beneficial effects achieved by the present invention are: 1. This technical solution constructs a user model through a master agent and derives multiple user twin agents. It can conduct multi-path simulation tests before teaching is implemented, select the optimal path and apply it to the user, significantly improving the match between the teaching plan and the user's learning characteristics, and realizing refined and differentiated teaching strategies.

[0014] 2. The main intelligent agent of this technical solution continuously collects user behavior data during the user's actual learning process and dynamically adjusts the user model, so that the subsequent twin intelligent agents can more accurately simulate the user status. The system can optimize the teaching content and rhythm in real time according to the changes in the user's learning status, enhancing the system's responsiveness and adaptability.

[0015] 3. The graph-structured database used in this technical solution has the ability to express knowledge point prerequisite relationships, semantic labels, and attribute weights. Combined with the graph interface module and reasoning module in the main intelligent agent, it can achieve semantic understanding and rapid construction of complex knowledge paths, improving the intelligent level of content scheduling and the logical integrity of path construction.

[0016] 4. The software and hardware components of the educational system of this technical solution adopt a modular design. The various working modules and components of the hardware part, as well as the instructions, parameters, and algorithms of the software part, can be easily replaced and / or upgraded later, thereby reducing the construction and maintenance costs of this system. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The present invention can be further understood from the following description in conjunction with the accompanying drawings. The components in the figures are not necessarily drawn to scale, but rather the emphasis is placed on illustrating the principles of the embodiments. In different views, the same reference numerals designate corresponding parts.

[0018] Explanation of the accompanying figures: 10-user; 100-main agent; 110-large language model module; 120-context modeling and memory module; 130-interaction management module; 140-graph interface module; 150-reasoning and evaluation module; 165-learning planning platform; 170-user model; 200-user twin agent; 230-database; Figure 1 This is a schematic diagram of the framework of the education system of the present invention; Figure 2 Schematic diagram of a database with a graph structure according to an embodiment of the present invention; Figure 3 Schematic diagram of the architecture of the master agent in an embodiment of the present invention; Figure 4 Schematic diagram of the structure of the user twin agent described in an embodiment of the present invention; Figure 5 Schematic diagram of the framework of the computer system used in the education system described in the embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention is further described in detail below in conjunction with its embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. For those skilled in the art, other systems, methods and / or features of the present embodiment will become apparent after reviewing the following detailed description. It is intended that all such additional systems, methods, features and advantages are included in this specification. Included within the scope of the present invention and protected by the appended claims. Additional features of the disclosed embodiments are described in the following detailed description, and these features will be apparent from the following detailed description.

[0020] The same or similar reference numerals in the drawings of the embodiments of the present invention correspond to the same or similar components. In the description of the present invention, it should be understood that if there are terms such as "upper", "lower", "left", "right", etc. indicating an orientation or positional relationship, they are based on the orientation or positional relationship shown in the drawings. This is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or component referred to must have a specific orientation. The terms used in the drawings to describe the positional relationship are only for illustrative purposes and cannot be understood as limiting this patent. For those skilled in the art, the specific meanings of the above terms can be understood according to the specific circumstances.

[0021] Example 1: Exemplarily, a personalized science education system driven by an AI agent, the education system includes: The main agent is configured to interact with the user, collect the user's knowledge status, behavioral characteristics and learning preferences, and build a user model; One or more user twin agents, generated by training the master agent based on the user model, to simulate the user's learning behavior and learning outcomes under a specified learning path; A learning planning platform, configured to receive input from the master agent and / or the user twin agent and output a multi-stage personalized learning path and learning resources; Among them, the learning planning platform includes a database running a graph structure; the database provides content data support for the learning path; the main intelligent agent obtains the content data support for the learning path from the learning planning platform based on the user model, formulates multiple learning paths, and drives one or more user twin intelligent agents to perform simulation learning, and based on the results of the simulation learning, selects one or more learning paths to implement personalized teaching for the user.

[0022] Preferably, the main agent includes the following functional modules: The large language model module is configured to perform natural language understanding and generation to enable language interaction and content expression with users; The context modeling and memory module is configured to record the user's historical interaction information, intent context, and behavior preferences, and maintain the user state; A graph interface module is configured to connect to the database and perform graph data retrieval, semantic query and learning path related information extraction; The reasoning and evaluation module is configured to analyze, judge, and optimize the current learning state based on the user model, database information, and task objectives; The interaction management module is used to control the dialogue process, information confirmation, feedback triggering and prompt delivery between the main intelligent agent and the user.

[0023] Preferably, the user model is constructed by the master agent, and the construction steps include: Collecting user behavior data during the learning process, including the user's learning time, answer accuracy, reaction time, error type, and operation path; Based on the behavioral data, analyze the user's mastery level and cognitive preferences on different knowledge points; Combining the user's historical learning records with the current learning task, a multi-dimensional vector representation is constructed to characterize the user's knowledge status, behavioral characteristics, and learning tendencies. The multi-dimensional vectorized representation is stored as a user model to drive the initialization of the user twin agent and the planning of personalized learning paths.

[0024] Preferably, the master agent determines the number of user twin agents to be generated based on multiple considerations, and the steps in this process include: S100: Analyze the behavioral characteristics of the user model to identify the user's fluctuations in various learning behaviors and learning results; S200: Calculate the uncertainty score based on the volatility of the performance, which indicates the consistency of the user's learning behavior; S300: Determine the number of generated user twin agents based on the uncertainty score; By taking the above considerations into account, the generated user twin intelligent agent can cover the different possible behavior patterns of the user.

[0025] Preferably, the graph-based database includes one or more knowledge graphs; the knowledge graphs are used to represent structured knowledge relationships of learning content, and the composition of the knowledge graphs includes: Multiple nodes, each node is used to represent an entity, including knowledge points, learning resources, learning goals, user status or skill modules; a plurality of edges, each edge being used to represent a semantic association relationship between two nodes, wherein the association relationship includes one of a prerequisite relationship, a dependency relationship, a similarity relationship, a feedback relationship, or a teaching sequence relationship; Each node and edge is attached with attribute information, including labels, contextual metadata, importance weights or vector embeddings, to support semantic retrieval and path planning.

[0026] Preferably, the master agent is further configured to formulate a plurality of candidate learning paths based on the user model and the database, which includes the following steps: Identifying in the database the node corresponding to the user's current knowledge state and the node corresponding to the target knowledge point; Based on the prerequisite relationship, semantic relevance and teaching sequence between each knowledge point in the database, multiple feasible paths from the current node to the target node are constructed.

[0027] As attached Figure 1 FIG2 shows an architectural block diagram of an exemplary embodiment of the educational system. The educational system operates in a computer system. The computer system may include one or more processors and memory, and may run an operating system and one or more application programs. The various components of the computer system may be physically, communicatively, and / or operationally interconnected to enable communication between the components.

[0028] Preferably, the operating system and the application are used to perform the various functions of this implementation scheme, including running one or more trained artificial intelligence-based main intelligent agents, and also including training other intelligent agents. Exemplarily, the computer system also includes running a learning planning platform 165. The learning planning platform 165 can receive input from one or more main intelligent agents, and generate corresponding learning plans and output corresponding learning content based on the learning plans in response to the inputs of these main intelligent agents. The output of the learning planning platform 165 can be fed back to the user through the output device of the computer system, or the output of the learning planning platform 165 can also be passed back to the intelligent agent as a new input, realizing a dynamic closed loop of information flow.

[0029] The computer system may receive user input through an input device and provide the user input to the learning planning platform 165. The output of the computer system may be output to a client device used by the user through a user interface.

[0030] The main agent, i.e., the artificial intelligence-based agent, in this technical solution specifically refers to the artificial intelligence software program used to drive the personalized science teaching interaction process. The agent can be trained to have the following functional characteristics: Environmental perception capabilities, which can collect and identify users' learning status, including behavioral data (clicks, stays, browsing, etc.), voice intonation, facial expressions, interaction preferences, etc. Knowledge modeling capabilities: Based on the user's existing knowledge graph and learning process data, a user-based user profile is constructed to generate a personalized cognitive model of the user, and ultimately a user twin intelligent agent that can simulate the user's learning, cognitive, and feedback capabilities; Task planning capability, i.e., dynamically generating recommended learning paths and learning tasks based on the teaching objectives or the guided learning plan of the learning planning platform, combined with the learning status and interests of the user and / or the user's household twin; Natural language interaction capabilities, that is, conducting question-and-answer, explanation, and guided dialogue with users through natural language to provide interactivity with humans; Multimodal fusion capability, that is, supporting content generation and output in multiple forms such as images, text, animation, and voice.

[0031] For example, in order to achieve deeper semantic understanding and efficient learning path recommendation, the education system further constructs a graph-based database 230 for the learning planning platform 165, which is used to store multi-dimensional information such as user characteristics, knowledge content, learning behavior, and ability maps, and form the underlying knowledge map of the education system.

[0032] Specifically, database 230 uses a massively scalable graph model structure to organize various types of educational data in the form of nodes and edges. Nodes represent entities, such as knowledge points, learning resources, and user status, while edges represent relationships between entities, such as prerequisite relationships, feedback relationships, and similarity relationships. Attribute information, such as contextual metadata, semantic tags, or vector embeddings, can be added to nodes and edges. Database 230 is a NoSQL database type that stores data in the form of nodes, edges, and properties, representing entities, relationships between entities, and attribute information, respectively. Therefore, database 230 is capable of efficiently storing, retrieving, and analyzing complex and highly interconnected data structures.

[0033] At the same time, the exemplary Figure 2 As shown, user 10 can also be represented as a node within database 230. For ease of presentation, this view only shows an area within database 230. By panning or zooming the view, other users can be included in this view. In this schematic view, user 10 includes an association with other nodes, such as a registered software developer node 20.

[0034] Illustratively, the logic for establishing and maintaining database 230 itself includes one or more software developer nodes 20. Illustratively, the connection between developer node 20 and user 10 can be expressed as "Founded by." Furthermore, software developer node 20 and multiple skills or expertise areas form "sub-field" connection paths, such as cloud computing node 21, full-stack JavaScript (JS) development node 22, mobile development node 23, back-end development node 24, database development node 25, and front-end development node 26.

[0035] Furthermore, database 230 also exemplarily establishes connection paths related to scientific learning, such as a biology node 30 and an astronomy node 40. For biology node 3, its subfields include an astrophysics node 31, an astronomical observation node 32, and a planetary geography node 33.

[0036] Furthermore, the technical personnel of the educational system of the present invention can create interactive and information-rich visual data representations using data visualization tools. These tools generally support multiple chart types and provide rich customization options to enhance the clarity and aesthetics of the graphical representations.

[0037] Furthermore, the graph-based database 230 can flexibly describe multi-level, multi-relational data between logical elements such as "user → age group → learning goal" or "concept → knowledge point → learning path." This facilitates semantic retrieval and path recommendations, for example, recommending closely related yet uncovered sub-fields based on the user's current learning content. When the graph-based database 230 is combined with a large language model, the agent can understand and apply the context and structure of the graph to generate content or learning path recommendations that better meet the learner's needs.

[0038] In an exemplary embodiment, the main agent 100's behavior is based on a large language model (LLM) and a generative AI architecture. The LLM provides powerful natural language understanding and generation capabilities, enabling it to learn linguistic knowledge from massive corpora and perform complex language reasoning. Furthermore, the main agent, combined with graph-based natural language processing (NLP) and knowledge retrieval capabilities from graph-based databases, can efficiently acquire, understand, and reason about specialized educational knowledge prepared by designers or acquired through its own searches.

[0039] For example, when a user raises and discusses a scientific question, the main intelligent agent 100 can extract the key concepts in the question through NLP methods, locate the corresponding nodes in the knowledge graph and obtain relevant facts or premise knowledge, and then use LLM to generate fluent and easy-to-understand explanations from these structured knowledge. This "large model + knowledge graph" technical route fully utilizes the advantages of both: large language models are good at language understanding and generation, and knowledge graphs are good at accurately storing and reasoning domain knowledge. The fusion of the two can build a more comprehensive and sophisticated educational knowledge system. Using this combination, the customizability of scientific education content and the accuracy of planning are improved. It can not only provide personalized learning paths and resource recommendations, but also reduce the risk of factual deviations or "hallucinations" in the answers of large language models under low constraints.

[0040] For example, as shown in the attached Figure 3 As shown, the main intelligent agent 100 constructs and implements the following functional modules through software architecture or modular system design; each functional module works together to support the operation of the main intelligent agent 100. Preferably, the multiple functional modules include: The large language model module 110, as the core language model, is the foundation for the main agent 100 to conduct conversations and generate content. The large language model module 110 has been pre-trained and specially tuned to undertake the main natural language processing tasks. The large language model module is configured to convert the user's natural language input into a structured internal representation, enabling the main agent to understand the user's intentions and the meaning of the expressed content, and generate corresponding natural language responses based on internal decisions. The large language model module 110 is the foundation for the main agent to conduct conversations and generate content. Its powerful language understanding and generation capabilities ensure smooth and contextually coherent interactions with users.

[0041] The context modeling and memory module 120 is responsible for modeling and long-term memory of conversation status and user learning capabilities. As the main agent interacts with the user in real time, the context modeling and memory module 120 continuously maintains the user model 170, including data such as the user's knowledge mastery, learning preferences, historical questions, and system-provided learning content. The context modeling and memory module 120 can continuously update the user profile based on conversation history and progress information obtained from the learning planning platform 165. For example, the context modeling and memory module 120 records the user's weaknesses or proficiency in knowledge points, aspects, and areas, and stores previous conversation content for subsequent reference. Based on the generation and reflection of contextual memory, the main agent remains sensitive to the user's state during long conversations, thereby generating personalized interactions for different audiences. Furthermore, the context modeling and memory module 120 utilizes the database 230 for memory storage and, through contextual retrieval technology, injects relevant historical information into the running context window of the large language model module 110, thereby achieving consistency and personalization in long-term conversations.

[0042] The interaction management module 130 is responsible for controlling the flow and strategy of communication between the main agent and the user. Based on its recognition of user intent and the current state of the conversation, the interaction management module 130 determines the main agent's interactive behavior. Exemplarily, this includes determining whether the current behavior is to answer user questions, provide prompts, raise new questions, or initiate an explanation session. The interaction management module incorporates strategic logic for educational conversations. Exemplarily, it organizes conversation strategies through Socratic questioning, error correction, and incentive feedback mechanisms to ensure effective interaction and adherence to pedagogical principles. Exemplarily, when a user asks a question, the interaction management module invokes the large language model module 110 and context modeling and memory module 120 to generate an answer. When a user solves a problem or exercises, the interaction management module invokes the reasoning and evaluation module 150 to determine correctness and provide feedback. If the user is confused or straying from the topic, the interaction management module determines whether to insert prompts, adjust the topic, or request re-planning from the learning planning platform. The interaction management module ensures the continuity and pedagogical purpose of the conversation, avoiding blind responses in a single round of conversation while gradually guiding the user toward their learning goals through multiple rounds of conversation.

[0043] The graph interface module 140 is configured as an interface for the main agent to interact with the database 230. The graph interface module converts natural language queries or internal decision-making requirements into retrieval operations on the graph database, and parses and filters the retrieval results for use by other modules. For example, when the large language model module 110 understands that the user question involves the concept of "photosynthesis", the graph interface module constructs a corresponding graph database query, such as a SPARQL query or API request, to retrieve the definition, attributes or relationship of the "photosynthesis" concept node with other concepts. The exemplary execution code is as follows: PREFIX edu:<http: / / example.org / edu#> SELECT ?subprocess ?label WHERE { edu:Photosynthesis edu:hasSubProcess ?subprocess . ?subprocess rdfs:label ?label . } In the above code snippet, edu:Photosynthesis refers to the URI of the concept “photosynthesis” in the graph-based database 230; edu:hasSubProcess refers to the custom "contains subprocess" predicate; After the code is executed, the query returns all "sub-process" nodes connected to the "photosynthesis" node and their labels (Chinese names or terms). For example, the following list can be returned, where the corresponding node path is returned as an HTTP address: subprocess label http: / / example.org / edu#LightAbsorption Absorb light energy http: / / example.org / edu#ElectronTransfer Electronic transmission http: / / example.org / edu#CarbonFixation Carbon fixation Combined with the structured results, the large language model module 110 generates a natural language answer, for example: "Photosynthesis mainly includes the following basic processes: first, the absorption of light energy, followed by the electron transfer process, and finally, carbon fixation. These steps together complete the process of converting light energy into chemical energy." For example, when the interaction management module 130 needs to arrange the next learning topic for user 10, it can request the graph interface module 140 to obtain a list of unmastered nodes adjacent to the current knowledge point. The graph interface module 130 is also responsible for invoking graph-structured NLP algorithms to add important information from the conversation to the user's knowledge graph. For example, it can mark knowledge points that the user has mastered or misunderstood in the graph. Through this interface, the main intelligent agent 100 achieves efficient retrieval and updating of domain knowledge, providing a basis for its responses and decisions.

[0044] The reasoning and evaluation module 150 is configured to perform: (1) complex reasoning calculations; and (2) user learning effect evaluation. In some embodiments, when multi-step deductive reasoning or mathematical calculations are required, the reasoning and evaluation module 150 assists the large language model module 110 in completing chained reasoning, such as decomposing problems and solving them step by step. The reasoning and evaluation module 150 can be considered a tool of the large language model or a chain-of-thought mechanism. In other embodiments, for content that requires logical verification, the reasoning and evaluation module 150 can use a rule engine or external computing library for verification to enhance the accuracy of the agent's answers.

[0045] In other embodiments, the main intelligent agent 100 performs quantitative analysis and judgment on the user's learning performance through the reasoning and evaluation module 150. For example, for the user's exercise answers, the reasoning and evaluation module 150 can compare with the standard answers or judge the correctness based on the relationship between the knowledge points in the graph; for the user's learning process data, the reasoning and evaluation module 150 performs statistical completion, accuracy, reaction time and other indicators. Based on these evaluation results, the reasoning and evaluation module 150 generates annotations and / or information for updating the user model. For example, such as marking "mastered" or "to be consolidated" on the knowledge point nodes, and providing decision-making basis for the interaction management module 130 and the learning planning platform 165. The reasoning and evaluation module 150 can also be configured to generate a brief learning report or confidence score for reference by users and teachers, so as to achieve measurability and explainability of the teaching process.

[0046] In summary, the above functional modules work closely together in the system architecture, with the large language model module 110 and the context modeling and memory module 120 ensuring dialogue understanding and personalization, the interaction management module 130 responsible for dialogue strategies and the overall teaching process, the graph interface module 140 providing authoritative knowledge support, and the reasoning and evaluation module 150 ensuring logical correctness and effect tracking.

[0047] In an exemplary embodiment, the master agent 100 needs to establish a user model 170 for each user. Exemplarily, the user model is a structured representation of information such as the user's learning ability, knowledge status, behavioral characteristics, and preference parameters. Preferably, this user data is represented in a relational data structure or graph format. However, the user model does not have an execution function, but rather serves as a basic data set for the master agent or other modules to use for judgment, decision-making, and matching.

[0048] Preferably, the content described in the user model includes one or more of the following: (1) Current knowledge mastery level. For example, the mastery level of each knowledge point can be scored to quantify the mastery level; (2) Learning ability indicators, such as memory retention rate, reaction time, and problem-solving accuracy; (3) Behavioral patterns, such as study time periods, content preferences, and exercise completion strategies; (4) Learning objectives and constraints, such as exam time, target level, etc.

[0049] The above are just some examples of indicators related to user learning areas, and are not intended to be limiting descriptions of the present technical solution. All indicators that can be quantified as parameters can be used as parameter items of the user model 100.

[0050] Preferably, the user model is expressed and stored in the form of an attribute table, a graph database node attribute set, a JSON document structure, etc.

[0051] In an exemplary embodiment, the user twin 200 is an intelligent, simulated learner built based on each user's user profile, representing the user's digital counterpart within a computer system. User twin 200 is capable of executing and "participating" in the learning content output by the learning planning platform, simulating learning task execution. It also predicts the user's learning behavior based on the results of the execution and reflects changes in the user's state. This means it has behavioral simulation, path testing, and feedback capabilities.

[0052] Furthermore, since the user twin agent 200 itself is an AI-based agent, it possesses the capabilities of perception, reasoning, and task execution, enabling it to perform "simulated learning" as a user mapping in a computer system, simulating the user's implementation of various learning paths and tasks, and estimating the user's mastery curve. At the same time, the learning path of the real user can be reversely optimized based on the behavioral results of the user twin agent 200. Furthermore, during the interaction between the main agent 100 and the user, the main agent 100 updates the user model in real time, ensuring that the capabilities and status of the user twin agent 200 reflect the user's actual situation as closely as possible, thereby realizing a digital twin.

[0053] The main working mechanism of the master agent 100 is described in detail below.

[0054] Exemplarily, the main intelligent agent 100 operates based on a closed-loop mechanism of "perception-decision-execution", dynamically updates the user model 170 and adjusts the teaching strategy during continuous interaction with the user.

[0055] The user model 170 is a structured digital representation of the user's learning ability, knowledge status, behavioral characteristics, and preference parameters, etc., as described in detail later.

[0056] The closed-loop mechanism may include the following execution modes: Preferably, during the interaction between the main agent 100 and the user 10, multiple trigger conditions can be used to trigger the main agent 100 to update the user model 170. For example, when a user answers a question, the reasoning and evaluation module 150 will determine whether it is correct or not and estimate the user's mastery of the knowledge points involved. For example, when the user 10 answers the question correctly and quickly, it is considered that the user 10 has a high degree of mastery of the relevant knowledge points, and errors are recorded as weaknesses. The context modeling and memory module 120 is responsible for writing the evaluation results into the user model and updating the mastery status of the knowledge point. Based on the ability to "remember" in which areas the user has difficulties, the main agent 100 will consciously strengthen explanations or exercises for these weaknesses in subsequent conversations. This operating mechanism ensures that the user model gradually approaches reality: as the interaction increases, the user model 170 can more and more accurately reflect the user's knowledge structure and learning behavior.

[0057] Preferably, the main agent 100 includes a function to identify user intent and determine if there is a discrepancy between the user's actual behavior and that expressed by the user model 170. The interaction management module 130 leverages the natural language understanding capabilities of the large language model to analyze the intent of each user input in real time. For example, this function assesses whether new questions raised by users express confusion or deviate from the current learning topic. In conjunction with the user model 170, the main agent 100 can also detect discrepancies between the user model 170's predicted state and actual behavior. For example, if the system predicts that a user has mastered a concept but repeatedly answers related questions incorrectly, this indicates a discrepancy between the user model 170 and reality. Alternatively, if a user asks questions that fall outside the scope of the current learning plan, indicating a shift in their interests or needs. By comparing the user model 170's predicted and actual interaction performance with the actual interaction performance, the main agent determines whether cognitive bias or interest shifts have occurred and adjusts its strategy accordingly. This process is similar to how a teacher adjusts teaching hypotheses based on classroom feedback to ensure that the teaching plan aligns with the user's actual situation.

[0058] Preferably, the main agent 100 includes a decision-making process for when to restart the conversation. Exemplarily, when the main agent 100 detects that the user has not responded for a long time, is obviously confused, or the topic is interrupted, it will proactively restart the conversation. For example, if the user has not responded to the previous content for a long time, the main agent 100 may choose to send a prompt or explain again in a different way to restart the communication. If the user continues to show misunderstandings, the main agent 100 may suspend the current explanation and ask questions to find out the cause of the misunderstanding, forming a new conversation round to clarify the concept. This conversation restart mechanism ensures the continuity and effectiveness of the interaction and prevents users from stopping learning in silence or misunderstanding.

[0059] Preferably, the master agent 100 decides whether to adjust the current learning task or the overall learning path based on the learning effect data provided by the reasoning and evaluation module 150 and the status of the user model 170. Preferably, the master agent 100 triggers a learning strategy adjustment in the following circumstances. For example, if the difficulty of the learning path deviates from the user's actual ability, as reflected by the user's performance exceeding expectations in the current task, such as completing exercises quickly and correctly multiple times, then the current learning task difficulty is low. The master agent 100 can skip redundant exercises or introduce more advanced content in advance. Conversely, if the user repeatedly fails, the difficulty is too high. The master agent 100 should return to supplement prerequisite knowledge or reduce the difficulty of the task. In some embodiments, when the learning plan output by the learning planning platform 165 does not match the user's actual ability assumptions, such as an unsuitable learning method or inconsistent interests, the master agent 100 can interact with the learning planning platform 165 to request a replanning of the learning path to better meet the user's needs. In some embodiments, if a user expresses a strong interest in other topics or proposes a new learning goal while implementing a learning plan, the master agent 100 will first identify the true intention behind this deviation. After confirming the new learning goal, it can interrupt the current task and collaborate with the learning planning platform 165 to restructure the plan. Through these mechanisms, the master agent 100 can adaptively adjust the teaching process, neither prematurely abandoning difficult points nor excessively repeating easy points. It can also dynamically optimize the next step based on user feedback, thereby ensuring that personalized learning remains effective and targeted.

[0060] Throughout the entire process, the master agent 100 operates within a closed-loop monitoring and feedback loop, i.e., a cycle of "perceiving user behavior → updating the user model → adjusting decision strategies → executing interactive actions → re-perceiving feedback." This cycle allows the agent to continuously self-correct and optimize its behavior, ensuring that the teaching process always focuses on the user's real-time needs and long-term goals.

[0061] Furthermore, in an exemplary embodiment, the master agent 100 and the learning planning platform 165 collaborate through a clear interface protocol to dynamically develop and update personalized learning paths. Preferably, the interaction mechanism between the master agent 100 and the learning planning platform 165 can be divided into the following stages: Initial learning path acquisition: When a new learning session begins or the user establishes a learning goal, the main intelligent body will send the user's key information to the learning planning platform 165, requesting the generation of an initial learning plan. Exemplarily, the information sent includes: the knowledge level assessment, learning preferences, target subjects and scope, and other limiting requirements already in the user model 170, such as exam dates. Based on this, the learning planning platform 165 calls its internal course planning algorithm and rule library, combines the knowledge graph of the learning field, analyzes and mines the user's learning needs, matches appropriate knowledge points and resources, and develops a customized learning path plan. Exemplarily, the learning planning platform 165 may identify several core concepts and skills that the user needs to master, and generate a sequence of learning units in order of prerequisite relationships, with each unit accompanied by recommended learning materials and practice questions. This initial plan, along with metadata such as priority and estimated time, will be returned to the main intelligent body.

[0062] For example, during the actual teaching interaction, the main agent 100 determines whether the original plan needs to be adjusted based on the user's performance. If the reasoning and evaluation module 150 detects that the user has mastered certain knowledge points in advance or has repeatedly encountered difficulties in a certain link, the main agent will send a mid-course adjustment request to the learning planning platform. This request includes the updated user model 170 to add mastery / weakness information in real time and update the current teaching progress. After receiving this data, the learning planning platform 165 recalculates the optimized path through an algorithm, including skipping mastered content, adding supplementary content before insertion, or changing the number of exercises. The modified path increment is then sent to the main agent. The main agent 100 uses this information to connect learning and switch teaching strategies. For example, if the platform returns the indication "The user has not mastered the prerequisite Y of X, and it is recommended to supplement Y," the main agent will pause the current topic in the conversation, introduce the teaching unit of Y instead, and then return to the original topic at the appropriate time.

[0063] Preferably, the interaction between the main agent 100 and the learning planning platform 165 is a continuous closed loop rather than a one-time one. After each major learning unit is completed, the main agent 100 summarizes the user 10's test scores, error types and other learning feedback in the unit and sends it to the learning planning platform 165. The learning planning platform 165 adjusts the difficulty and rhythm of the subsequent path based on the latest data. For example, if the user performs well in a certain unit, the platform may reduce the number of subsequent related exercises; otherwise, it will increase the consolidation link. In this way, the learning path is calibrated and optimized every time a stage is completed. The main agent 100 periodically synchronizes with the learning planning platform 165 to ensure that subsequent teaching decisions are always based on the latest and most accurate user models and data analysis.

[0064] Preferably, in order to achieve the above-mentioned collaboration, efficient communication interfaces and protocols are necessary. The main agent and the learning planning platform 165 usually communicate through an internal API or message queue, and use a unified data format to describe the user model and learning path. Exemplarily, a JSON format file can be used, which internally describes the user ID, knowledge point ID, status, recommended operations, etc. of the object. When processing a request, the learning planning platform 165 queries the database 230 to understand the knowledge point relationship or find available resources, and then makes a comprehensive decision. The above-mentioned communication processing process can be transparent to the user, and the final result is presented to the user 10 by the main agent 100. The main agent 100 ensures that the adjustments recommended by the learning planning platform 165 are naturally integrated into the human-computer dialogue, for example, guiding the user into a newly recommended learning activity through dialogue.

[0065] Through the above-mentioned interactive mechanism, the learning planning platform 165, as the core processing department of the overall learning planning relative to the main intelligent entity 100, is responsible for in-depth learning path calculation and macro planning; while the main intelligent entity 100, as the front-end execution department, implements these plans in specific teaching dialogues, and the two work closely together.

[0066] Furthermore, in an exemplary embodiment, the master agent 100 makes full use of the domain knowledge graph stored in the database 230 and realizes accurate acquisition of teaching content and instant answering of questions through a collaborative retrieval mechanism. Preferably, the master agent 100 and the database 230 collaborate in the following ways: Database 230 provides knowledge query and answer support for the master agent 100. When a user raises a question related to scientific knowledge or requires detailed information about a concept during a conversation, the master agent 100 calls the graph interface module 140 and generates a corresponding search request to the database 230. For example, when a user is learning the principles of photosynthesis, the master agent 100 extracts the keyword "photosynthesis" from the learning topic, locates the concept node in the knowledge graph, and then extracts its definition, related reaction formulas, and upstream and downstream knowledge. The structured results returned by database 230 are summarized and organized by the master agent 100, and a coherent explanation and answer is generated for the user through the large language model module. This process ensures the accuracy and reliability of the answer content because the information comes directly from a well-organized and verified knowledge base, rather than relying entirely on model memory. In addition, the master agent 100 can also expand relevant knowledge points for the user based on the associations provided by database 230, such as using an analogy with respiration related to photosynthesis to enrich the coherence and systematic nature of the teaching content.

[0067] Furthermore, the graph structure of database 230 supports multi-hop knowledge reasoning and associative retrieval by the main agent 100. For example, when a user poses a complex question that requires the integration of multiple concepts, such as an interdisciplinary question or a topic requiring deduction, the main agent 100 can perform a multi-step query in database 230. For example, the main agent 100 first searches for the various concept nodes involved in the question, and then traces the relationship paths between them layer by layer. For example, if the current learning topic is "Battery performance changes with seasons," the main agent 100 can search for the "battery performance" node, then find the relationship related to temperature, and infer the chain of events in which low temperatures affect the chemical reaction rate, leading to a decrease in battery power. In each step of the search, the agent uses graph-based NLP technology to convert intermediate results into new query constraints, continuously expanding a subgraph on the knowledge graph of database 230, and ultimately finding the complete relationship chain required to answer the question. Through this collaborative search, the main agent 100 can disassemble and analyze a deep learning topic that requires spanning multiple knowledge points and implicitly integrate the reasoning process into the answer.

[0068] Furthermore, the master agent 100 can retrieve prior knowledge from the database 230 to generate a personalized learning path for the user. When planning personalized learning content, the master agent 100 includes accessing the user's node graph chain in the database 230 to obtain the knowledge set relevant to the user. For example, for a weak knowledge point A marked in a user model 170, the agent can query the graph to find A's prerequisite knowledge B and C, as well as advanced application D, to determine the teaching sequence, for example, completing B and C first, then continuing with A, and finally extending to D. For example, when selecting exercises for the user, the agent can search the knowledge graph for questions or case resources marked with the same knowledge point based on the current learning node, and then select items of appropriate difficulty from these resources and push them to the user. Database 230 not only supports the provision of knowledge for the problem to be solved, but also serves as an index for teaching resources. The master agent 100 can rely on the database 230 to quickly locate lecture notes, video teaching resources, examples, etc. related to the current learning content, avoiding blind searches and greatly improving the accuracy and efficiency of personalized recommendations.

[0069] In some embodiments, the main agent 100 includes necessary maintenance of the database 230. In addition to acquiring knowledge from the database 230, the main agent 100 will also update the knowledge graph information of the database 230 under specified circumstances to form a collaborative evolution mechanism. Preferably, after the user completes a knowledge assessment, the main agent 100 can write the results back to the user's user model 170 to mark the degree of mastery. Preferably, if the system introduces new teaching content or finds that the knowledge graph needs to be corrected and supplemented, the main agent 100 or the main agent 100 cooperates with other applications to update the database 230 to ensure that the graph knowledge base 230 is always authoritative and up-to-date. These updates ensure the quality of subsequent retrieval and a complete description of the user portrait.

[0070] Furthermore, in an exemplary embodiment, the construction of the master agent 100 relies on the training paradigm of a large language model, and is customized for specific educational scenarios through pre-training, fine-tuning, and continuous optimization. Preferably, the overall generation and training process of the master agent 100 can be divided into the following stages: Initial pre-training phase: During this phase, the system's core large-scale language model undergoes unsupervised pre-training using a vast corpus of general-purpose data. For example, by training the GPT series of models on encyclopedia articles, textbooks, popular science materials, and internet text, they learn the grammatical and semantic aspects of language, as well as a broad range of common sense and scientific knowledge. This process typically produces a pre-trained model with hundreds of millions or even hundreds of billions of parameters, demonstrating strong natural language understanding and generation capabilities but not yet optimized for specific educational tasks. The pre-training phase endows the model with general intelligence and knowledge storage, giving it a good starting point for downstream tasks.

[0071] Specific Fine-Tuning Phase: After obtaining the pre-trained model, the system conducts supervised fine-tuning to adapt it to the specific role of the "educational agent" and form an agent with certain practical capabilities. Preferably, this phase further trains the model using a large amount of domain data and interaction data, including: textbooks and question banks in science education to strengthen the model's professional knowledge depth; historical teaching dialogue materials to simulate teacher-student dialogues, allowing the model to learn the strategies and tone of teaching conversations; and data on converting knowledge graph triples into text to enable the agent to transform structured knowledge into explanatory language, etc.

[0072] Preferably, during the fine-tuning process, the pre-trained model is guided to learn teaching strategies such as how to provide layered explanations based on user questions, how to ask questions to guide student thinking, and when to provide prompts or clues. This ensures that the model's output more closely aligns with the intelligent tutor's desired behavior patterns. Furthermore, a method called Reinforcement Learning with Human Feedback (RLHF) may be employed: educational experts provide scoring feedback on the quality of the model's responses, further fine-tuning model parameters to ensure that responses are both accurate and compliant with the curriculum and ethical requirements. After fine-tuning, the master agent 100 is obtained, and its large language model module 110 is optimized for knowledge accuracy, educational tone, and interaction strategies, making it more capable of handling conversational scenarios involving personalized teaching.

[0073] Preferably, after the main agent 100 is actually deployed and put into operation, a continuous learning and optimization mechanism is set up to adapt to the ever-changing user data and needs. On the one hand, the response is personalized through small-scale learnable parameters or context adaptation. Exemplarily, the education system can maintain a unique preference vector or prompt template for each user and incorporate this part of the content into the user model 170. When generating answers, the user preferences are embedded in the context, so that the main agent 100 understands the style and known areas of a specific user. In this way, the main model parameters of the main agent 100 can achieve customized output for a single user without changing the global parameters.

[0074] On the other hand, some implementations include continuous optimization of the global model of the master agent 100. For example, the system collects a large amount of real user interaction logs for regular retraining or fine-tuning of the model. This includes analyzing the logs to find that the model repeatedly gives incorrect or invalid responses to certain types of questions. The development team will then add corresponding training data (or modify the knowledge graph) to fine-tune the model again to correct the deviation. Alternatively, when new knowledge is updated in the scientific field, the system will incorporate this new knowledge into the training data to ensure that the model keeps pace with the times. Continuous optimization also includes performance tuning (such as distilling small models to speed up responses) and security tuning (such as strengthening filtering responses to bad questions and answers). In addition, the master agent may cooperate with an adaptive learning algorithm during operation to adjust dialogue policy parameters based on real-time user feedback (this is more like policy-level training, such as adjusting the algorithmic weight of question difficulty, rather than directly changing LLM parameters).

[0075] In general, the continuous optimization mechanism ensures that the main intelligent body becomes smarter as it is used: it can not only accumulate global experience to improve overall teaching capabilities, but also provide targeted adaptation to individuals, accompanying users in their long-term growth.

[0076] Embodiment 2: This embodiment should be understood to include at least all the features of any one of the aforementioned embodiments and to be further improved thereon.

[0077] For example, as shown in the attached Figure 4 As shown, the composition principle of the user twin intelligent agent 200 is schematically illustrated. The generation of the user twin intelligent agent 200 is based on the existing user model 170. The user model 170 includes a large amount of multi-faceted data description of the user 10, including: Knowledge state refers to the user's mastery level in various knowledge points or skill areas, which can be expressed as a proficiency vector or probability distribution of knowledge concepts.

[0078] Behavioral characteristics refer to the typical behavioral patterns of users during the learning process, such as problem-solving speed, habitual error types, and distribution of learning time.

[0079] Learning preferences refer to users' preferences for learning content and methods, such as preference for visual explanations or text descriptions, preference for independent exploration or guided teaching, and preference for the difficulty of exercises.

[0080] Furthermore, in some embodiments, the master agent 100 is used to train one or more user twin agents 200. The goal of the training is to make the twin's response in a dynamic learning environment consistent with that of a real user, that is, to fit the learning performance of a real user. Preferably, the training process includes the following stages: First, the initial parameters of the user twin agent 200 are set. Exemplarily, based on the information of the user model 170, the initial parameters of the user twin agent 200 are extracted from the user model 170 by parameter copying or mapping. In other words, when creating the user twin agent 200, the user 10's knowledge mastery is filled into the corresponding internal representation of the user twin agent 200, such as the initial weight value of the neural network or the knowledge graph node weight, so as to convert the user 10 behavioral characteristics into the behavioral strategy parameters of the user twin agent 200, such as decision thresholds, reaction delay distribution, etc., and convert the learning preferences into the twin agent's tendency configuration for selecting learning resources or paths in simulation. After this initialization, the twin should maintain a high degree of consistency with the user in known characteristics in the initial state, laying the foundation for subsequent simulation training.

[0081] The master agent then drives the user twin 200 to perform a simulated learning task, which may include a series of learning tasks or exercises, preferably tasks or exercises previously performed by a real user. Based on its current level of training, the user twin 200 responds truthfully to each task and generates simulation results, including numerical records such as the time taken to answer the questions and whether the answers were correct or incorrect.

[0082] Furthermore, the fitting parameters of the user twin agent 200 are updated. The main agent 100 compares the simulation results of the user twin agent 200 with the expected user performance, calculates the error signal, and adjusts the internal parameters of the user twin agent 200 accordingly, so that the output of the twin agent is closer to the real behavior of the mapped user. In some embodiments, such adjustments can be made using supervised learning or reinforcement learning methods. Preferably, for continuous numerical differences (such as completion time errors), gradient descent can be used to minimize the mean square error; for matching discrete decisions such as multiple-choice questions, a reward function can be designed to encourage the twin agent's decision to be consistent with the user's choice. The main agent 100 continuously iterates the "simulation-evaluation-update" cycle to gradually reduce the deviation between the twin agent's prediction and the user's actual performance.

[0083] In an exemplary embodiment, when the user twin agent 200 completes training, it is likely to be able to simulate the feedback of the user after executing a learning plan at the digital level. Therefore, the main agent 100 uses the trained user twin agent 200 to conduct trial simulations on multiple candidate learning paths to evaluate the possible evolution of the user's knowledge state under different teaching strategies. The educational system can use the main agent 100 in a virtual environment to instruct one or more user twin agents 200 of the user to execute different learning content sequences or paths in sequence or synchronously, and record the changes in their knowledge mastery and forgetting.

[0084] For each candidate learning path, to ensure that simulations of different learning paths begin with the same initial state, the master agent 100 takes a snapshot of the state of the user twin 200 and derives multiple copies. The master agent 100 then drives the user twin 200 to complete each learning unit or exercise in the order defined by the learning path. After each learning unit is completed, the master agent 100 records the changes in the user twin's 200's mastery of the relevant knowledge points.

[0085] For example, in a simulation implementation, a forgetting curve model can be introduced during the path learning process to simulate the decay of knowledge over time. In some implementations, a decay function can be applied to the knowledge parameters of the user twin agent 200 according to the Ebbinghaus forgetting curve theory. For example, let Δt be the time interval between two learnings, and introduce the factor To reduce the proficiency of the corresponding knowledge and simulate the forgetting effect. Among them, Δt represents the time interval between the last learning of the relevant knowledge point and the current time; λ represents the forgetting rate constant. The larger the value, the faster the forgetting. It can be specifically set in the user model 170 according to the actual situation of the user; finally, f(Δt) represents the proficiency ratio of the knowledge point retained when the time interval is Δt. In other words, in the simulation, every time the user twin intelligent agent 200 fails to review a certain knowledge point for a period of time, its corresponding mastery will decrease according to the forgetting function, which approximately simulates the phenomenon of human memory decline over time.

[0086] After completing the entire path, the user twin agent 200's knowledge mastery over time (or learning steps) is recorded, along with the final knowledge level and skill mastery. The master agent 100 can compare the results of different paths, analyze the knowledge consolidation and forgetting processes of each learning path, and prioritize learning paths with higher memory efficiency.

[0087] Similarly, in some embodiments, the user twin intelligent agent 200 can be used to simulate the efficiency of knowledge acquisition, for example, to compare the time or computing power cost of using different learning paths to master the same program for each knowledge point; shorter time or lower computing power cost indicates that the load generated by the implementation of the learning path is lower and the pressure is relatively light.

[0088] Preferably, the knowledge change and forgetting process is modeled as an explicit mathematical function or algorithmic process. For example, a data structure is used to record the user's twin intelligent agent 200's proficiency in each knowledge point, which decreases over time according to the forgetting curve formula. When a learning event occurs, the corresponding proficiency is improved according to the learning gain model. This multi-path simulation provides a basis for optimizing the teaching path and also provides rich data for evaluating the consistency of the twin model's dynamic behavior with the user's actual situation.

[0089] Furthermore, in an exemplary embodiment, the degree of fidelity with which the user twin 200 simulates the mapped user needs to be verified. Therefore, the education system includes a matching evaluation mechanism. Periodic learning evaluations can be conducted with the user to obtain the user's actual learning outcomes. Under the same learning conditions, the predicted output of the user twin 200 can be compared with the actual performance of the real user, and the difference between the two can be quantified.

[0090] Exemplary, key comparison indicators include: Task completion time difference: Compare the difference or ratio between the time it takes the user twin 200 to complete a learning task and the actual time it takes the user 10. The smaller the time difference, the more accurately the twin simulates the user's rhythm.

[0091] Correctness Deviation: Compare the difference between the simulated accuracy of user twin 200 on a set of practice questions and the actual accuracy of user 10. This can be broken down into the accuracy deviation for each knowledge point, used to check whether user twin 200 has adequately mapped user 10's strengths and weaknesses.

[0092] Learning Curve Similarity: Compare the curve of user twin 200's simulated learning effect over the amount of practice with the actual learning curve of user 10, specifically the trajectory of knowledge and ability improvement with practice. For example, methods such as Dynamic Time Warping (DTW) distance and Pearson correlation coefficient can be used to quantify the similarity between the two curves to assess the degree of fit between the twins in terms of the learning progress rhythm.

[0093] Furthermore, for the above indicators, a comprehensive fit score or loss function L is defined to quantify the overall fit. For example, the loss function L is designed to be in the following form: ; In the above formula, the task time difference Δt and the accuracy difference Δacc are weighted and summed. The smaller the value, the closer the twin and the user's performance is. The loss can then be converted into a matching score. An example of a possible matching score S is: S=1 / (1+L), The matching score S is normalized to a value between 0 and 1, with 1 indicating a perfect match. During the training and evaluation phases, the master agent 100 monitors the fidelity of the user twin agent 200 by calculating the matching score S. When the user twin agent 200 is sufficiently accurate, all deviation indicators are small and the matching score is close to 1, indicating that the twin agent's predictions are almost identical to the user's real data.

[0094] Furthermore, in an exemplary embodiment, when the matching score S of the user twin agent 200 is found to be lower than a preset threshold during the training process or normal operation, the main agent 100 will initiate a corresponding retraining or parameter correction mechanism to improve the mapping accuracy of the user twin agent 200.

[0095] Preferably, the master agent 100 retrains the user twin agent 200 through an error feedback drive. Specifically, the master agent can collect the difference data between the simulation results of the current user twin agent 200 and the user's actual performance, use it as a new training sample, and perform targeted additional training on the user twin agent 200. For example, if it is found that the user twin agent 200 systematically exceeds the user's learning rate and assessment accuracy rate in a certain learning content, these cases can be used to adjust the model so that its simulation of such questions is corrected to its actual performance.

[0096] Furthermore, in a preferred embodiment, the model parameters of the user twin 200 can be adjusted based on the details of the deviation. For example, if analysis reveals that the knowledge decay is too slow or too fast due to an improper setting of the forgetting rate of the user twin 200, the parameter λ of the forgetting curve can be modified; or if the user twin 200's response sensitivity to a certain behavioral characteristic of the user is too low, the feature weight of that characteristic in the user twin 200 can be increased. If necessary, new feature dimensions can even be introduced to enrich the twin's input, making the model more comprehensive in fitting the user.

[0097] Furthermore, in an exemplary embodiment, the master agent 100 includes multiple user twin agents 200 representing the same user. Simulating multiple candidate learning paths, the system predicts the user's mastery efficiency, forgetting rate, and learning cost under different paths, ultimately selecting the optimal learning plan for the real user. To ensure a balance between simulation representativeness and computational efficiency, the system appropriately sets the number Nt of user twin agents 200 and the number Np of candidate learning paths, and statistically evaluates and selects the path results.

[0098] Exemplarily, the following steps are included: Step 1: Set the number of user twins, Nt. These twins simulate the behavioral and cognitive characteristics of user 10 during the learning process. Considering differences in attention, stability, and consistency of knowledge acquisition, the system introduces a user uncertainty score, U∈[0,1], to assess the degree of behavioral fluctuation. In some implementations, the uncertainty score, U, is calculated based on a combination of characteristics such as the variance of a user's historical answers, changes in learning time, and fluctuations in preferences.

[0099] Based on the score, the master agent 100 dynamically determines the number of derived user twin agents 200, calculated as follows: ; In the above formula, Nt base is the baseline number of user twin agents, such as 10, 15 or 20; γ is the uncertainty magnification factor, which can be preferably set to 10 or other suitable values. After rounding up the product of γ and U, the final Nt value is calculated.

[0100] This computational strategy ensures that fewer user twin agents are generated for users with stable behavior to save computing resources, while for users with large behavioral fluctuations, the number of user twin agents is increased to cover their potential behavior distribution, thereby improving the reliability of simulation predictions.

[0101] Step 2: Set the number of candidate learning paths Np. The number of candidate learning paths Np is mainly determined based on the prior path structure between the user's current knowledge state and the target knowledge point on the knowledge graph, as well as the computational budget allowed by the system.

[0102] In some embodiments, the master agent 100 first forms a path candidate set based on the learning planning platform and all possible pre-trained paths between the user's current knowledge node and the target node that can be enumerated in the database 230; then, to control the complexity, the system applies restriction rules to the path candidate set, such as limiting the length of the learning path steps, the length of the learning time, the learning difficulty, and filtering repeated nodes or irrelevant branches. Then, combined with the simulation budget B max , calculate the number of acceptable paths: ; Among them, B max Refers to the total computing power consumed by the system in completing a learning path selection process, such as B max The total number of simulation implementations can be set to 100 times, and the number of user twin agents Nt written is set to 5, then the number of learning paths allocated to each user twin agent 200 is 20.

[0103] In other embodiments, B max It can refer to total power or total operation time, etc.

[0104] Step 3: Perform performance analysis on the simulation results of all learning paths. For example, the master agent 100 drives all user twin agents to fully simulate each candidate learning path and record the key indicators during the path execution. Preferably, these key indicators may include: M ij , refers to the final knowledge mastery index of the j-th user twin agent after going through the i-th learning path; F ij , refers to the knowledge forgetting index simulated by the j-th user twin agent after experiencing the i-th learning path; T ij , refers to the time taken by the j-th user twin agent to complete the entire learning path after going through the i-th learning path.

[0105] The above three key indicators are dimensionless values ​​after normalization of the original values.

[0106] Furthermore, the average performance score is calculated for each learning path: ; In the above formula, η and ε are the forgetting penalty coefficient and the time penalty coefficient, respectively, to balance the weights of different evaluation dimensions. The specific values ​​of the two can be set by relevant technical personnel.

[0107] Preferably, the stability of the learning path results is evaluated, the fluctuation degree of each path under each user's twin agent is statistically analyzed, and the performance score of the i-th learning path is calculated. i Variance Var i .

[0108] Finally, the master agent 100 can calculate the performance score of all candidate learning paths. i And the variance of its score Var i Make a comprehensive selection. Among them, the selected strategy can first consider the performance score Score i Higher learning path while allowing the variance Var to be within an acceptable range i A slightly higher risk can be reflected in choosing a learning path with high volatility but relatively high returns. In other selection strategies, a learning path with low volatility is preferred in order to pursue stable learning results and returns.

[0109] Embodiment 3: This embodiment should be understood to include at least all the features of any one of the aforementioned embodiments and to be further improved thereon.

[0110] Further, as attached Figure 5 As shown, an exemplary embodiment of the computer system used in the educational system is described; the computer system 500 can be used to identify and judge the data storage, calculation and result output processes of each working module in the system.

[0111] Illustratively, computer system 500 includes a bus 502 or other communication mechanism for communicating information, and one or more processors 504 coupled with bus 502 for processing information; processor 504 may be, for example, one or more general-purpose microprocessors.

[0112] The computer system 500 also includes a main memory 506, such as a random access memory (RAM), a cache, and / or other dynamic storage device, coupled to the bus 502 for storing information and instructions to be executed by the processor 504; the main memory 506 may also be used to store temporary variables or other intermediate information during execution of the instructions to be executed by the processor 504; these instructions, when stored in a storage medium accessible to the processor 504, present the computer system 500 as a special-purpose machine customized to perform the operations specified in the instructions.

[0113] The computer system 500 may also include a read-only memory (ROM) 508 or other static storage device coupled to the bus 502 for storing static information and instructions for the processor 504; a storage device 510 such as a magnetic disk, optical disk, or USB drive (flash drive) will be coupled to the bus 502 for storing information and instructions.

[0114] Furthermore, coupled to the bus 502 may also include a display 512 for displaying various information, data, media, etc., and an input device 514 for allowing a user of the computer system 500 to control, manipulate, and / or interact with the computer system 500 .

[0115] A preferred way to interact with the management system may be through a cursor control device 516, such as a computer mouse or similar control / navigation mechanism.

[0116] Furthermore, the computer system 500 may further include a network device 518 coupled to the bus 502; wherein the network device 518 may include, for example, a wired network card, a wireless network card, a switching chip, a router, a switch, and other components; In general, the terms "engine," "component," "system," "database," etc., used herein may refer to logic embodied in hardware or firmware, or to a collection of software instructions, possibly with entry and exit points, written in a programming language such as Java, C, or C++; software components may be compiled and linked into executable programs, installed in a dynamic link library, or may be written in an interpreted programming language (such as BASIC, Perl, or Python); it will be understood that software components may be callable from other components or from themselves, and / or may be called in response to detected events or interrupts.

[0117] Software components configured to execute on a computing device may be provided on a computer-readable medium, such as a compact disc, digital video disc, flash drive, magnetic disk, or any other tangible medium, or as a digital download (and may be initially stored) in a compressed or installable format that requires installation, decompression, or decryption prior to execution); such software code may be stored in part or in whole on a memory device of the executing computing device for execution by the computing device; software instructions may be embedded in firmware, such as an EPROM; it will also be understood that hardware components may be composed of connected logic units (such as gates and flip-flops), and / or may be composed of programmable units (such as a programmable gate array or processor).

[0118] Computer system 500 may implement the techniques described herein using custom hard-wired logic, one or more ASICs or FPGAs, firmware, and / or program logic that, in combination with the computer system, renders computer system 500 a special-purpose computing device.

[0119] In accordance with one or more embodiments, the techniques herein are performed by computer system 500 in response to processor 504 executing one or more sequences of one or more instructions contained in main memory 506; such instructions may be read into main memory 506 from another storage medium, such as storage device 510; execution of the sequences of instructions contained in main memory 506 causes processor 504 to perform the process steps described herein; in alternative embodiments, hard-wired circuitry may be used in place of or in combination with software instructions.

[0120] As used herein, the term "non-transitory media" and similar terms refer to any medium that stores data and / or instructions that cause a machine to operate in a specific manner; such non-transitory media may include non-volatile media and / or volatile media; non-volatile media include, for example, optical or magnetic disks, such as storage device 510; and volatile media include dynamic memory, such as main memory 506.

[0121] Among them, common forms of non-transitory media include, for example, floppy disks, diskettes, hard disks, solid-state drives, magnetic tape or any other magnetic data storage medium, CD-ROMs, any other optical data storage medium, any physical medium with a pattern of holes, RAM, PROM and EPROM, FLASH-EPROM, NVRAM, any other memory chip or cartridge, and networked versions thereof.

[0122] Non-transient media are distinct from, but may be used in conjunction with, transmission media; transmission media participate in the transmission of information between non-transient media; for example, transmission media include coaxial cables, copper wires, and optical fibers, including the wires that comprise bus 502; transmission media may also take the form of sound or light waves, such as radio waves and infrared data communications.

[0123] Although the present application has been described above with reference to various embodiments, it will be understood that many changes and modifications may be made without departing from the scope of the present application. That is, the methods, systems, and devices discussed above are examples. Various configurations may omit, replace, or add various processes or components as appropriate. For example, in alternative configurations, the methods may be performed in an order different from that described, and / or various components may be added, omitted, and / or combined. Moreover, features described with respect to certain configurations may be combined in various other configurations, such as different aspects and elements of the configurations may be combined in a similar manner. Furthermore, as technology develops, the elements therein may be updated, i.e., many of the elements are examples and do not limit the scope of the present disclosure or the claims.

[0124] Specific details are given in the description to provide a thorough understanding of the exemplary configurations, including implementations. However, the configurations can be practiced without these specific details. For example, well-known circuits, processes, algorithms, structures, and techniques have been shown without unnecessary detail to avoid obscuring the configurations. This description provides only example configurations and does not limit the scope, applicability, or configurations of the claims. Instead, the foregoing description of the configurations will provide those skilled in the art with an enabling description for implementing the described techniques. Various changes may be made to the function and arrangement of the elements without departing from the spirit or scope of the present disclosure.

[0125] In summary, it is intended that the above detailed description be considered illustrative rather than restrictive, and it should be understood that the above embodiments are intended to be merely illustrative of the present invention and not to limit the scope of protection of the present invention. After reading the contents of the present invention, a skilled person may make various changes or modifications to the present invention, and these equivalent changes and modifications also fall within the scope defined by the claims of the present invention.

Claims

1. A personalized science education system driven by AI agents, characterized by: The said education system includes; The main agent is configured to interact with the user, collect the user's knowledge status, behavioral characteristics and learning preferences, and build a user model; One or more user twin agents, generated by training the master agent based on the user model, to simulate the user's learning behavior and learning outcomes under a specified learning path; A learning planning platform, configured to receive input from the master agent and / or the user twin agent and output a multi-stage personalized learning path and learning resources; Among them, the learning planning platform includes a database running a graph structure; the database provides content data support for the learning path; the main intelligent agent obtains the content data support for the learning path from the learning planning platform based on the user model, formulates multiple learning paths, and drives one or more user twin intelligent agents to perform simulation learning, and based on the results of the simulation learning, selects one or more learning paths to implement personalized teaching for the user.

2. The educational system according to claim 1, wherein: The main intelligent body includes the following functional modules: The large language model module is configured to perform natural language understanding and generation to enable language interaction and content expression with users; The context modeling and memory module is configured to record the user's historical interaction information, intent context, and behavior preferences, and maintain the user state; A graph interface module is configured to connect to the database and perform graph data retrieval, semantic query and learning path related information extraction; The reasoning and evaluation module is configured to analyze, judge, and optimize the current learning state based on the user model, database information, and task objectives; The interaction management module is used to control the dialogue process, information confirmation, feedback triggering and prompt delivery between the main intelligent agent and the user.

3. The educational system according to claim 1, wherein: The user model is constructed by the main agent, and the construction steps include: Collecting user behavior data during the learning process, including the user's learning time, answer accuracy, reaction time, error type, and operation path; Based on the behavioral data, analyze the user's mastery level and cognitive preferences on different knowledge points; Combining the user's historical learning records with the current learning task, a multi-dimensional vector representation is constructed to characterize the user's knowledge status, behavioral characteristics, and learning tendencies. The multi-dimensional vectorized representation is stored as a user model to drive the initialization of the user twin agent and the planning of personalized learning paths.

4. The educational system according to claim 1, wherein: The master agent determines the number of user twin agents to be generated based on multiple considerations. The steps in this process include: S100: Analyze the behavioral characteristics of the user model to identify the user's fluctuations in various learning behaviors and learning results; S200: Calculate the uncertainty score based on the volatility of the performance, which indicates the consistency of the user's learning behavior; S300: Determine the number of generated user twin agents based on the uncertainty score; By taking the above considerations into account, the generated user twin intelligent agent can cover the different possible behavior patterns of the user.

5. The educational system according to claim 1, wherein: The graph-based database includes one or more knowledge graphs; the knowledge graph is used to represent the structured knowledge relationships of learning content, and the knowledge graph is composed of: Multiple nodes, each node is used to represent an entity, including knowledge points, learning resources, learning goals, user status or skill modules; a plurality of edges, each edge being used to represent a semantic association relationship between two nodes, wherein the association relationship includes one of a prerequisite relationship, a dependency relationship, a similarity relationship, a feedback relationship, or a teaching sequence relationship; Each node and edge is attached with attribute information, including labels, contextual metadata, importance weights or vector embeddings, to support semantic retrieval and path planning.

6. The educational system according to claim 1, wherein: The master agent is further configured to formulate a plurality of candidate learning paths based on the user model and the database, which includes the following steps: Identifying in the database the node corresponding to the user's current knowledge state and the node corresponding to the target knowledge point; Based on the prerequisite relationship, semantic relevance and teaching sequence between each knowledge point in the database, multiple feasible paths from the current node to the target node are constructed.

Images