Personalized English grammar learning system and method based on artificial intelligence

Abstract

The invention relates to the technical field of artificial intelligence education, and discloses a personalized English grammar learning system based on artificial intelligence. Comprising a multi-modal data acquisition module, a deep user portrait construction module, a dynamic association knowledge graph module, a multi-modal intelligent diagnosis module, a meta-cognitive guide type learning path planning module, a personalized content generation module, an emotional real-time feedback module, a meta-cognitive training module, a social cooperation module and a closed-loop optimization module. The multi-modal data acquisition module is used for acquiring multi-dimensional and multi-modal data of a learner, and covering related data such as basic information. According to the personalized English grammar learning system and method based on artificial intelligence, by setting a multi-modal data acquisition module and a deep user portrait construction module, basic information, cognitive features, meta-cognitive levels and emotional preferences of learners are captured in all directions; and constructing a dynamically updated deep user portrait by means of a multi-dimensional data fusion algorithm and a time sequence analysis algorithm.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence education technology, specifically to a personalized English grammar learning system and method based on artificial intelligence. Background Technology

[0002] With the deep integration of artificial intelligence and educational technology, English grammar learning systems have gradually introduced intelligent algorithms to optimize the learning experience. They are transforming from the traditional, single-mode "rule explanation + rote memorization" to an intelligent paradigm of "data-driven + personalized adaptation." Leveraging core technologies such as natural language processing, machine learning, and computer vision, existing systems have achieved basic functions such as automatic answer correction, preliminary identification of knowledge gaps, and visualization of learning progress. This breaks down the limitations of time and space, allowing learners to engage in self-directed learning anytime, anywhere. Whether completing grammar-specific exercises through a mobile app or obtaining instant feedback through a webpage, the intervention of intelligent technology significantly improves the convenience and efficiency of learning, especially demonstrating outstanding advantages in the integration of massive question banks and the standardized breakdown of knowledge points, providing more flexible learning options for learners with different levels of proficiency.

[0003] However, existing English grammar learning systems still have significant shortcomings: data collection is mostly limited to surface information such as test results, failing to fully capture learners' cognitive characteristics and emotional preferences; knowledge is presented in a fragmented and static manner, lacking contextual connections and interdisciplinary integration; error diagnosis can only identify explicit problems, making it difficult to pinpoint deeper root causes such as cognitive biases and insufficient metacognition; learning paths lack dynamic adjustment and cannot adapt to personalized needs; feedback is mostly simple result notification, lacking emotional guidance and targeted remediation; and there is a lack of precise collaborative learning mechanisms, making it difficult for system models to achieve continuous optimization through data iteration, thus failing to meet learners' deep and personalized learning needs. Summary of the Invention

[0004] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a personalized English grammar learning system and method based on artificial intelligence. It boasts advantages such as precise multi-dimensional data collection, dynamic user profiling, comprehensive knowledge graph association, root cause-based error diagnosis, personalized learning paths, emotional feedback, collaborative learning process, and closed-loop system optimization. This solves the problems of existing systems, such as limited data collection, isolated knowledge presentation, inaccurate error diagnosis, rigid learning paths, lack of targeted feedback, absence of collaborative mechanisms, and insufficient continuous optimization capabilities, making it difficult to meet learners' personalized and in-depth grammar learning needs.

[0005] (II) Technical Solution To achieve the aforementioned goals of precise multi-dimensional data collection, dynamic user profiling, three-dimensional knowledge graph association, root cause analysis of errors, personalized learning paths, emotional feedback, collaborative learning process, and closed-loop system optimization, this invention provides the following technical solution: a personalized English grammar learning system based on artificial intelligence, comprising a multimodal data collection module, a deep user profiling construction module, a dynamically associated knowledge graph module, a multimodal intelligent diagnosis module, a metacognitive guided learning path planning module, a personalized content generation module, an emotional real-time feedback module, a metacognitive training module, a social collaboration module, and a closed-loop optimization module; The multimodal data acquisition module is used to collect learners' multidimensional and multimodal data, covering basic information, learning objectives, cognitive feature data, multimodal learning behavior data, and metacognitive related data; The deep user profile building module uses information from the multimodal data acquisition module and employs a multi-dimensional data fusion algorithm to build a dynamically updated deep user profile. The dynamic association knowledge graph module constructs a four-dimensional dynamic knowledge graph consisting of grammatical rules, scenario applications, interdisciplinary topics, and metacognitive cues. The multimodal intelligent diagnostic module uses a multimodal fusion deep learning model to accurately diagnose and analyze the root causes of grammatical errors in the learner's multimodal input content. The metacognitive guided learning path planning module is based on deep user profiles, dynamic association knowledge graphs, and multimodal intelligent diagnostic results. It uses reinforcement learning algorithms to integrate metacognitive guidance strategies and plans dynamic personalized learning paths. The personalized content generation module: Based on the metacognitive guided learning path, it adopts generative adversarial networks (GAN) combined with multimodal content generation technology to customize personalized learning content; The emotional real-time feedback module uses emotional computing and personalized feedback algorithms to provide emotional and targeted real-time feedback on learners' learning process and practice results; The metacognitive training module includes four sub-modules: self-planning, self-monitoring, self-reflection, and self-assessment. The social collaboration module supports precise matching of learning partners based on deep user profiles, formation of thematic learning groups, and multi-form collaborative learning, and incorporates incentive mechanisms such as points rewards, selection of reflection stars, and collaborative task achievement badges. The closed-loop optimization module continuously collects learners' interaction data in each module and iteratively optimizes the system model by combining machine learning algorithms.

[0006] Preferably, the basic information includes age, education stage, English proficiency level (CEFR A1-C2), and learning scenario; the learning objectives include short-term and long-term goals; the cognitive characteristic data is obtained through specialized tests, including cognitive style, information processing speed, and attention span; the multimodal learning behavior data includes the content and process data of text input, voice input, and handwriting input, which are automatically recorded by the system and collected through multi-terminal (mobile phone, tablet, computer) adaptation; the metacognitive related data includes learners' error reflection records, learning plan execution status, self-assessment results, and frequency and depth of questions raised, which are obtained through a combination of user-initiated input and automatic system analysis.

[0007] Preferably, the profile includes basic dimensions (age, English proficiency, learning goals), cognitive dimensions (cognitive style, information processing characteristics, attention level), knowledge dimensions (mastery of grammar points, distribution of weaknesses, error type preferences), metacognitive dimensions (self-monitoring ability, depth of reflection, planning and execution ability), and emotional dimensions (intensity of learning motivation, resilience, and preference for feedback type), and the dynamic changes of the profile are tracked through a time-series analysis algorithm.

[0008] Preferably, the dynamically associated knowledge graph module uses major grammatical categories (tense, voice, clauses, etc.) as first-level nodes, specific grammatical rules (simple present tense, relative clauses, etc.) as second-level nodes, and core test points, common mistakes, and fixed collocations as third-level nodes. It matches diverse application scenario tags and corresponding corpus examples to each third-level node; it associates with interdisciplinary themes, integrating high-frequency grammatical applications under those themes; it embeds metacognitive prompts, including common reflection directions for the knowledge point, self-checking methods for common mistakes, and learning strategy suggestions (metacognitive prompt dimension); and, based on massive real-time corpora (academic literature, social media, exam questions), it dynamically updates the association strength of knowledge points, adds emerging application scenarios, and high-frequency error points through machine learning algorithms.

[0009] Preferably, the multimodal fusion deep learning model is based on a bidirectional LSTM+CRF architecture that integrates visual feature extraction (for handwriting input, including writing conventions and symbol usage) and speech feature extraction (for speech input, including tense, intonation, and grammatical pauses). This enables dual recognition of both explicit and implicit grammatical errors. Error types are further subdivided into five major categories and 32 subcategories: grammatical rules, fixed collocations, contextual adaptation, cognitive biases, and expression norms. By combining deep user profiles and dynamically associated knowledge graphs, the model identifies the knowledge gaps, cognitive shortcomings, and metacognitive deficiencies corresponding to the errors, generating a precise diagnostic report that includes error location, error type, root cause analysis, and links to related knowledge points.

[0010] Preferably, the dynamic personalized learning path has three objectives: knowledge acquisition, ability enhancement, and metacognitive development, and combines the following principles: adapting learning forms according to cognitive styles; designing reflection tasks according to metacognitive levels; prioritizing core knowledge points according to learning objectives and scenario tags; the dynamic personalized learning path includes the order of knowledge point learning, allocation of learning time, combination of practice types, arrangement of reflection tasks, and setting of review cycles, and has a dynamic adjustment mechanism.

[0011] Preferably, the personalized content generation module includes knowledge point explanation materials, graded practice questions, metacognitive training tasks, and scenario-based application tasks, and all content is labeled with corresponding knowledge graph nodes and metacognitive prompts.

[0012] Preferably, the feedback forms in the emotional real-time feedback module include: result feedback, error feedback, emotional feedback, and interactive feedback. At the same time, it will push knowledge point links, metacognitive training tasks, and consolidation exercises related to the error points.

[0013] Preferably, the self-planning submodule, based on in-depth user profiles and learning goals, guides learners to develop personalized learning plans, provides plan templates and adjustment suggestions, tracks plan execution, and provides reminders; the self-monitoring submodule monitors in real time for distractions and progress delays during the learning process, helping learners get back into focus through pop-up prompts and suggestions for adjusting learning pace; the self-reflection submodule provides structured error reflection tools, knowledge point summary templates, and learning strategy optimization record forms; and the self-assessment submodule periodically generates multi-dimensional self-assessment scales, guiding learners to conduct self-assessments against their profiles and learning goals, and analyzing gaps based on system assessment results.

[0014] The steps for using AI-based personalized English grammar learning are as follows: S1: Multimodal data acquisition and profile initialization; After registering and logging in, learners fill in basic information and learning objectives, complete cognitive feature tests and initial multimodal diagnostic tasks, and the system integrates data to generate an initial in-depth user profile; S2: Multimodal intelligent diagnosis and root cause localization; the system completes the initial calibration by dynamically associating knowledge graphs, performs multimodal analysis on the initial diagnosis task, identifies error types and root causes, and generates a diagnosis report that includes a list of weaknesses and capability ratings; S3: Metacognitive Guided Path Generation; The path planning module generates personalized paths based on profiles, diagnostic reports, and maps, including the order of knowledge points, learning duration, and reflection tasks, and supports learners to adjust and provide feedback for optimization. S4: Personalized learning and metacognitive training; learners acquire suitable learning content according to the path, complete graded exercises and scenario-based tasks, the system provides emotional real-time feedback, and the metacognitive training module guides self-planning, monitoring and reflection. S5: Learning Data Collection and Dynamic Optimization: The closed-loop optimization module continuously collects data such as learning behavior, practice quality, and reflection depth, updates user profiles and knowledge graphs, and adjusts learning paths and content parameters. S6: Phase Evaluation and Closed-Loop Iteration: After each learning unit, conduct multi-dimensional evaluations, combine the evaluation results to optimize subsequent learning plans, repeat steps S4 and S5, forming a closed loop of collection-diagnosis-learning-feedback-optimization until the learning objectives are achieved.

[0015] (III) Beneficial Effects Compared with existing technologies, this invention provides a personalized English grammar learning system and method based on artificial intelligence, which has the following beneficial effects: 1. This AI-based personalized English grammar learning system and method captures learners' basic information, cognitive characteristics, metacognitive level, and emotional preferences in an all-round way by setting up a multimodal data acquisition module and a deep user profile construction module. It also constructs a dynamically updated deep user profile with the help of multi-dimensional data fusion algorithms and time-series analysis algorithms.

[0016] 2. This AI-based personalized English grammar learning system and method, by setting up a four-dimensional dynamically related knowledge graph module and a multimodal intelligent diagnostic module, breaks through the limitations of traditional static knowledge graphs with a four-dimensional architecture of "grammar rules - scenario application - interdisciplinary topics - metacognitive prompts", achieving deep binding and dynamic updates of knowledge with scenarios, disciplines and metacognition; on the other hand, it adopts a bidirectional LSTM+CRF architecture multimodal fusion deep learning model to accurately identify explicit and implicit grammatical errors and trace the root causes at the knowledge, cognition and metacognition levels.

[0017] 3. This AI-based personalized English grammar learning system and method integrates metacognitive development into the entire learning process by setting up a metacognitive guided learning path planning module and a metacognitive training module. It dynamically plans the learning path based on the learner's cognitive style, metacognitive level, and learning goals, and combines it with systematic training in four sub-modules: self-planning, monitoring, reflection, and evaluation. This not only achieves accurate transmission of grammatical knowledge, but also helps learners build self-learning abilities.

[0018] 4. This AI-based personalized English grammar learning system and method, through the setting of an emotional real-time feedback module, a social collaboration module, and a closed-loop optimization module, not only adapts to learners' emotional preferences to provide multi-form, targeted feedback, but also creates a collaborative learning atmosphere through precise partner matching, topic groups, and incentive mechanisms. At the same time, it relies on full-process data collection and machine learning algorithms to continuously iterate the system model. Attached Figure Description

[0019] Figure 1 This is a schematic block diagram of the system structure in this invention; Figure 2 This is a schematic block diagram illustrating the closed-loop operation of the present invention. Figure 3 This is a flowchart of the method in this invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figure 1-3 The AI-based personalized English grammar learning system includes a multimodal data acquisition module, a deep user profile construction module, a dynamic knowledge graph module, a multimodal intelligent diagnosis module, a metacognitive guided learning path planning module, a personalized content generation module, an emotional real-time feedback module, a metacognitive training module, a social collaboration module, and a closed-loop optimization module. In the case implementation, the multimodal data acquisition module was used to collect learners' multidimensional and multimodal data, covering basic information, learning objectives, cognitive characteristic data, multimodal learning behavior data, and metacognitive data. Basic information included age, education stage, English proficiency level (CEFR A1-C2), and learning scenario, all obtained through user-initiated form submissions. Learning objectives included short-term and long-term goals, which could be entered and selected by the user in the system's goal setting module. Cognitive characteristic data was obtained through specialized tests, including cognitive style, information processing speed, and attention span, obtained through the system's built-in standardized cognitive assessment question bank, combined with analysis of user test responses and reaction times. Multimodal learning behavior data included text input, voice input, and... The content and process data of handwritten input are automatically recorded by the system and collected through multi-terminal (mobile phone, tablet, computer) adaptation. Text input data is captured in real time through the terminal input box, capturing the content and key timing. Voice input data is captured through the terminal microphone, capturing audio and extracting voice features. Handwritten input data is captured through tablet stylus or computer graphics tablet, capturing writing trajectory, pen pressure, and writing time. Metacognitive data includes learners' error reflection records, learning plan execution status, self-assessment results, and frequency and depth of questions raised. This data is obtained through a combination of user-initiated input (error reflection text, self-filled assessment forms) and automatic system analysis (study plan completion rate statistics, frequency and semantic analysis of question submission records). In the case implementation, the deep user profile construction module, based on information from the multimodal data acquisition module, uses a multi-dimensional data fusion algorithm to construct dynamically updated deep user profiles. These profiles include basic dimensions (age, English proficiency, learning goals), cognitive dimensions (cognitive style, information processing characteristics, attention level), knowledge dimensions (mastery of grammar points, distribution of weaknesses, error type preferences), metacognitive dimensions (self-monitoring ability, depth of reflection, planning and execution ability), and affective dimensions (intensity of learning motivation, resilience, and preference for feedback). The basic dimension data is directly taken from the learner's registration information and learning goal settings. Cognitive dimension data is derived through quantitative analysis of specific test results; for example, field-dependent / field-independent cognitive styles are determined through cognitive tests with embedded scenario-based options, and information processing characteristics are comprehensively extracted from data such as the number of pauses and task completion time in multimodal learning behavior. Knowledge dimension data is based on error types identified by the multimodal intelligent diagnostic module, the knowledge point associations corresponding to incorrect questions, and statistics on practice completion quality, clarifying the learner's performance in terms of tense, etc. The study analyzes differences in the mastery of grammar modules such as clauses and the tendency of high-frequency errors. The metacognitive dimension is constructed by analyzing the completeness of learners' error reflection records (such as whether they include the reasons for errors and improvement methods), the execution rate of learning plans (such as whether daily learning tasks are completed as preset), and the consistency between self-assessment results and system assessment results. The affective dimension is generated by combining the learners' task persistence time (resilience), interaction frequency with feedback (feedback preference), and willingness to continue learning after achieving goals (intensity of learning motivation) and other behavioral performance and preference survey results during the learning process. The dynamic changes of the profile are tracked through time-series analysis algorithms. For example, the mastery rate of grammar knowledge points and the distribution of weak points are updated in real time as the learning progress progresses. The metacognitive ability rating is adjusted according to the depth of improvement of error reflection records. The affective dimension parameters are optimized based on emotional feedback in learning tasks (such as persistence after consecutive errors). The feature labels of the cognitive and affective dimensions are supplemented by the interaction performance in the social collaboration module to ensure that the profile always accurately matches the learner's real-time status and provides dynamic and comprehensive core basis for subsequent accurate diagnosis and path planning. In the implementation of the case, the dynamic association knowledge graph module constructs a four-dimensional dynamic knowledge graph consisting of grammatical rules, scenario applications, interdisciplinary topics, and metacognitive prompts. The module uses major grammatical categories (tense, voice, clauses, etc.) as first-level nodes, specific grammatical rules (simple present tense, relative clauses, etc.) as second-level nodes, and core test points, common mistakes, and fixed collocations as third-level nodes. It matches diverse application scenario tags and corresponding corpus examples to each third-level node; it associates with interdisciplinary topics, integrating high-frequency grammatical applications under those topics; and it embeds metacognitive prompts, including common reflection directions for the knowledge point, self-checking methods for common mistakes, and learning strategy suggestions (metacognitive prompt dimension). Simultaneously, based on massive real-time corpora (academic literature, social media, and past exam questions), it dynamically updates the association strength of knowledge points, adds emerging application scenarios, and identifies high-frequency error points through machine learning algorithms. The three-level nodes of the grammar rules dimension are determined by extracting data from authoritative grammar textbooks, past exam outlines, and teaching practice data to ensure that core test points and common mistakes are not missed, and fixed collocations are all derived from standardized corpora and mainstream authoritative reference materials. The tags of the scenario application dimension are set according to the common learning scenarios of learners (academic English, daily communication, exam preparation, etc.). The corpus examples prioritize typical expressions in real contexts, such as matching dialogue examples of shopping and asking for directions in daily communication scenarios, and selecting examples such as paper abstracts and report texts in academic writing scenarios. The association of interdisciplinary topics is based on the statistical results of frequently used grammatical structures in various fields. For example, the science and technology topic focuses on integrating the application forms of passive voice and complex relative clauses, and the business topic focuses on the use of conditional adverbial clauses and subjunctive mood in negotiations and emails. Metacognitive prompts are generated by combining the root causes of common learners' errors and summaries of efficient learning experiences. For example, for the knowledge point of "subjunctive mood", the prompt is "Reflect on whether there is a hypothetical meaning in the context. When checking yourself, you can try to replace it with real mood to verify the rationality of the expression." By communicating with the multimodal intelligent diagnostic module and the personalized content generation module in real time, the knowledge graph nodes are precisely matched with learners' error types, cognitive styles, and learning goals. For example, for learners preparing for exams, the application scenarios and interdisciplinary cases corresponding to the "high-frequency test points" tags in the third-level nodes are activated first, and the corpus correlation strength of the application scenario dimension is strengthened for field-dependent learners. At the same time, based on the group learning data fed back by the closed-loop optimization module, the correlation weight of nodes in each dimension is continuously adjusted. For example, when the error rate of a certain emerging application scenario (such as short video oral communication) increases significantly, the scenario tag and corresponding corpus are quickly added to ensure that the knowledge graph always fits the actual application needs of learners, providing dynamic and efficient knowledge support for accurate diagnosis, path planning, and personalized content generation.

[0022] In the implementation of the case, the multimodal intelligent diagnosis module adopts a multimodal fusion deep learning model to accurately diagnose and analyze the root causes of grammatical errors in learners' multimodal input. The multimodal fusion deep learning model is based on a bidirectional LSTM+CRF architecture that integrates visual feature extraction (for handwriting input, writing norms, and symbol usage) and speech feature extraction (for speech input, tense, intonation, and grammatical pauses). It achieves dual recognition of explicit and implicit grammatical errors and subdivides error types into five major categories and 32 subcategories: grammatical rule type, fixed collocation type, context adaptation type, cognitive bias type, and expression norm type. Combined with deep user profiles and dynamic related knowledge graphs, it locates the knowledge weaknesses, cognitive shortcomings, and metacognitive deficiencies corresponding to the errors and generates an accurate diagnostic report that includes error location, error type, root cause analysis, and links to related knowledge points. The identification of explicit grammatical errors focuses on direct violations of grammatical rules, such as tense confusion, subject-verb disagreement, and misuse of prepositions, which can be quickly determined through rule matching. It relies on a bidirectional LSTM+CRF architecture to perform sequence annotation and verification of grammatical structures in text, speech, and handwriting input. The identification of implicit errors combines multimodal contextual information and deep features, such as grammatically correct text input that does not conform to the interdisciplinary topic expression habits, semantic ambiguity caused by mismatch between tense and intonation changes in spoken input, and improper logical connection in handwriting practice. It achieves accurate capture by integrating visual and speech features with text semantic analysis. The 32 subcategories of the five major error categories are subdivided based on high-frequency error scenarios in teaching practice. For example, the grammatical rule category includes 8 subcategories such as tense misuse and voice confusion, and the cognitive bias category includes 5 subcategories such as scenario judgment errors and logical connection errors. Each subcategory corresponds to a specific third-level node and metacognitive prompt information in the dynamically associated knowledge graph. By associating and matching the diagnostic results of a multimodal fusion deep learning model with deep user profiles, and combining learners' cognitive styles, knowledge gap distribution, and metacognitive ability ratings, the root causes of errors are identified. For example, if context-dependent learners frequently make context-appropriate errors, the root cause is determined to be insufficient cognitive style and context-related ability. If learners with low metacognitive abilities repeatedly make similar grammatical rule errors, the root cause points to a lack of self-reflection and self-checking ability. At the same time, the association information of the knowledge points corresponding to the errors in the dynamically associated knowledge graph is called into the knowledge point links in the diagnostic report, so as to achieve precise matching of "error-root cause-remedial resources", providing a direct basis for subsequent metacognitive guided learning path planning and personalized content generation.

[0023] In the case implementation, the metacognitive guided learning path planning module, based on deep user profiles, dynamically linked knowledge graphs, and multimodal intelligent diagnostic results, employs reinforcement learning algorithms to integrate metacognitive guidance strategies and plan dynamic personalized learning paths. These paths aim for three objectives: knowledge acquisition, ability enhancement, and metacognitive development, adhering to the following principles: adapting learning formats to cognitive styles; designing reflective tasks based on metacognitive levels; prioritizing core knowledge points based on learning objectives and scenario tags. The dynamic personalized learning path includes the order of knowledge point learning, allocation of learning time, combination of practice types, arrangement of reflective tasks, and setting of review cycles, and also possesses a dynamic adjustment mechanism. Specifically, "adapting learning formats to cognitive styles" means: for field-dependent learners, emphasizing scenario-based case studies, interdisciplinary thematic exercises, and group collaborative tasks to enhance understanding through interaction with peers in authentic contexts; for field-independent learners, providing logical breakdowns of grammar rules, knowledge point framework diagrams, and self-directed inquiry tasks to meet their independent analysis needs; for visually-oriented learners, generating illustrated explanations, animated demonstrations, and mind maps; for auditory-oriented learners, providing audio explanations, dialogue audio, and listening-related grammar exercises; and for kinesthetic-oriented learners, designing interactive grammar games, handwritten practice tasks, and scenario simulation exercises. "Designing reflective tasks based on metacognitive levels" specifically means: learners with low metacognitive abilities are provided with structured error reflection templates (clearly specifying error types, knowledge gaps, etc.). Improvement steps, key points for the next self-check), fixed-cycle learning plan templates, and daily learning review tips; for learners with intermediate to high metacognitive abilities, open-ended reflection topics, tools for self-developing learning plans, and phased self-assessment scales for learning effectiveness are provided; "Prioritizing core knowledge points based on learning objectives and scenario tags" specifically means: test preparation learners (such as middle school entrance exams, IELTS preparation) should prioritize knowledge points corresponding to the "high-frequency test point" tags in the dynamic relational knowledge graph, combined with past exam questions adapted for practice and exam scenario simulation tasks; academic English learners should focus on in-depth learning of knowledge points such as complex clauses and academic voice, integrated into scenario application tasks such as literature reading and paper writing; daily communication learners should prioritize planning high-frequency oral grammar (such as tense application and simple clauses), combined with dialogue practice in daily scenarios such as shopping, travel, and socializing. By continuously iterating path adaptation strategies through reinforcement learning algorithms, and using "knowledge mastery achievement rate, metacognitive ability improvement, and scenario application proficiency" as core reward signals, the algorithm dynamically optimizes path parameters. For example, when a learner achieves the mastery rate of knowledge points learned according to the path three times consecutively, and the metacognitive reflection record meets the depth requirements, the algorithm will increase the proportion of self-planning tasks. At the same time, it links real-time data from the multimodal intelligent diagnosis module and the closed-loop optimization module to trigger a dynamic adjustment mechanism: when the mastery rate of a certain knowledge point reaches 90% or more, it automatically unlocks the comprehensive application task of related knowledge points and extends the review cycle of that module; when the error rate of the same type exceeds 40% for two consecutive times, it suspends the learning of subsequent new knowledge points and inserts cognitive bias correction courses, targeted reflection training, and basic consolidation exercises until the error rate drops below 20%; when the metacognitive ability rating improves by one level, it reduces structured guidance content and increases open learning tasks and self-planning space, ensuring that the learning path is always accurately matched with the learner's real-time status and ability improvement pace, achieving synergistic advancement of knowledge, ability, and metacognition.

[0024] In the implementation of the case, the personalized content generation module: based on the metacognitive guided learning path, it adopts generative adversarial network (GAN) combined with multimodal content generation technology to customize personalized learning content. The personalized content generation module includes knowledge point explanation materials, graded practice questions, metacognitive training tasks and scenario-based application tasks, and all content is labeled with corresponding knowledge graph nodes and metacognitive prompts. Among them, the knowledge point explanation materials are generated in the form of images, videos and other adapted formats according to cognitive styles, and interdisciplinary thematic cases are integrated; the graded practice questions are divided into basic, advanced and other levels, gradually linking knowledge points with real-world scenarios; metacognitive training tasks include tools such as error reflection templates; and scenario-based application tasks simulate real communication, examination and other scenarios. By optimizing content fit through generative adversarial networks (GANs), combining multimodal content generation technology to match learning style preferences, and annotating knowledge graph nodes and metacognitive cues, we can ensure that the content is accurately aligned with the path goals and user profiles, providing targeted resource support for personalized learning.

[0025] In the implementation of the case, the emotional real-time feedback module uses affective computing and personalized feedback algorithms to provide learners with emotional and targeted real-time feedback on their learning process and practice results. The feedback forms in the emotional real-time feedback module include: result feedback, error feedback, emotional feedback and interactive feedback. At the same time, it will push knowledge point links, metacognitive training tasks and consolidation exercises related to the error points. The results feedback uses visual charts to present the accuracy rate of answers and the progress of knowledge mastery, accompanied by encouraging language to enhance the sense of accomplishment in learning; error feedback adopts a five-step model of "error location - multi-dimensional root cause analysis - rule review - scenario-based examples - metacognitive self-check tips" to accurately break down problems; emotional feedback is tailored to the learner's emotional profile, such as using gentle and encouraging language for highly sensitive learners and direct and efficient feedback for goal-oriented learners; interactive feedback supports multiple forms of interaction such as text, voice, and handwriting, and can answer questions instantly; By analyzing learners' learning status (such as frustration or apathy) using affective computing algorithms, and combining this with personalized feedback algorithms to match feedback formats and content, the system synchronously links dynamically with knowledge graphs and metacognitive training modules. This allows for the precise delivery of links to knowledge points corresponding to errors, appropriate metacognitive training tasks, and tiered reinforcement exercises, helping learners to immediately fill in knowledge gaps and improve their learning motivation and effectiveness.

[0026] In the case implementation, the metacognitive training module includes four sub-modules: self-planning, self-monitoring, self-reflection, and self-assessment. The self-planning sub-module, based on in-depth user profiles and learning goals, guides learners to develop personalized learning plans, provides plan templates and adjustment suggestions, tracks plan execution, and provides reminders. The self-monitoring sub-module monitors in real time for distractions and progress delays during the learning process, helping learners get back into focus through pop-up prompts and suggestions for adjusting the learning pace. The self-reflection sub-module provides structured error reflection tools, knowledge point summary templates, and learning strategy optimization record forms. The self-assessment sub-module regularly generates multi-dimensional self-assessment scales, guiding learners to conduct self-assessments against their profiles and learning goals, and analyzes the gaps based on the system's assessment results. The self-planning sub-module features a plan template designed in layers according to short-term and long-term goals, with adjustment suggestions generated based on learners’ attention span, execution rate, and other profile characteristics. The self-reflection sub-module’s structured tools clearly define the dimensions for filling in errors, such as reasons for errors and knowledge gaps, and are adapted to different error types. The self-assessment scale covers core dimensions corresponding to the profile, such as knowledge mastery and depth of reflection. By linking with in-depth user profiles and multimodal intelligent diagnostic results, the training intensity is dynamically adapted—strengthening template guidance and real-time reminders for those with low metacognitive levels, and increasing self-planning space for those with intermediate and advanced levels; at the same time, the training data is synchronized to the closed-loop optimization module to update the metacognitive dimension parameters of the profile, providing a basis for adjusting the learning path and generating content.

[0027] In the implementation of the case, the social collaboration module supports precise matching of learning partners based on deep user profiles, the formation of topic learning groups, and multi-form collaborative learning, and incorporates incentive mechanisms such as points rewards, the selection of reflection stars, and collaborative task achievement badges. Among them, precise learning partner matching is based on the complementarity of learning goals, English proficiency and cognitive style. For example, test takers are matched with learners with the same goals, and scene-dependent learners are matched with scene-independent partners. Thematic learning groups are formed around interdisciplinary themes (science and technology, business, etc.) or learning scenarios (high school entrance examination preparation, daily communication, etc.), and collaborative learning forms include mutual review of wrong questions, co-construction of learning plans, and sharing of strategies. By recording learners' participation in collaborative tasks and the quality of their contributions (such as the practicality of their reflections and the value of their shared content), corresponding points and badges are awarded. These points can be redeemed for learning resources. "Reflection Stars" are selected and publicized based on the depth and frequency of their reflections. At the same time, collaborative data is synchronized to the deep user profile and closed-loop optimization module to supplement the profile with social interaction characteristics, provide a reference for optimizing learning paths and content generation, and improve learning enthusiasm and collaborative effectiveness.

[0028] In the implementation of the case, the closed-loop optimization module continuously collects the interaction data of learners in each module and iteratively optimizes the system model by combining machine learning algorithms; The collected interactive data covers multiple dimensions of information, including the frequency of clicks on learning content, the quality of practice completion (accuracy rate, answering time), the depth of reflection records (whether the reasons for errors and improvement methods are fully covered), feedback satisfaction rating, changes in metacognitive ability (plan execution rate, self-check accuracy rate), and social collaboration participation (task completion rate, value of shared content), comprehensively capturing learners' learning status and system adaptation effect. Machine learning algorithms are used to classify, analyze, and mine patterns in the collected data: For the multimodal intelligent diagnostic module, the accuracy of error identification and root cause analysis is optimized to reduce the rate of missed implicit errors; for the dynamic knowledge graph module, the strength of node associations is adjusted, and scenario cases and interdisciplinary thematic materials corresponding to high-frequency errors are added; for the personalized content generation module, adaptation parameters are updated to make the explanation format and exercise difficulty more in line with learners' cognitive styles and knowledge levels; for the learning path planning module, the reward mechanism of the reinforcement learning algorithm is optimized to make path adjustments more in line with the learners' ability improvement pace; at the same time, common patterns in group learning data (such as high-frequency weaknesses and efficient learning strategies of learners at specific levels) are mined to feed back into knowledge graph updates and module function upgrades, achieving a two-way promotion of individual adaptation optimization and overall system iteration, and continuously improving the accuracy and effectiveness of the system.

[0029] The steps for using AI-based personalized English grammar learning are as follows: S1: Multimodal data acquisition and profile initialization; After registering and logging in, learners fill in basic information and learning objectives, complete cognitive feature tests and initial multimodal diagnostic tasks, and the system integrates data to generate an initial in-depth user profile; In this process, learners fill in basic information and learning objectives through the multimodal data collection module, and complete a cognitive characteristic test and a learning preference survey; the system adapts to multiple terminals to collect learners' initial multimodal input data; the deep user profile construction module integrates the above data to generate an initial deep user profile, clarifying the learner's cognitive style, initial English grammar level, metacognitive ability baseline, and emotional preferences.

[0030] S2: Multimodal intelligent diagnosis and root cause localization; the system completes the initial calibration by dynamically associating knowledge graphs, performs multimodal analysis on the initial diagnosis task, identifies error types and root causes, and generates a diagnosis report that includes a list of weaknesses and capability ratings; Among them, the multimodal intelligent diagnosis module pushes customized multimodal diagnosis tasks. After the learner completes the task, the system analyzes the input content through a multimodal fusion deep learning model, identifies explicit and implicit errors, locates knowledge gaps, cognitive shortcomings and metacognitive deficiencies, generates a detailed diagnostic report and provides feedback to the learner.

[0031] S3: Metacognitive Guided Path Generation; The path planning module generates personalized paths based on profiles, diagnostic reports, and maps, including the order of knowledge points, learning duration, and reflection tasks, and supports learners to adjust and provide feedback for optimization. The knowledge points are arranged according to the principle of "prioritizing weak points and progressively addressing related knowledge points", prioritizing the knowledge points corresponding to the high-frequency errors marked in the diagnostic report; the learning time is allocated based on the learner's attention span, information processing speed and other cognitive characteristics, with more time extended for more difficult grammar points; the reflection tasks are designed according to the differences in metacognitive levels, with structured templates for low-level learners and open-ended topics for intermediate and high-level learners. By integrating reinforcement learning algorithms with metacognitive guidance strategies, dynamic adjustments are triggered in conjunction with real-time learning data—automatic advancement is achieved when the knowledge point mastery rate reaches over 90%, and corrective training is inserted when the error rate of the same type exceeds 40%; at the same time, learner adjustment suggestions are collected, and the path adaptability is optimized in combination with feedback to ensure that the path not only fits the profile and diagnostic results, but also meets personalized needs.

[0032] S4: Personalized learning and metacognitive training; learners acquire suitable learning content according to the path, complete graded exercises and scenario-based tasks, the system provides emotional real-time feedback, and the metacognitive training module guides self-planning, monitoring and reflection. The learning content is generated by Generative Adversarial Networks (GANs) combined with multimodal technology, and customized according to cognitive style—visual learners are given text / image / animation materials, auditory learners are given audio explanations / dialogue audio, and kinesthetic learners are given interactive games / handwriting tasks, and interdisciplinary thematic cases are incorporated; the graded exercises are divided into basic questions (consolidation of a single knowledge point), advanced questions (integration of 2-3 related knowledge points), comprehensive questions (real-world scenarios + interdisciplinary applications), and innovative questions (open-ended grammar applications); the scenario-based tasks are aligned with the learning objectives, such as exam takers focusing on adapting real exam questions, academic English learners focusing on practicing essay writing paragraphs, and daily communicators designing dialogue creation tasks around scenarios such as shopping and travel; The emotional real-time feedback module adopts a "result visualization + five-step error analysis + emotionally appropriate tone + multi-form interaction" model, simultaneously pushing error-related knowledge point links and metacognitive self-check prompts; the metacognitive training module is linked to the deep user profile, providing structured templates (error reflection, plan formulation) and real-time reminders for low-level learners, and providing open tasks and self-planning tools for intermediate and advanced learners. All training data is synchronized to the deep user profile and closed-loop optimization module to support subsequent path adjustments and content iterations.

[0033] S5: Learning Data Collection and Dynamic Optimization: The closed-loop optimization module continuously collects data such as learning behavior, practice quality, and reflection depth, updates user profiles and knowledge graphs, and adjusts learning paths and content parameters. The collected data specifically covers key dimensions of the entire learning process, including the frequency of clicks on learning content, the duration of time spent on knowledge points, the accuracy and completion time of answering practice questions, the completeness of reflection records on incorrect questions (whether they include the reasons for the errors and improvement methods), the execution rate of metacognitive training tasks, feedback satisfaction scores, participation in social collaboration tasks, and the quality of contributions. By using machine learning algorithms to deeply analyze the collected data, we can dynamically update the deep user profile—for example, adjusting the distribution of weaknesses in the knowledge dimension based on practice quality, upgrading the metacognitive ability rating based on the depth of reflection, and optimizing the emotional dimension preference tags based on feedback and interaction performance. On the other hand, we can optimize the dynamically linked knowledge graph, supplement the scenario cases and interdisciplinary materials corresponding to high-frequency errors, and adjust the node association strength. At the same time, we can iterate the core parameters of learning path planning and personalized content generation. For example, for knowledge points with a high error rate of over 40%, we can extend the learning time and increase consolidation exercises. For learners with improved metacognitive abilities, we can reduce structured guidance and increase self-planning tasks to achieve real-time adaptation between the system and the learner's state.

[0034] S6: Phase Evaluation and Closed-Loop Iteration: After each learning unit, conduct multi-dimensional evaluations, combine the evaluation results to optimize subsequent learning plans, repeat steps S4 and S5, forming a closed loop of collection-diagnosis-learning-feedback-optimization until the learning objectives are achieved; The multi-dimensional assessment covers knowledge, metacognitive, and affective dimensions: the knowledge dimension uses multimodal unit tests (including text-based quizzes, oral grammar application, and handwriting practice) and scenario-based application tasks (such as simulated academic writing and cross-cultural dialogue creation) to test the mastery and flexible application of knowledge points; the metacognitive dimension assesses self-learning ability through data such as the depth of reflection records, the accuracy of self-assessment, and the execution rate of learning plans; and the affective dimension captures learning attitude through indicators such as learning motivation change questionnaires, feedback satisfaction scores, and task persistence. By linking and analyzing the assessment results with in-depth user profiles, multimodal intelligent diagnostic results, and dynamically related knowledge graphs: if the knowledge dimension does not meet the standards, the difficulty gradient and exercise type of personalized learning content are adjusted first, and the scenario-based training of weak knowledge points is strengthened; if the metacognitive dimension improves slowly, structured reflection templates and real-time monitoring reminders are added; if the emotional dimension feedback is negative, the tone and incentive mechanism of the emotional real-time feedback are optimized; at the same time, the assessment data is synchronized to the closed-loop optimization module to iterate the core model parameters of the system, so that the learning path, content generation, and diagnostic accuracy continuously adapt to the learner's state, ensuring the dynamic optimization effect of the closed loop.

[0035] In summary, this AI-based personalized English grammar learning system and method, by setting up a multimodal data acquisition module and a deep user profile construction module, comprehensively captures learners' basic information, cognitive characteristics, metacognitive level, and emotional preferences. It utilizes multi-dimensional data fusion algorithms and time-series analysis algorithms to construct dynamically updated deep user profiles. This solves the problems of existing systems where data acquisition dimensions are singular, user profiles are superficial, and they fail to accurately depict learners' individual differences and dynamic changes, resulting in a lack of targeted core basis for subsequent diagnosis, path planning, and content generation.

[0036] Furthermore, by setting up a four-dimensional dynamically related knowledge graph module and a multimodal intelligent diagnostic module, on the one hand, the four-dimensional architecture of "grammar rules - scenario application - interdisciplinary topics - metacognitive prompts" breaks through the limitations of traditional static knowledge graphs, realizing deep binding and dynamic updates of knowledge with scenarios, disciplines and metacognition; on the other hand, a multimodal fusion deep learning model with a bidirectional LSTM+CRF architecture is adopted to accurately identify explicit and implicit grammatical errors and trace the root causes at the knowledge, cognition and metacognition levels, solving the problems of existing systems where knowledge is presented in an isolated and static manner, lacks scenario-based and interdisciplinary connections, and error diagnosis only stays on the surface and is difficult to locate deep root causes.

[0037] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0038] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A personalized English grammar learning system based on artificial intelligence, characterized by: It includes a multimodal data acquisition module, a deep user profile construction module, a dynamic association knowledge graph module, a multimodal intelligent diagnosis module, a metacognitive guided learning path planning module, a personalized content generation module, an emotional real-time feedback module, a metacognitive training module, a social collaboration module, and a closed-loop optimization module; The multimodal data acquisition module is used to collect learners' multidimensional and multimodal data, covering basic information, learning objectives, cognitive feature data, multimodal learning behavior data, and metacognitive related data; The deep user profile building module uses information from the multimodal data acquisition module and employs a multi-dimensional data fusion algorithm to build a dynamically updated deep user profile. The dynamic association knowledge graph module constructs a four-dimensional dynamic knowledge graph consisting of grammatical rules, scenario applications, interdisciplinary topics, and metacognitive cues. The multimodal intelligent diagnostic module uses a multimodal fusion deep learning model to accurately diagnose and analyze the root causes of grammatical errors in the learner's multimodal input content. The metacognitive guided learning path planning module is based on deep user profiles, dynamic association knowledge graphs, and multimodal intelligent diagnostic results. It uses reinforcement learning algorithms to integrate metacognitive guidance strategies and plans dynamic personalized learning paths. The personalized content generation module: Based on the metacognitive guided learning path, it adopts generative adversarial networks (GAN) combined with multimodal content generation technology to customize personalized learning content; The emotional real-time feedback module uses emotional computing and personalized feedback algorithms to provide emotional and targeted real-time feedback on learners' learning process and practice results; The metacognitive training module includes four sub-modules: self-planning, self-monitoring, self-reflection, and self-assessment. The social collaboration module supports precise matching of learning partners based on deep user profiles, formation of thematic learning groups, and multi-form collaborative learning, and incorporates incentive mechanisms such as points rewards, selection of reflection stars, and collaborative task achievement badges. The closed-loop optimization module continuously collects learners' interaction data in each module and iteratively optimizes the system model by combining machine learning algorithms.

2. The personalized English grammar learning system based on artificial intelligence according to claim 1, characterized in that: The basic information includes age, education stage, English proficiency level (CEFR A1-C2), and learning scenario; the learning objectives include short-term and long-term goals; the cognitive characteristic data is obtained through specialized tests, including cognitive style, information processing speed, and attention span. The multimodal learning behavior data includes the content and process data of text input, voice input, and handwriting input, which are automatically recorded by the system and collected through multi-terminal (mobile phone, tablet, computer) adaptation; the metacognitive related data includes learners' records of incorrect answers and reflections, learning plan execution, self-assessment results, frequency and depth of questions raised, which are obtained through a combination of user-initiated input and automatic system analysis.

3. The personalized English grammar learning system based on artificial intelligence according to claim 1, characterized in that: The profile includes basic dimensions (age, English proficiency, learning goals), cognitive dimensions (cognitive style, information processing characteristics, attention level), knowledge dimensions (mastery of grammar points, distribution of weaknesses, error type preferences), metacognitive dimensions (self-monitoring ability, depth of reflection, planning and execution ability), and emotional dimensions (intensity of learning motivation, resilience, and preference for feedback type). The dynamic changes of the profile are tracked through time-series analysis algorithms.

4. The personalized English grammar learning system based on artificial intelligence according to claim 1, characterized in that: The dynamically linked knowledge graph module uses major grammatical categories (tense, voice, clauses, etc.) as first-level nodes, specific grammatical rules (simple present tense, relative clauses, etc.) as second-level nodes, and core test points, common mistakes, and fixed collocations as third-level nodes. It matches diverse application scenario tags and corresponding corpus examples to each third-level node; it links to interdisciplinary themes, integrating high-frequency grammatical applications within those themes; it embeds metacognitive prompts, including common reflection directions for the knowledge point, self-checking methods for common mistakes, and learning strategy suggestions (metacognitive prompt dimension); and based on massive real-time corpora (academic literature, social media, and past exam questions), it dynamically updates the association strength of knowledge points, adds emerging application scenarios, and identifies high-frequency errors through machine learning algorithms.

5. The personalized English grammar learning system based on artificial intelligence according to claim 1, characterized in that: The multimodal fusion deep learning model is based on a bidirectional LSTM+CRF architecture that integrates visual feature extraction (for handwriting input, including writing conventions and symbol usage) and speech feature extraction (for speech input, including tense, intonation, and grammatical pauses). It achieves dual recognition of explicit and implicit grammatical errors and subdivides error types into five major categories and 32 subcategories: grammatical rules, fixed collocations, context adaptation, cognitive biases, and expression norms. Combining deep user profiles and dynamically associated knowledge graphs, it locates the knowledge gaps, cognitive shortcomings, and metacognitive deficiencies corresponding to errors and generates an accurate diagnostic report that includes error location, error type, root cause analysis, and links to related knowledge points.

6. The personalized English grammar learning system based on artificial intelligence according to claim 1, characterized in that: The dynamic personalized learning path aims at three goals: knowledge acquisition, ability enhancement, and metacognitive development, and combines the following principles: adapting learning formats according to cognitive styles; Design reflection tasks based on metacognitive levels; The core knowledge points are prioritized based on learning objectives and scenario tags; the dynamic personalized learning path includes the order of knowledge point learning, allocation of learning time, combination of practice types, arrangement of reflection tasks and setting of review cycles, and has a dynamic adjustment mechanism.

7. The personalized English grammar learning system based on artificial intelligence according to claim 1, characterized in that: The personalized content generation module includes knowledge point explanation materials, graded practice questions, metacognitive training tasks, and scenario-based application tasks, and all content is labeled with corresponding knowledge graph nodes and metacognitive prompts.

8. The personalized English grammar learning system based on artificial intelligence according to claim 1, characterized in that: The feedback formats in the emotional real-time feedback module include: result feedback, error feedback, emotional feedback, and interactive feedback. It also pushes knowledge point links, metacognitive training tasks, and reinforcement exercises related to the error points.

9. The personalized English grammar learning system based on artificial intelligence according to claim 1, characterized in that: The self-planning submodule, based on in-depth user profiles and learning goals, guides learners to develop personalized learning plans, provides plan templates and adjustment suggestions, tracks plan execution and provides reminders; The self-monitoring submodule monitors in real time situations such as distraction and slow progress during the learning process, and helps learners get back into the zone through pop-up prompts and suggestions for adjusting the learning pace; the self-reflection submodule provides a structured tool for reflecting on wrong answers, a template for summarizing knowledge points, and a form for recording learning strategy optimization; the self-assessment submodule regularly generates a multi-dimensional self-assessment scale to guide learners to conduct self-assessment by comparing their profile with their learning goals, and analyze the gaps by combining the system's assessment results.

10. A personalized English grammar learning method based on artificial intelligence, comprising the personalized English grammar learning system based on artificial intelligence as described in claims 1-9, characterized in that: The operation steps are as follows: S1: Multimodal data acquisition and profile initialization; After registering and logging in, learners fill in basic information and learning objectives, complete cognitive feature tests and initial multimodal diagnostic tasks, and the system integrates data to generate an initial in-depth user profile; S2: Multimodal intelligent diagnosis and root cause localization; the system completes the initial calibration by dynamically associating knowledge graphs, performs multimodal analysis on the initial diagnosis task, identifies error types and root causes, and generates a diagnosis report that includes a list of weaknesses and capability ratings; S3: Metacognitive Guided Path Generation; The path planning module generates personalized paths based on profiles, diagnostic reports, and maps, including the order of knowledge points, learning duration, and reflection tasks, and supports learners to adjust and provide feedback for optimization. S4: Personalized learning and metacognitive training; learners acquire suitable learning content according to the path, complete graded exercises and scenario-based tasks, the system provides emotional real-time feedback, and the metacognitive training module guides self-planning, monitoring and reflection. S5: Learning Data Collection and Dynamic Optimization: The closed-loop optimization module continuously collects data such as learning behavior, practice quality, and reflection depth, updates user profiles and knowledge graphs, and adjusts learning paths and content parameters. S6: Phase Evaluation and Closed-Loop Iteration: After each learning unit, conduct multi-dimensional evaluations, combine the evaluation results to optimize subsequent learning plans, repeat steps S4 and S5, forming a closed loop of collection-diagnosis-learning-feedback-optimization until the learning objectives are achieved.

Images