A language model-based adaptive learning state control method and system

Abstract

The application provides a language model-based adaptive learning state control method and system, the learning state control method comprises the following steps: starting a learning session, the core learning state comprises a teaching state and an understanding verification state, a knowledge point mapping table corresponding to a learning goal is constructed, and the knowledge point mapping table at least contains a mastery state level field of the knowledge point; teaching content is generated in the teaching state and output to the learner; the system monitors the trigger condition of state switching in real time, judges whether to switch the core learning state; if the understanding verification state is switched, the explanatory input of the learner is received; the explanatory input of the learner is analyzed; it is judged whether the semantic determination result meets the passing condition; if not, it is judged whether the failure condition is reached, if yes, the failure backtracking reteaching is triggered. The beneficial effect is that the automatic switching of the teaching and understanding verification process is realized, the multi-dimensional semantic judgment improves the accuracy of understanding evaluation, and the learning interruption risk is reduced through the failure backtracking mechanism.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence teaching technology, specifically to an adaptive learning state control method and system based on a language model. Background Technology

[0002] Artificial intelligence-based online education systems have been widely used in scenarios such as intelligent tutoring, personalized learning, and comprehension assessment. Current mainstream solutions mostly adopt linear question-and-answer or single dialogue interaction modes, directly output teaching content through language models, or insert test questions at preset fixed nodes to assess learners' comprehension through simple question-and-answer feedback or quantitative scoring.

[0003] However, such systems have significant technical flaws: First, they lack dynamic and refined state modeling of the learning process, making it difficult to effectively distinguish between key stages such as "knowledge point teaching" and "understanding verification," resulting in a lack of clear stage divisions and controllability in the learning process. Second, the judgment of learners' understanding level relies on single question-and-answer results or single-dimensional scoring, failing to construct a multi-dimensional, closed-loop verification logic, and thus cannot accurately reflect learners' true grasp of the core principles and conceptual relationships of knowledge points, leading to insufficient reliability in understanding assessment. Third, when learners experience misunderstandings, explanation failures, or exhibit negative emotions such as frustration or giving up, the system lacks an adaptive regression mechanism and targeted re-teaching strategies, making it unable to adjust teaching content and pace in a timely manner. This can easily lead to interruptions in the learning process and a decline in the learning experience, making it difficult to meet the actual needs of personalized learning and reliable understanding assessment.

[0004] Therefore, there is an urgent need for a technical solution based on the semantic judgment results of language models to realize the stage division, state judgment and automatic switching of the learning process, so as to solve the problem of unstable understanding and evaluation in the existing system and improve the stability and practicality of the teaching process. Summary of the Invention

[0005] To address the problems in the prior art, this invention provides an adaptive learning state control method based on a language model.

[0006] The present invention provides an adaptive learning state control method based on a language model, comprising the following steps: Step S1: The system starts a learning session. The learning session includes at least two core learning states, namely the teaching state and the understanding verification state, which are used to realize the teaching output of knowledge points and the verification of the degree of understanding, respectively. The current core learning state is initialized as the teaching state, and a knowledge point mapping table corresponding to the learning objectives is constructed. The knowledge point mapping table contains at least a knowledge point mastery level field. Step S2: In teaching mode, call the first language model instance, generate teaching content for the corresponding knowledge points according to the knowledge point mapping table, and output it to the learner; Step S3: The system monitors in real time whether the triggering conditions for state switching are met. If the triggering conditions are met, the system automatically switches to the corresponding core learning state. If the conditions are not met, the system maintains the current core learning state, thereby achieving continuous modeling of the learning process state. Step S4: If the core learning state is switched to the understanding and verification state, switch the current core learning state to the understanding and verification state and receive the learner's explanatory input. Step S5: Call the second language model instance to perform semantic analysis on the received learner interpretive input and output the comprehension judgment result; Step S6: Determine whether the judgment result meets the pass conditions; Step S7: If the semantic judgment result is not passed, then determine whether the failure condition has been met. If the failure condition has been met, trigger the failure backtracking and re-teaching, and switch the core learning state to the teaching state.

[0007] In a further improvement to this invention, in step S6, if the semantic judgment result is passed, the mastery level of the knowledge point is marked as understanding or transfer, the core learning state is switched to the teaching state, and the teaching of the next knowledge point begins.

[0008] The present invention is further improved in that, in step S7, if the failure condition is not met, the understanding and verification state is maintained.

[0009] In a further improvement to this invention, in step S5, the semantic analysis includes at least one or more of the following algorithms: Semantic anchor coverage analysis: Extract the set of key semantic anchors corresponding to the target knowledge points and determine whether the learner's explanation covers the semantic anchors; Conceptual Relationship Consistency Analysis: Determines whether the conceptual relationships described in the learner's explanation are consistent with the pre-defined relational structure of the target knowledge points; Simplification of expression analysis: Determine whether the explanation expresses the core meaning without directly referencing standard definitions or technical terms; Emotional semantic feature analysis: Detecting whether negative emotional semantic features appear during the interpretation process.

[0010] In a further improvement to this invention, in step S3, the triggering conditions include at least one or more of the following: the teaching content of the current knowledge point is completed; the learner actively requests to verify their understanding of the current knowledge point; and the mastery level of the corresponding knowledge point in the knowledge point mapping table is the initial value.

[0011] The present invention is further improved in that, in step S6, the conditions include: the learner's explanation of the content covers a preset proportion of semantic anchor points; the expression of conceptual relationships is consistent with the target knowledge points; and the explanation is completed without relying on technical terms.

[0012] In a further improvement, in step S7, the retrospective re-teaching process identifies the knowledge points that learners have not correctly understood based on the semantic analysis results, and makes targeted adjustments to the teaching content. The adjustments include: breaking down and explaining the uncovered semantic anchors; introducing more basic or specific examples; and adjusting the order or granularity of the teaching expression.

[0013] In a further improvement, in step S4, the explanatory input is limited to: descriptive expressions for assumed unknown objects; explanations that do not rely on the definition of technical terms; and restates of the basic structure and principles of knowledge points.

[0014] In a further improvement to this invention, in step S1, each knowledge point in the knowledge point mapping table is associated with a set of predefined semantic anchor points.

[0015] The present invention also provides an adaptive learning state control system based on semantic determination of a language model, comprising: The state initialization module is used to start a learning session, initialize the current learning state to the teaching state, and build a knowledge point mapping table corresponding to the learning objectives. The knowledge point mapping table must contain at least a knowledge point mastery level field. The trigger condition setting module is used to set the trigger conditions for state switching; The teaching status module is used to generate teaching content based on the current knowledge point and output it to the learner. The state switching module is used to monitor the fulfillment of trigger conditions in real time. If the trigger conditions are met, it automatically switches to the corresponding core learning state. If the conditions are not met, it maintains the current core learning state, thereby realizing continuous modeling of the learning process state. Explanatory Input Understanding Module: Used to perform semantic analysis on learners' explanatory input and output the understanding judgment result; The backtracking teaching module is used to trigger a failure backtracking and re-teaching when the learner's interpretive input meets the failure condition, and the core learning state is switched to the teaching state.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention provides an adaptive learning state control method based on a language model, which can effectively solve the problem that existing artificial intelligence teaching systems are difficult to meet the actual needs of personalized learning and reliable understanding assessment. By using this method, the automatic switching between teaching and understanding verification processes can be realized, improving process controllability. Based on the multi-dimensional semantic judgment of the language model, the accuracy and stability of understanding assessment can be improved. At the same time, the failure backtracking mechanism reduces the risk of learning interruption and improves user experience. Attached Figure Description

[0017] To more clearly illustrate the solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a block diagram of the adaptive learning state control system based on semantic determination of a language model according to the present invention. Figure 2 This is a flowchart of the adaptive learning state control method based on semantic determination of a language model according to the present invention. Detailed Implementation

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having” and any variations thereof in the specification, claims and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0020] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0021] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0022] like Figure 1 As shown, the present invention provides an adaptive learning state control system based on semantic determination of a language model: including a state initialization module, used to start a learning session, initialize the current learning state as a teaching state, and construct a knowledge point mapping table corresponding to the learning objective, wherein the knowledge point mapping table contains at least a knowledge point mastery level field. The trigger condition setting module is used to set the trigger conditions for state switching; The teaching status module is used to generate teaching content based on the current knowledge point and output it to the learner. The state switching module is used to monitor the fulfillment of trigger conditions in real time. If the trigger conditions are met, it automatically switches to the corresponding core learning state. If the conditions are not met, it maintains the current core learning state, thereby realizing continuous modeling of the learning process state. Explanatory Input Understanding Module: This module performs semantic analysis on learners' explanatory input and outputs the understanding judgment result. It also calls the language model to perform semantic analysis on the input through preset prompt templates. The backtracking teaching module is used to trigger backtracking and re-teaching when the learner's interpretive input meets the failure condition, and the core learning state is switched to the teaching state. Reversible evolution control module: Used to receive semantic analysis results and update the mastery level of each knowledge point through conditional branching logic.

[0023] The initial startup process involves the state initialization module defining the initial state and mapping relationship (Map) of the system, and inputting this initial information into the core knowledge point mapping table (knowledge_map).

[0024] The knowledge map, as the core data layer of the system, primarily undertakes three tasks: Provide the current knowledge point data to the teaching status module to drive the teaching process to the next stage; It has built-in logic for dynamically adjusting content, which can change the content of knowledge points in real time based on teaching feedback; The system saves the mastery level field for each knowledge point and updates the mastery level in real time, which can directly pinpoint students' comprehension obstacles and provide a basis for subsequent review teaching.

[0025] After receiving the knowledge point data, the teaching status module will trigger two actions in parallel: The language model instance is invoked to push teaching content to students through the output teaching content module, and the output progress is fed back synchronously. The driver trigger condition setting module presets rules to prepare for state switching.

[0026] The state transition determination module will perform conditional judgments based on preset rules: If the conditions are not met: Maintain the current teaching status and return the process to the teaching status module to continue; If the conditions are met: the system enters Feynman mode and triggers the interpretive input comprehension module; At the same time, when it is necessary to update the students' mastery level of knowledge points, it will be directly reflected in the knowledge point mapping table (knowledge_map).

[0027] The interpretive input comprehension module performs multi-dimensional semantic analysis on student input and determines whether the analysis passes or fails. If the judgment is successful: the result is fed back to the knowledge point mapping table (knowledge_map) to update the mastery level of the knowledge point; If the test fails: check if the failure threshold has been reached; if so, trigger the backtracking teaching module.

[0028] The retrospective teaching mechanism uses a knowledge map to locate students' comprehension obstacles, then triggers a retrospective adjustment of the teaching process, reorganizing the teaching content for the corresponding knowledge points.

[0029] like Figure 2 As shown, the present invention also provides an adaptive learning state control method based on a language model, such as... Figure 1 As shown, this method is used in the above system. The detailed implementation method of adaptive learning state control based on language model in this example is as follows: 1. The system starts a learning session, which includes at least two core learning states: teaching state and comprehension verification state. These are used to implement the teaching output of knowledge points and the verification of comprehension level, respectively. The current core learning state is initialized as the teaching state, and a knowledge point mapping table corresponding to the learning objectives is constructed. The knowledge point mapping table must contain at least a knowledge point mastery level field.

[0030] Each knowledge point in the knowledge point mapping table is associated with a set of predefined semantic anchor points.

[0031] The mastery status level field includes Unlearned, Pending Verification, and Mastered.

[0032] The teaching state is used to output teaching content of target knowledge points to learners; the understanding and verification state is used to receive reverse interpretive input from learners and perform semantic judgment.

[0033] The system provides each knowledge point Maintain a knowledge point state vector that is updated in real time. .

[0034] Mastering the status level : Scalar value, mapped to five levels from 0 to 4.

[0035] Table 1: Mastery Level Explanation at each level Coverage The calculation formula is: .

[0036] Independence The language model is based on the "proportion of text not quoted". Scoring is used to identify mechanical repetition.

[0037] consistency : Boolean value (True / False), determines whether the causal logic in the input matches the preset logical structure.

[0038] II. In teaching mode, the system according to The corresponding teaching content template calls the first language model instance, generates the corresponding knowledge point teaching content according to the knowledge point mapping table, and outputs it to the learner.

[0039] The first language model instance is a language model with complete knowledge expression capabilities. Its prompt configuration is set to output the concept definition, basic principle or example description of knowledge points in a structured manner.

[0040] Initial Mastery Level If the value is 0, the generated teaching content is organized based on the current knowledge point and its dependencies in the knowledge point mapping table and output to the learner.

[0041] In teaching mode: The system invokes the first language model for teaching and outputs structured explanation content based on the structured semantic structure, including: explanation of preconditions, explanation of cause and effect, and supplementary explanation of examples.

[0042] Third, the system monitors in real time whether the trigger conditions for state switching are met. If the trigger conditions are met, the system automatically switches to the corresponding core learning state. If the conditions are not met, the system maintains the current core learning state, thereby achieving continuous modeling of the learning process state.

[0043] Triggering conditions include at least one or more of the following: the teaching content for the current knowledge point has been completed; the learner actively requests to verify their understanding of the current knowledge point; the mastery level of the corresponding knowledge point in the knowledge point mapping table. The initial value is 0.

[0044] After the teaching is completed, the system will not immediately enter the verification state. Instead, it will call the "routing module" to continuously monitor the learner's input and determine whether the state switching conditions are triggered.

[0045] The system determines that the state switching conditions are met and switches the core learning state to the understanding and verification state.

[0046] 4. If the core learning state switches to the understanding and verification state, switch the current core learning state to the understanding and verification state and receive the learner's explanatory input.

[0047] Explanatory input is limited to: descriptive expressions for assumed unknown objects; explanations that do not rely on the definitions of technical terms; and restates of the basic structure and principles of knowledge points.

[0048] This reverse interpretive input is used to simulate the learner's process of re-expressing knowledge points to support the verification of the level of understanding.

[0049] 5. Call the second language model instance to perform semantic analysis on the received learner interpretive input and output the comprehension judgment result.

[0050] The output comprehension assessment result should include at least the following structured metrics: key knowledge point coverage. Independence of expression and logical consistency .

[0051] When the learner input is received, the interpretive input understanding module fills the input text into the following prompt template for the second language model instance: "Please analyze the learner input, compare it with the anchor set {A1, A2...}, and output a JSON-formatted metric: {coverage:%, independence: 0-1, logic_consistent: bool}" The reversible evolution control module performs calculations based on the JSON metrics returned by the prompt template: Transition logic: If and and Then execute In this way, when learners express the core meaning without relying on standard definitions, the system triggers... arrive The leap.

[0052] Degradation logic (reversible mechanism): If detected in subsequent dialogues If a "logic conflict" flag appears, the system will execute immediately. This enables forced rollback of the state.

[0053] This method enables the system to have the downgrade logic to identify "false knowledge": Logical conflict downgrade: If a learner demonstrates a causal description in subsequent higher-order validations that contradicts previously mastered knowledge, the system determines that the original mastery level is invalid and performs a forced downgrade of the cognitive level (e.g., ).

[0054] Negative Emotion Regression: When the system detects severe frustration, confusion, or feelings of giving up in learners, it proactively reduces the difficulty of the task and regresses to the basic teaching state, reorganizing the teaching path.

[0055] Based on the analysis results, the system assigns a mastery level to each knowledge point in the knowledge point mapping table. Update the status. For example: initial, ambiguous, repetitive, understood, transition. These status levels are dynamically updated after each round of explanation, and may either increase or decrease.

[0056] And the system will based on the current Dynamically select prompt words strategy. For example, if (Repetition level) The explanatory input understanding module will automatically push "anti-template verification prompts" to the second language model, forcing the model to generate an interactive instruction that "please do not use technical terms to explain".

[0057] When the mastery of all core knowledge points meets the preset passing conditions (such as...) ≥3), the system performs overall verification. Through comprehensive semantic analysis, it determines whether the learner has completed the explanation of the entire set of principles without relying on terminology. If successful, the learning session is considered to have globally converged and the process ends.

[0058] Semantic analysis includes at least one or more of the following algorithms: Semantic anchor coverage analysis: Extract the set of key semantic anchors corresponding to the target knowledge points and determine whether the learner's explanation covers the semantic anchors; Conceptual Relationship Consistency Analysis: Determines whether the conceptual relationships described in the learner's explanation are consistent with the pre-defined relational structure of the target knowledge points; Simplification of expression analysis: Determine whether the explanation expresses the core meaning without directly referencing standard definitions or technical terms; Emotional semantic feature analysis: Detecting whether negative emotional semantic features appear during the interpretation process.

[0059] Negative emotional semantic features include frustration, denial, or giving up.

[0060] Once the system enters the understanding and verification state, it receives interpretive input from the learner, such as: "Newton's first law states that an object will maintain its state of motion." The system performs structural mapping analysis on the interpretable input, specifically as follows: 1. Constructing the learner's semantic structure The system calls the semantic analysis module to extract the structure of the input: extracting semantic nodes and causal or logical relationships between nodes to construct the learner's semantic structure.

[0061] 2. Structural Comparison Analysis The system compares and analyzes learners' semantic structures with structured semantic structures, including: (1) Calculation of anchor point coverage Determine whether the expression contains A1, A2, and A3. (2) Relationship Consistency Analysis Determine whether the premise causal relationship A1→A2 is established; (3) Analysis of expression independence Determine whether to directly quote from the textbook; (4) Emotional semantic analysis Determine if negative expressions such as "chaos" or "abandonment" are present; The system generates knowledge point state vectors. When learners can independently connect real-life examples and correctly describe causal logic, the system directly upgrades their mastery level to the "understanding" or "transfer" level.

[0062] By constructing a structured semantic structure, we can effectively avoid misjudgments based on superficial similarity, and improve the accuracy of understanding and evaluation without relying on single text similarity.

[0063] In this specific embodiment, semantic analysis is performed using a second language model instance or by multiple language model instances in parallel, wherein at least one of them is a restricted semantic model instance, which is configured not to actively output the standard definition of the target knowledge point, but to guide the learner to continue to interpret by posing basic questions.

[0064] VI. Determine whether the judgment result meets the passing conditions.

[0065] The conditions for passing the test include: the learner's explanation covers a predetermined proportion of semantic anchor points; the expression of conceptual relationships is consistent with the target knowledge points; and the explanation is completed without relying on technical jargon.

[0066] If the semantic judgment result is passed, the mastery level of the knowledge point is marked as understanding or transfer, the core learning state is switched to teaching state, and the teaching of the next knowledge point begins.

[0067] In this embodiment, the system uses the knowledge point state vector Determine whether state transition is allowed.

[0068] If the following conditions are met: 1. Node coverage ≥ preset threshold; 2. Complete causal relationship; 3. Independent expression; 4. Stable emotional state.

[0069] Then, the mastery level of that knowledge point will be raised to the "understanding" or "transfer" level, and the core learning state will be switched to the teaching state, and the teaching of the next knowledge point will begin. If not satisfied: The system will remain in the understanding and verification state and will not update the mastery level of that knowledge point.

[0070] Unlike traditional scoring mechanisms, this embodiment requires complete structural relationships, rather than just textual similarity.

[0071] 7. If the semantic judgment result is "not passed", then determine whether the failure condition has been met. If the failure condition has been met, then trigger the failure backtracking and re-teaching, and switch the core learning state to the teaching state.

[0072] The system determines failure conditions and whether they meet a failure threshold. The failure threshold includes at least one or more of the following: a certain knowledge point remains stuck at a certain level for an extended period. or Continuous decline; repeated failure to pass comprehension judgment; cumulative emotional semantic features exceeding the preset threshold.

[0073] If the failure condition is not met, the understanding and verification state is maintained.

[0074] In this implementation, when a learner fails repeatedly or expresses negative emotions, the system calculates the difference between the current learner's semantic structure and the structured semantic structure.

[0075] For example: missing A1 (assuming zero external force) and missing the causal relationship A1→A2.

[0076] The system will forcibly switch the core learning state to the repair state.

[0077] The system can prevent learning interruptions by introducing the analysis of emotional variable characteristics.

[0078] In retrospective re-teaching, the system calculates the difference between the current cognitive state and the target structure, locates the missing semantic anchors, and forcibly reconstructs the teaching path to enter a targeted remedial teaching phase, rather than repeating the entire process.

[0079] The adjustments include: breaking down and explaining the uncovered semantic anchors; introducing more basic or specific examples; and adjusting the order or granularity of the teaching presentation.

[0080] The adjusted teaching content is regenerated using a first language model instance and output to learners, while simultaneously updating the status field of the corresponding knowledge point in the knowledge point mapping table.

[0081] For example, in the repair state: the system does not repeat the complete teaching content, but only explains the missing nodes; it emphasizes the missing causal relationships; and it provides a special example to illustrate that "zero external force is a prerequisite".

[0082] After the repair is completed, the structure will be verified again in the understanding and verification state.

[0083] In the repair mode, learners can receive precise supplementary lessons without repeating the entire lesson, thus improving learning efficiency.

[0084] The learning process ends when the learning objective is achieved or the session termination conditions are met.

[0085] Compared with the prior art, the present invention has the following innovative features: 1. By using an explicit learning state machine, the automatic switching between teaching and understanding verification processes can be achieved, improving process controllability; 2. Improve the accuracy of understanding assessment through multi-dimensional semantic judgment based on language models; 3. By implementing a failure backtracking mechanism, the risk of learning interruption is reduced, and the user experience is improved; A closed-loop control mechanism based on semantic structure consistency was constructed, which is applicable to various teaching scenarios and has good scalability and engineering feasibility.

[0086] As can be seen from the above, the present invention provides an adaptive learning state control method based on a language model, which can effectively solve the problem that existing artificial intelligence teaching systems are unable to meet the actual needs of personalized learning and reliable understanding assessment. By using this method, the automatic switching between teaching and understanding verification processes can be realized, improving process controllability. Based on the multi-dimensional semantic judgment of the language model, the accuracy and stability of understanding assessment can be improved. At the same time, the failure backtracking mechanism reduces the risk of learning interruption and improves user experience.

[0087] The specific embodiments described above are preferred embodiments of the present invention and are not intended to limit the specific scope of the present invention. The scope of the present invention includes, but is not limited to, these specific embodiments. All equivalent changes made in accordance with the present invention are within the protection scope of the present invention.

Claims

1. An adaptive learning state control method based on a language model, characterized in that, Includes the following steps: Step S1: The system starts a learning session. The learning session includes at least two core learning states, namely a teaching state and a comprehension verification state, which are used to realize the teaching output of knowledge points and the verification of the degree of comprehension, respectively. The current core learning state is initialized as the teaching state, and a knowledge point mapping table corresponding to the learning objectives is constructed. The knowledge point mapping table at least includes a knowledge point mastery level field. Step S2: In teaching mode, call the first language model instance, generate teaching content for the corresponding knowledge points according to the knowledge point mapping table, and output it to the learner; Step S3: The system monitors in real time whether the triggering conditions for the state switching are met. If the triggering conditions are met, the system automatically switches to the corresponding core learning state. If the conditions are not met, the system maintains the current core learning state, thereby achieving continuous modeling of the learning process state. Step S4: If the core learning state is switched to the understanding and verification state, switch the current core learning state to the understanding and verification state and receive the learner's explanatory input. Step S5: Call the second language model instance to perform semantic analysis on the received learner interpretive input and output the comprehension judgment result; Step S6: Determine whether the judgment result meets the pass conditions; Step S7: If the semantic judgment result is not passed, then determine whether the failure condition has been met. If the failure condition has been met, trigger the failure backtracking and re-teaching, and switch the core learning state to the teaching state.

2. The adaptive learning state control method based on a language model according to claim 1, characterized in that: In step S6, if the semantic judgment result is passed, the mastery level of the knowledge point is marked as understanding or transfer, the core learning state is switched to the teaching state, and the teaching of the next knowledge point begins.

3. The adaptive learning state control method based on a language model according to claim 1, characterized in that: In step S7, if the failure condition is not met, the understanding verification state is maintained.

4. The adaptive learning state control method based on a language model according to claim 1, characterized in that: In step S5, the semantic analysis includes at least one or more of the following algorithms: Semantic anchor coverage analysis: Extract the set of key semantic anchors corresponding to the target knowledge point and determine whether the learner's explanation covers the semantic anchors; Conceptual Relationship Consistency Analysis: Determines whether the conceptual relationships described in the learner's explanation are consistent with the pre-defined relational structure of the target knowledge points; Simplification of expression analysis: Determine whether the explanation expresses the core meaning without directly referencing standard definitions or technical terms; Emotional semantic feature analysis: Detecting whether negative emotional semantic features appear during the interpretation process.

5. The adaptive learning state control method based on a language model according to claim 1, characterized in that: In step S3, the triggering conditions include at least one or more of the following: the teaching content for the current knowledge point is completed; the learner actively requests to understand and verify the current knowledge point. The mastery level of the corresponding knowledge point in the knowledge point mapping table is the initial value.

6. The adaptive learning state control method based on a language model according to claim 1, characterized in that: In step S6, the passing conditions include: the learner's explanation of the content covers a preset proportion of semantic anchors; the expression of conceptual relationships is consistent with the target knowledge points; and the explanation is completed without relying on technical terms.

7. The adaptive learning state control method based on a language model according to claim 1, characterized in that: In step S7, the retrospective re-teaching process identifies the knowledge points that learners have not correctly understood based on the semantic analysis results, and makes targeted adjustments to the teaching content. The adjustments include: breaking down and explaining the uncovered semantic anchors; introducing more basic or specific examples; and adjusting the order or granularity of the teaching expression.

8. The adaptive learning state control method based on a language model according to claim 1, characterized in that: In step S4, the interpretive input is defined as: a descriptive expression oriented towards a hypothetical unknown object; Explanations that do not rely on definitions of technical terms; A restatement of the basic structure and principles of the knowledge points.

9. The adaptive learning state control method based on a language model according to claim 1, characterized in that: In step S1, each knowledge point in the knowledge point mapping table is associated with a set of predefined semantic anchor points.

10. An adaptive learning state control system based on semantic determination of a language model, used to implement the adaptive learning state control method based on a language model as described in any one of claims 1-9, characterized in that: The adaptive learning state control system based on language model semantic determination includes: The state initialization module is used to start a learning session, initialize the current learning state to the teaching state, and construct a knowledge point mapping table corresponding to the learning objectives. The knowledge point mapping table at least includes a knowledge point mastery level field. The trigger condition setting module is used to set the trigger conditions for state switching; The teaching status module is used to generate teaching content based on the current knowledge point and output it to the learner. The state switching module is used to monitor the fulfillment of the triggering conditions in real time. If the triggering conditions are met, the module automatically switches to the corresponding core learning state. If the conditions are not met, the module maintains the current core learning state, thereby enabling continuous modeling of the learning process state. Explanatory Input Understanding Module: Used to perform semantic analysis on learners' explanatory input and output the understanding judgment result; The backtracking teaching module is used to trigger a failure backtracking and re-teaching when the learner's interpretive input meets the failure condition, and the core learning state is switched to the teaching state.

Images