Intelligent homework tutoring system and method based on multi-modal interaction and adaptive learning

Abstract

The invention relates to the technical field of artificial intelligence education, in particular to an intelligent homework tutoring system and method based on multi-modal interaction and adaptive learning, and the system comprises a multi-modal input analysis module, a cognitive state dynamic evaluation module, an intelligent decision explanation engine, and a learning effect visualization closed-loop module. The method has the beneficial effects that a three-dimensional adaptive system of explanation granularity-presentation form-interaction frequency is used, and teaching strategies such as visual derivation, concept metaphor and the like can be automatically matched according to cognitive styles of students. Secondly, a'backtracking reinforcement-lateral expansion 'double-intervention mechanism is innovatively used, and the recurrence rate of similar errors is effectively reduced through error real-time detection and correlation knowledge contrast teaching. And finally, constructing a dynamic knowledge graph and an interactive learning report, realizing visual tracking of a learning path, and helping students to establish a systematic knowledge framework.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence education, in particular to an intelligent homework tutoring system and method based on multi-modal interaction and adaptive learning. BACKGROUND

[0002] With the development of artificial intelligence technology and large language models and inference models, the development of artificial intelligence in the field of education is advancing at an unprecedented speed. From basic education to higher education, AI technology is reshaping teaching methods, optimizing learning experiences, and promoting educational equity and individualization. Currently, online education platforms and intelligent tutoring systems have been widely used, but there are still significant limitations:

[0003] Unidirectional: Traditional systems mostly use a one-way knowledge infusion model, lacking dynamic perception of students' cognitive state.

[0004] Insufficient adaptability: The interaction with students is single, and the teaching strategy is fixed. Existing intelligent tutoring tools usually use linear explanation strategies, which cannot adjust the teaching logic in real time (such as backtracking reinforcement or correlation expansion) according to student feedback.

[0005] Closed loop missing: Learning effect evaluation lags behind, lacking systematic data analysis and visual feedback, making it difficult for students to intuitively grasp their knowledge weaknesses and progress trajectory.

[0006] Although educational large models and multi-modal technologies are gradually applied, existing solutions still fail to effectively integrate dynamic cognitive diagnosis, adaptive strategy generation, real-time interaction, and closed-loop optimization, resulting in insufficient tutoring efficiency and individualization. SUMMARY

[0007] The purpose of the present application is to provide an intelligent homework tutoring system and method based on multi-modal interaction and adaptive learning to solve the problems raised in the background.

[0008] To achieve the above purpose, the present application provides the following technical solution: an intelligent homework tutoring system based on multi-modal interaction and adaptive learning, comprising:

[0009] Multi-modal input analysis module: integrates text input box, speech recognition and image acquisition module, supports students to input questions in text, speech and image; relies on multi-modal large model to perform word segmentation, part-of-speech tagging, named entity recognition and semantic vector construction on input content, completes semantic analysis; combined with sentiment analysis technology to capture student emotional state; for mathematical formulas, geometric figures and other subject-specific images, use specially trained recognition models for feature extraction and analysis, convert image information into computer understandable symbols and structures, accurately interpret image questions combined with context semantics, and align with education knowledge graph, output structured problem representation;

[0010] Cognitive state dynamic evaluation module: based on the interactive record, the process of multi-dimensional evaluation data such as job behavior trajectory and test response heat map, the individual knowledge graph of students is dynamically constructed by using the probabilistic graph model, and the cognitive topology network with breadth and depth is formed; Through the emotion calculation module and the cognitive load monitoring algorithm, the emotional fluctuations and thinking obstacles in the learning process are captured in real time; A three-dimensional dynamic adjustment system of "explanation granularity-presentation form-interactive frequency" is constructed, and teaching strategies are intelligently matched according to different learner characteristics, and the dynamic iteration optimization of learning strategies is realized through the metacognitive feedback closed loop;

[0011] Intelligent decision-making explanation engine: Through multi-modal feedback analysis technology, the multi-modal feedback information such as student's voice tone, facial expression, mouse operation or pen touch interaction is captured in real time; When detecting the correct answer of the student, automatically promote the knowledge node; When the error or confusion signal is recognized, trigger the "backtracking reinforcement-lateral expansion" double strategy, form the "accurate diagnosis-layered intervention" mechanism through variant explanation and comparison teaching of related knowledge network;

[0012] Learning effect visualization closed loop module: record the multi-modal interaction data throughout the process, generate an interactive learning report containing error trace, thinking path analysis and personalized improvement suggestions; Build a dynamic knowledge mastery map to visually present the mastery of each knowledge point by students in different colors and node sizes, and show the learning mistakes and progress trajectory; Through the same type training generator, the variant questions are intelligently matched based on the error question mode, and the gradient training scheme is automatically generated to strengthen the consolidation of knowledge points.

[0013] Preferably, in the multi-modal input analysis module, the multi-modal large model deeply excavates semantic relationship through attention mechanism, constructs accurate semantic vector, and completes high-precision semantic analysis; For the recognition model of subject-specific images, the image information is converted into symbols and structures through feature extraction and analysis, and the image problem is accurately interpreted in combination with the context semantics, the recognition result is aligned with the education knowledge graph, the core knowledge points of the problem are extracted and the structured problem representation is output.

[0014] Preferably, in the cognitive state dynamic evaluation module, the three-dimensional dynamic adjustment system includes: activating the "thinking scaffold" mode for students with weak foundation, decomposing complex concepts into interactive modular components, and assisting with progressive derivation animations; Real-time generation of dynamic topology chart and three-dimensional cognitive map for visual learners; For students with abstract thinking, adopt the concept metaphor mode to build an understanding bridge through interdisciplinary analogy; The metacognitive feedback closed loop continuously tracks the effectiveness of the learning strategy through the Bayesian optimization algorithm, realizing the dynamic iteration of "monitoring-intervention-optimization".

[0015] Preferably, in the intelligent decision-making explanation engine, the "backtracking reinforcement-lateral expansion" dual strategy includes: backtracking reinforcement: conducting variant explanation on previously explained steps, helping students re-understand knowledge points by changing the conditions, expression methods or problem-solving ideas of the problem; lateral expansion: calling the associated knowledge network for comparative teaching, introducing other concepts, theorems or examples related to the current knowledge point, and deepening students' understanding and memory of the knowledge point through comparative analysis.

[0016] Preferably, in the learning effect visualization closed loop module, the working process of the same type training generator includes: feature extraction of the students' wrong questions, analysis of the knowledge points, question types and difficulty elements involved in the wrong questions; screening questions with similar characteristics from the question bank, and processing the questions in variant, such as changing the data and adjusting the problem angle; automatically generating gradient training scheme, and pushing training questions to students in order from easy to difficult, and strengthening knowledge point consolidation.

[0017] A method for an intelligent homework tutoring system based on multi-modal interaction and adaptive learning, comprising the following steps:

[0018] Multi-modal input analysis: build a multi-channel input interface, integrate a text input box, a voice recognition module and an image acquisition module, support students to input questions in text, voice and image; rely on multi-modal large models to perform word segmentation, part-of-speech tagging, named entity recognition and semantic vector construction on input content, complete high-precision semantic analysis; use sentiment analysis technology to capture the emotional state of students when asking questions; for subject-specific images, use specially trained recognition models for feature extraction and analysis, convert image information into symbols and structures, accurately interpret the question in combination with the context semantics, and align with the educational knowledge graph to output structured question representation;

[0019] Cognitive state dynamic evaluation: based on multi-dimensional process evaluation data such as interactive records, homework behavior trajectories and test response heat maps, use a probabilistic graph model to dynamically build a student individual knowledge graph; through the emotion calculation module and the cognitive load monitoring algorithm, real-time capture emotional fluctuations and thinking obstacles in the learning process; build a "explanation granularity-presentation form-interactive frequency" three-dimensional dynamic adjustment system, intelligently match teaching strategies for different learners, and realize dynamic iterative optimization of learning strategies through metacognitive feedback closed loop;

[0020] Intelligent decision-making explanation: real-time capture of multi-modal feedback information such as students' voice tone, facial expression, mouse operation or pen touch interaction through multi-modal feedback analysis technology; when detecting correct answers from students, automatically advance the knowledge node; when recognizing errors or confusion signals, trigger the "backtracking reinforcement-lateral expansion" dual strategy, form a "precise diagnosis-layered intervention" mechanism through variant explanation and comparative teaching of associated knowledge network;

[0021] Learning effect visualization closed loop: record multi-modal interaction data throughout the process, generate interactive learning reports containing error traceability, thought path analysis, and personalized improvement suggestions; build a dynamic knowledge mastery map to visually present students' mastery of each knowledge point and show learning mistakes and progress trajectories; through the same type of training generator, intelligently match variable problems based on error patterns, automatically generate gradient training plans, and strengthen knowledge point consolidation;

[0022] Full-process intelligent integration: through the cooperation of the above steps, the full-process intelligence of "multi-modal input → adaptive explanation → closed-loop feedback" is realized, and the traditional homework tutoring is upgraded to a complete learning system.

[0023] Preferably, in the multi-modal input analysis step, the multi-modal large model uses attention mechanism to deeply mine semantic relationships, constructs accurate semantic vectors, and completes high-precision semantic analysis; for subject-specific image recognition models, image information is converted into symbols and structures through feature extraction and analysis, and the recognition results are aligned with the education knowledge graph to extract the core knowledge points of the question and output structured question representations.

[0024] Preferably, in the cognitive state dynamic evaluation step, the three-dimensional dynamic adjustment system includes: activating the "thinking scaffold" mode for students with weak foundation, breaking down complex concepts into interactive modular components, and assisting with progressive deduction animations; real-time generation of dynamic topology charts and three-dimensional cognitive maps for visual learners; for students with abstract thinking, use the concept metaphor mode to build understanding bridges through interdisciplinary analogy; the meta-cognitive feedback loop continuously tracks the effectiveness of learning strategies through the Bayesian optimization algorithm, realizing the dynamic iteration of "monitoring - intervention - optimization".

[0025] Preferably, in the intelligent decision explanation step, the "backtracking reinforcement - lateral expansion" double strategy includes: backtracking reinforcement: variable explanation of previously explained steps, by changing the conditions, expression methods or problem solving ideas of the problem, to help students re-understand the knowledge points; lateral expansion: call the associated knowledge network for comparative teaching, introduce other concepts, theorems or examples related to the current knowledge points, and deepen students' understanding and memory of knowledge points through comparative analysis.

[0026] Preferably, in the learning effect visualization closed loop step, the workflow of the same type of training generator includes: feature extraction of students' wrong questions, analysis of the knowledge points, question types and difficulty elements involved in the wrong questions; screen similar characteristic questions from the question bank, and perform variable processing on the questions, such as changing data and adjusting problem angles; automatically generate gradient training plans and push training questions to students in order from easy to difficult to strengthen knowledge point consolidation.

[0027] Compared with the prior art, the present application has the following advantages:

[0028] The intelligent homework tutoring system and method based on multi-modal interaction and adaptive learning provided by the application use a three-dimensional adaptive system of "explanation granularity-presentation form-interactive frequency", which can automatically match visual deduction, concept metaphor and other teaching strategies according to the cognitive style of students. Secondly, the dual intervention mechanism of "backtracking reinforcement-lateral expansion" is innovatively used, and through real-time error detection and associated knowledge comparison teaching, the recurrence rate of similar errors is effectively reduced. Finally, a dynamic knowledge graph and an interactive learning report are constructed to realize the visual tracking of the learning path and help students establish a systematic knowledge framework. The system upgrades the traditional correction to a closed-loop learning ecology of "evaluation-intervention-feedback", providing a scalable technical paradigm for intelligent education and having good popularization value. BRIEF DESCRIPTION OF DRAWINGS

[0029] Figure 1 The method flowchart of the application. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical scheme of the application clear, complete and more clearly understand the advantages, the embodiments of the application will be further described in detail below in combination with the drawings. It should be understood that the specific embodiments described herein are part of the embodiments of the application, not all embodiments, and are only used to explain the embodiments of the application, and do not limit the embodiments of the application. All other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0031] Embodiment one, please refer to Figure 1 The application provides a technical scheme: an intelligent homework tutoring system based on multi-modal interaction and adaptive learning, comprising:

[0032] The multi-modal input analysis module: integrates a text input box, a speech recognition module and an image acquisition module, supports students to input problems in the form of text, speech and image; relies on a multi-modal large model to perform word segmentation, part-of-speech tagging, named entity recognition and semantic vector construction on the input content, completes semantic analysis; combines emotion analysis technology to capture the emotional state of students; for mathematical formulas and geometric figure discipline-specific images, a specially trained recognition model is used for feature extraction and analysis, image information is converted into computer understandable symbols and structures, image problems are accurately interpreted in combination with context semantics, and are aligned with an education knowledge graph, and structured problem representation is output; the multi-modal large model deeply excavates semantic relationships through an attention mechanism, constructs accurate semantic vectors, and completes high-precision semantic analysis; for discipline-specific image recognition models, image information is converted into symbols and structures through feature extraction and analysis, image problems are accurately interpreted in combination with context semantics, the recognition results are aligned with the education knowledge graph, core knowledge points of the problems are extracted, and structured problem representation is output.

[0033] Cognitive state dynamic assessment module: Based on the interactive records, job behavior trajectory, test response heat map multi-dimensional process evaluation data, use probability graph model to dynamically construct student individual knowledge graph, form a cognitive topology network with breadth and depth; Through the emotional computing module and the cognitive load monitoring algorithm, real-time capture of emotional fluctuations and thinking obstacles in the learning process; Build a "explanation granularity-presentation form-interactive frequency" three-dimensional dynamic adjustment system, intelligently match teaching strategies for different learners, and realize dynamic iterative optimization of learning strategies through metacognitive feedback closed loop; The three-dimensional dynamic adjustment system includes: activating the "thinking scaffold" mode for students with weak foundation, breaking down complex concepts into interactive modular components, and assisting with progressive derivation animations; Real-time generation of dynamic topology charts and three-dimensional cognitive maps for visual learners; For students with abstract thinking, use the concept metaphor mode to build understanding bridges through interdisciplinary analogy; The metacognitive feedback closed loop continuously tracks the effectiveness of learning strategies through Bayesian optimization algorithm, realizing the dynamic iteration of "monitoring-intervention-optimization".

[0034] Intelligent decision-making explanation engine: Real-time capture of students' voice tone, facial expressions, mouse operations or pen touch interaction multi-modal feedback information through multi-modal feedback analysis technology; When detecting correct answers, automatically advance the knowledge node; When identifying errors or confusion signals, trigger the "backtracking reinforcement-lateral expansion" double strategy, through variant explanation and comparison teaching of related knowledge network, form the mechanism of "accurate diagnosis-layered intervention"; "Backtracking reinforcement-lateral expansion" double strategy includes: backtracking reinforcement: variable explanation of previous steps, change the conditions, expression methods or problem solving ideas of the problem to help students understand the knowledge points again; Lateral expansion: call related knowledge network for comparison teaching, introduce other concepts, theorems or examples related to the current knowledge point, deepen students' understanding and memory of knowledge points through comparison and analysis.

[0035] Learning effect visualization closed loop module: Record multi-modal interaction data throughout the process, generate interactive learning reports containing error trace, thinking path analysis, and personalized improvement suggestions; Build a dynamic knowledge mastery map to visually present students' mastery of each knowledge point with different colors and node sizes, and show learning mistakes and progress trajectory; Through the same type training generator, variable questions are intelligently matched based on the error mode, and gradient training schemes are automatically generated to strengthen knowledge point consolidation; The workflow of the same type training generator includes: feature extraction of students' wrong questions, analysis of the knowledge points, question types and difficulty elements involved in the wrong questions; From the question bank, filter out questions with similar characteristics, and process the questions, such as changing the data and adjusting the problem angle; Automatically generate gradient training schemes, push training questions to students in order from easy to difficult, and strengthen knowledge point consolidation.

[0036] In Example Two, based on Example One, an intelligent homework tutoring method based on multi-modal interaction and adaptive learning is proposed, including the following steps:

[0037] First step: multi-modal input analysis

[0038] The system constructs a multi-channel input interface, integrates a text input box, a speech recognition module, and an image acquisition module, supports students to input problems in various ways such as text, speech, and image, and realizes flexible collection of multi-modal data. Relying on the powerful processing capability of the multi-modal large model, the system performs basic operations such as word segmentation, part-of-speech tagging, and named entity recognition on the content after text and speech transcription. In this process, with the help of attention mechanism, the model deeply excavates semantic relationship, constructs accurate semantic vector, and completes high-precision semantic analysis. At the same time, the system uses sentiment analysis technology to sensitively capture the emotional state of students when asking questions, providing an important basis for the subsequent development of personalized tutoring strategies.

[0039] In the image recognition and understanding link, the multi-modal large model uses specially trained recognition models to extract and analyze features of images such as mathematical formulas and geometric figures, and converts image information into symbols and structures that computers can understand. Combined with the context semantics, the model accurately interprets the problems implied by the image, and aligns the recognition results with the education knowledge graph. Again, with the help of attention mechanism, the core knowledge points of the problem are extracted from complex information, and finally the structured problem representation is output, laying a solid foundation for subsequent cognitive assessment, solution strategy, etc. Links ensure that the system can completely and accurately understand the questions raised by students.

[0040] Second step: dynamic assessment of cognitive state

[0041] The cognitive state dynamic assessment module innovatively integrates semantic analysis engine and knowledge point nanoscale disassembly technology by constructing an intelligent education big model, realizing precise navigation of the tutoring process. Based on multi-dimensional process evaluation data, including interaction records, homework behavior trajectory, test response heat map, etc., the system uses a probabilistic graph model to dynamically construct a student individual knowledge graph, forming a cognitive topology network with breadth and depth. This graph not only accurately quantifies the knowledge mastery level, but also captures emotional fluctuations and thought barriers in the learning process through emotion calculation module and cognitive load monitoring algorithm.

[0042] In terms of adaptive adjustment mechanism, the system innovatively constructs a "explanation granularity-presentation form-interactive frequency" three-dimensional dynamic adjustment system. According to different learner characteristics, the system intelligently matches the optimal teaching strategy: for students with weak foundation, it automatically activates the "thinking scaffold" mode, breaking down complex concepts into interactive modular components and providing progressive derivation animations; for visual learners, it generates dynamic topological charts and three-dimensional cognitive maps in real time; for students with abstract thinking, it uses the concept metaphor mode to build understanding bridges through interdisciplinary analogy.

[0043] The system specially designs a metacognitive feedback closed loop, continuously tracks the effectiveness of learning strategies through Bayesian optimization algorithm, and realizes the dynamic iteration of "monitoring-intervention-optimization". This kind of neural plasticity intelligent design promotes the transformation of education services from standardized delivery to personalized growth, and provides each learner with an AI tutor with full-stack cognitive ability.

[0044] Step 3: Intelligent decision-making explanation engine:

[0045] This engine captures student status in real time through multi-modal feedback analysis technology. During the interaction between students and the system, the system collects multi-modal feedback information such as students' voice tone, facial expressions, mouse operations or pen touch interactions through sensors and data analysis technology.

[0046] When the student correctly answers, the system automatically advances to the next knowledge node and enters the explanation of the next knowledge point according to the preset knowledge path. When the error or confusion signal is recognized, the "backtracking reinforcement-lateral expansion" double strategy is triggered. In terms of backtracking reinforcement, the system re-explains the previous steps by changing the conditions, expressions or problem-solving ideas of the questions, helping students understand the knowledge points again; in terms of lateral expansion, it calls the associated knowledge network for comparison teaching, introduces other concepts, theorems or examples related to the current knowledge point, and deepens students' understanding and memory of the knowledge point through comparison and analysis, forming an intelligent decision-making mechanism of "accurate diagnosis-layered intervention".

[0047] Step 4: Learning effect visualization closed loop:

[0048] The system records multi-modal interaction data throughout the process, including students' question input, system's explanation content, students' feedback information, etc. After cleaning and preprocessing, these data are sent to the education big model for deep mining. The big model generates an interactive learning report through natural language processing and data analysis technology. The report includes error tracing, which analyzes the reasons for students' errors, whether it is due to weak knowledge grasp, incorrect problem-solving approach or carelessness, etc.; thinking path analysis, which restores students' thinking track by recording their operation steps and thinking process during problem-solving; personalized improvement suggestions, which develop targeted learning plans and improvement measures for students based on their learning situation and problem analysis.

[0049] The system constructs a dynamic knowledge mastery graph based on multi-modal interaction data. The graph intuitively presents the students' mastery of each knowledge point with different colors and node sizes, clearly shows the learning mistakes and progress trajectory, helps students understand the learning situation comprehensively, and effectively improves the students' metacognitive ability.

[0050] In addition, the auxiliary module intelligently matches the variable question based on the wrong question mode through the same question type training generator. The system first extracts the characteristics of the students' wrong questions, analyzes the knowledge points, question types, difficulty and other elements involved in the wrong questions; then selects questions with similar characteristics from the question bank, and appropriately processes the questions, such as changing data, adjusting the problem angle, etc.; finally, a gradient training scheme is automatically generated, and the training questions are pushed to the students in order from easy to difficult, strengthening the consolidation of knowledge points.

[0051] Although embodiments of the present application have been shown and described, it will be understood by those having ordinary skill in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent homework tutoring system based on multi-modal interaction and adaptive learning, characterized in that: Comprise: Multi-modal input analysis module: integrated text input box, speech recognition and image acquisition module, support students to input questions in the form of text, voice, image; rely on multi-modal large model to perform word segmentation, part-of-speech tagging, named entity recognition and semantic vector construction on input content, complete semantic analysis; combined with emotion analysis technology to capture student emotional state; for mathematical formulas, geometry discipline specific images, use specially trained recognition model for feature extraction and analysis, convert image information into computer understandable symbols and structures, combined with context semantics to accurately interpret image problems, and align with the educational knowledge graph, output structured problem representation; Cognitive state dynamic evaluation module: based on multi-dimensional process evaluation data such as interactive records, homework behavior trajectory and test response heat map, use probabilistic graph model to dynamically build student individual knowledge graph, form a cognitive topology network with breadth and depth; through emotion computing module and cognitive load monitoring algorithm, real-time capture emotional fluctuations and thinking obstacles in the learning process; build a "explanation granularity-presentation form-interactive frequency" three-dimensional dynamic adjustment system, intelligently match teaching strategies for different learners, and realize dynamic iterative optimization of learning strategies through metacognitive feedback loop; Intelligent decision-making explanation engine: through multi-modal feedback analysis technology, real-time capture of student voice tone, facial expression, mouse operation or pen touch interaction multi-modal feedback information; when detecting correct response, automatically advance knowledge nodes; when recognizing errors or confusion signals, trigger "backtracking reinforcement-lateral expansion" dual strategy, through variant explanation and comparison teaching of associated knowledge network, form "accurate diagnosis-layered intervention" mechanism; Learning effect visualization closed loop module: record multi-modal interaction data throughout the process, generate interactive learning report containing error trace, thinking path analysis, and personalized improvement suggestions; build dynamic knowledge mastery map to visually present students' mastery of each knowledge point with different colors and node sizes, show learning mistakes and progress trajectory; through the same type training generator, intelligently match variant questions based on error patterns, automatically generate gradient training schemes to strengthen knowledge point consolidation. 2.The intelligent homework tutoring system based on multi-modal interaction and adaptive learning of claim 1, wherein: In the multi-modal input analysis module, the multi-modal large model deeply excavates semantic relationships through attention mechanism, constructs accurate semantic vectors, and completes high-precision semantic analysis; for discipline-specific image recognition models, image information is converted into symbols and structures through feature extraction and analysis, and combined with context semantics to accurately interpret image problems, align the recognition results with the educational knowledge graph, extract the core knowledge points of the problem and output structured problem representation. 3.The intelligent homework tutoring system based on multi-modal interaction and adaptive learning of claim 2, wherein: In the cognitive state dynamic evaluation module, the three-dimensional dynamic adjustment system includes: activating the "thinking scaffold" mode for students with weak foundation, breaking down complex concepts into interactive modular components, and assisting with progressive derivation animations; for visual learners, real-time generation of dynamic topology charts and three-dimensional cognitive maps; for abstract thinkers, use the concept metaphor mode to build understanding bridges through interdisciplinary analogy; the meta-cognitive feedback loop continuously tracks the effectiveness of learning strategies through Bayesian optimization algorithms, achieving a dynamic iteration of "monitoring-intervention-optimization". 4.The intelligent homework tutoring system based on multi-modal interaction and adaptive learning of claim 3, wherein: In the intelligent decision-making explanation engine, the "retrospective reinforcement-lateral expansion" dual strategy includes: retrospective reinforcement: variable explanation of previously explained steps, by changing the conditions, expression methods or problem-solving ideas of the problem, to help students better understand the knowledge points; lateral expansion: call the associated knowledge network for comparative teaching, introduce other concepts, theorems or examples related to the current knowledge point, and deepen students' understanding and memory of the knowledge point through comparative analysis. 5.The intelligent homework tutoring system based on multi-modal interaction and adaptive learning of claim 4, wherein: In the learning effect visualization closed loop module, the working process of the same type of training generator includes: feature extraction of students' wrong questions, analysis of the knowledge points, question types and difficulty factors involved in the wrong questions; selecting questions with similar characteristics from the question bank, and processing the questions by changing the data and adjusting the problem angle; automatically generating a gradient training plan and pushing training questions to students in order from easy to difficult to strengthen knowledge point consolidation.

6. A method for the intelligent homework tutoring system based on multi-modal interaction and adaptive learning as claimed in claim 5, wherein: The method comprises the following steps: Multi-modal input analysis: build a multi-channel input interface, integrate a text input box, a voice recognition module, and an image acquisition module, support students to input questions in text, voice, and image modes; rely on multi-modal large models to perform word segmentation, part-of-speech tagging, named entity recognition, and semantic vector construction on input content to complete high-precision semantic analysis; use sentiment analysis technology to capture the emotional state of students when asking questions; for subject-specific images, use a specially trained recognition model to extract and analyze features, convert image information into symbols and structures, and accurately interpret the question in combination with the context semantics, and align with the educational knowledge graph to output structured question representation; Cognitive state dynamic evaluation: based on multi-dimensional process evaluation data such as interactive records, homework behavior trajectories, and test response heat maps, use a probabilistic graph model to dynamically build a student individual knowledge graph; Through the emotion calculation module and the cognitive load monitoring algorithm, real-time capture of emotional fluctuations and thinking obstacles in the learning process; build a "explanation granularity-presentation form-interactive frequency" three-dimensional dynamic adjustment system, intelligently match teaching strategies for different learners, and realize dynamic iteration optimization of learning strategies through meta-cognitive feedback loop; Intelligent decision-making explanation: through multi-modal feedback analysis technology, real-time capture of students' voice tone, facial expressions, mouse operations or pen touch interaction multi-modal feedback information; when detecting correct answers, automatically advance the knowledge node; when recognizing errors or confusion signals, trigger the "retrospective reinforcement-lateral expansion" dual strategy, through variable explanation and comparative teaching of associated knowledge networks, form a "precise diagnosis-layered intervention" mechanism; Learning effect visualization closed loop: record multi-modal interaction data throughout the process, generate interactive learning reports containing error traceability, thought path analysis, and personalized improvement suggestions; build a dynamic knowledge mastery map to visually present students' mastery of each knowledge point and show learning mistakes and progress trajectories; generate gradient training schemes by matching variable problems based on error patterns using the same problem type training generator, and automatically generate gradient training schemes to reinforce knowledge points; Full-process intelligent integration: through the cooperation of the above steps, realize the full-process intelligence of "multi-modal input → adaptive explanation → closed-loop feedback", and upgrade traditional homework tutoring to a complete learning system.

7. A method according to claim 6, characterised in that: In the multi-modal input analysis step, the multi-modal large model uses attention mechanisms to deeply analyze semantic relationships, construct accurate semantic vectors, and complete high-precision semantic analysis; for subject-specific image recognition models, feature extraction and analysis are used to convert image information into symbols and structures, and the recognition results are aligned with the education knowledge graph to extract core knowledge points and output structured problem representations.

8. A method according to claim 7, characterised in that: In the cognitive state dynamic evaluation step, the three-dimensional dynamic adjustment system includes: activating the "thinking scaffold" mode for students with weak foundations, breaking down complex concepts into interactive modular components, and providing progressive deduction animations; generating dynamic topology charts and three-dimensional cognitive maps for visual learners in real time; for abstract thinkers, use the concept metaphor mode to build understanding bridges through interdisciplinary analogies; meta-cognitive feedback loop continuously tracks the effectiveness of learning strategies using Bayesian optimization algorithms to achieve a dynamic iteration of "monitoring - intervention - optimization".

9. A method according to claim 8, characterised in that: In the intelligent decision-making explanation step, the "backtracking reinforcement - lateral expansion" dual strategy includes: backtracking reinforcement: variable explanation of previously explained steps, by changing the conditions, expression methods or problem-solving ideas of the problem, to help students understand the knowledge points again; lateral expansion: call the associated knowledge network for comparative teaching, introduce other concepts, theorems or examples related to the current knowledge points, and deepen students' understanding and memory of knowledge points through comparative analysis.

10. A method according to claim 9, wherein: In the learning effect visualization closed loop step, the workflow of the same problem type training generator includes: feature extraction of students' wrong questions, analysis of the knowledge points, question types, and difficulty elements involved in the wrong questions; filter out questions with similar characteristics from the question bank, and process the questions by changing the data and adjusting the problem angle; automatically generate gradient training schemes, and push training questions to students in order from easy to difficult to reinforce knowledge point consolidation.

Images