A cognitive diagnosis and reverse thinking training education system based on dual AI collaboration
The cognitive diagnosis and reverse thinking training education system based on dual AI collaboration solves the problems of personalized learning and dynamic teaching loop in basic education, realizes accurate diagnosis of students' cognitive status and personalized training, and improves teaching effectiveness.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANGHAI FANGLUEMENKOU EDUCATION TECH CO LTD
- Filing Date
- 2026-01-26
- Publication Date
- 2026-05-26
AI Technical Summary
Existing educational technologies struggle to accurately capture differences in students' cognitive levels during basic education, lack multi-dimensional cognitive diagnostic tools, fail to provide personalized learning support, and lack reverse thinking training and dynamic teaching loops, resulting in poor teaching outcomes.
A cognitive diagnosis and reverse thinking training education system based on dual AI collaboration is adopted. Through multimodal learning data preprocessing, multimodal cognitive diagnosis model and dynamic division of cognitive level, a personalized training task matrix is generated, the vulnerability node is located and a progressive question sequence is generated, and a dynamic mastery evaluation index system is constructed.
It enables accurate diagnosis of students' cognitive status and personalized training, enhances students' learning initiative and knowledge transfer ability, dynamically adjusts learning paths, and improves the accuracy and efficiency of teaching.
Smart Images
Figure CN121581820B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the fields of educational technology and artificial intelligence technology, specifically relating to an educational system for cognitive diagnosis and reverse thinking training based on dual AI collaboration. Background Technology
[0002] In the development of AI-enabled education, while existing educational technologies and traditional classrooms have attempted to integrate intelligent elements to some extent, significant shortcomings remain in supporting precise teaching and personalized learning. This is particularly evident in basic education (such as K-12) classroom settings, where the intelligent requirements of a "teaching-driven learning" model are difficult to meet. The specific core issues are as follows:
[0003] The challenge of adapting to individual cognitive differences: In large-class teaching scenarios, teachers are limited by their energy and teaching resources, making it difficult to accurately capture the differences in each student's cognitive level and learning style. Existing intelligent teaching systems mostly rely on a unified knowledge push logic and lack a dynamic response mechanism for individual student characteristics, making it difficult to break through the "one-size-fits-all" teaching model and failing to provide suitable learning support for students with different cognitive levels.
[0004] The bottleneck of making implicit thinking explicit: Implicit thinking problems in students during the learning process, such as logical leaps, conceptual confusion, and knowledge transfer gaps, are difficult to effectively observe through traditional methods such as text-based answers and after-class tests. Existing technologies lack cognitive diagnostic tools that can integrate multi-dimensional information, and cannot transform unstructured learning process data (such as flaws in spoken expression, logical deviations in handwritten annotations, and defects in flowchart construction) into structured cognitive problem analysis results. This results in a lack of accurate data basis for teaching and tutoring, making it difficult to pinpoint the root causes of learning weaknesses.
[0005] The lack of deep thinking training models: Most intelligent teaching systems remain at the level of one-way knowledge transmission, with students passively receiving explanations of knowledge points and exercises, lacking training scenarios for active thinking and creative reconstruction. Existing technologies have not built an interactive framework that can guide students to engage in reverse thinking (such as role reversal and logical inversion), failing to effectively cultivate students' knowledge transfer ability, metacognitive ability, and problem-solving ability, and making it difficult to achieve the learning goal shift from "knowledge memorization" to "thinking construction."
[0006] The dynamic teaching closed loop fails to function: the existing assessment system relies mainly on static post-class tests, with assessment results lagging behind the learning process and unable to reflect students' cognitive changes in real time. Simultaneously, the lack of a technological engine to automatically generate personalized learning paths based on real-time assessment results prevents dynamic adjustments to the difficulty of learning tasks, the order of content modules, and training methods based on students' cognitive diagnostic data. This leads to a break in the "learning-assessment-optimization" teaching closed loop, hindering dynamic iteration and continuous optimization of the teaching process and failing to provide teachers with real-time and accurate data to support their teaching decisions. Summary of the Invention
[0007] To address the aforementioned problems in existing technologies, this invention provides a cognitive diagnosis and reverse thinking training education system based on dual AI collaboration.
[0008] The objective of this invention can be achieved through the following technical solutions:
[0009] Includes: cognitive hierarchical matching unit, error and omission targeted correction unit, pre-reinforcement unit, and dynamic capability iteration unit;
[0010] The cognitive hierarchical matching unit acquires and preprocesses multimodal learning data during the student's learning process; inputs the multimodal learning data into a pre-trained multimodal cognitive diagnostic model, and outputs a structured multidimensional cognitive feature vector; based on the multidimensional cognitive feature vector, dynamically divides students into different cognitive levels and constructs differentiated cognitive profiles; and generates a personalized training task matrix according to a preset cognitive level and task mapping rule base.
[0011] The error-targeting correction unit, based on the personalized training task matrix, uses a rule engine and pattern matching algorithm to locate vulnerability nodes and typical logical error paradigms; according to the identified error paradigms, it calls and generates a progressive question sequence from a preset metacognitive question chain library.
[0012] Based on the cognitive profile, scenario-based verification tasks are generated, forming a set of error correction and knowledge reinforcement.
[0013] The pre-reinforcement unit constructs a group cognitive graph based on the error correction and knowledge reinforcement set; based on the group cognitive graph, it dynamically reconstructs the logical order of the teaching content, puts the common cognition and related ability training modules in advance, and generates a teaching training sequence; based on the teaching training sequence, it builds a centralized special training module to gradually promote the extended learning of related knowledge.
[0014] The dynamic ability iteration unit constructs a dynamic mastery assessment index system; it quantitatively assesses students' knowledge mastery status based on the dynamic mastery assessment index system; it switches training modes according to the assessment results, and finally generates a multi-dimensional ability feedback report.
[0015] As a preferred technical solution of the present invention, the specific process of the cognitive hierarchical matching unit is as follows: First, multimodal learning data acquisition and preprocessing are carried out, collecting data such as students' speech, text answers, handwritten annotations, and flowchart drawing during learning; the speech data is denoised and transcribed, and semantic and rhythmic features are extracted; redundant information is removed from the text data, core viewpoints are extracted and knowledge point associations are marked; the handwritten and flowchart data are image-enhanced, key markers and logical connections are identified, and a standardized dataset is formed by unifying the format; the standardized data is input into a pre-trained multimodal cognitive diagnostic model, and the model integrates the core information of each modality through feature extraction and fusion, and solves the problem with reference to the preset cognitive dimension framework. The system analyzes knowledge mastery and logical thinking completeness to output structured multidimensional cognitive feature vectors. Based on these feature vectors, it extracts core evaluation dimensions such as knowledge reserves, logical reasoning ability, and knowledge transfer ability, sets judgment criteria for each dimension, and dynamically classifies cognitive levels by comparing feature vectors with standard thresholds. Simultaneously, it tracks changes in subsequent learning data to adjust level assignments in real time. Key information such as knowledge structure weaknesses and thinking style tendencies are selected from the feature vectors and transformed into concrete labels to construct differentiated cognitive profiles. Finally, based on the cognitive level, the system retrieves corresponding task templates from the mapping rule base, adjusts task details by combining weaknesses in the profile with learning style labels, supplements targeted reinforcement content, and generates a personalized training task matrix.
[0016] Specifically, the process of acquiring and preprocessing multimodal learning data during the student learning process is as follows: noise reduction and text transcription are performed on the speech data to extract semantic information and rhythmic features from the speech; redundant information is removed from the text data and the relationships between knowledge points are marked; image enhancement processing is performed on handwritten annotations and flowchart data to identify key markers and logical connections; and a standardized dataset that can be directly used as model input is formed after unifying the data format.
[0017] Specifically, the pre-trained multimodal cognitive diagnostic model is used to output structured cognitive features. The specific process is as follows: feature extraction is performed on the standardized data of each modality; the core information of different modalities is integrated through feature fusion technology; according to the preset cognitive feature dimension framework, the knowledge mastery status, the integrity of thinking logic, and the coherence of expression are structured and analyzed, and finally a multidimensional cognitive feature vector reflecting the student's cognitive status is output.
[0018] Specifically, the process of dynamically classifying students into different cognitive levels is as follows: based on multidimensional cognitive feature vectors, core assessment dimensions are extracted, including knowledge reserves, logical reasoning ability, and knowledge transfer ability; judgment criteria for each dimension are set; by comparing feature vectors with preset standard thresholds, students are classified into corresponding cognitive levels; at the same time, changes in student learning data are tracked, and level assignments are adjusted in real time.
[0019] Specifically, the differentiated cognitive profile is used to visualize students' cognitive characteristics. The specific process is as follows: extract key information from the multidimensional cognitive feature vector, including weaknesses in knowledge structure, thinking tendencies, and learning style preferences; match the key information with a preset feature tag library and calculate the matching degree of each tag; select the matched tags to form a basic profile framework.
[0020] Specifically, the cognitive level and task mapping rule base is used to match training tasks. The specific process is as follows: the cognitive level and task mapping rule base presets the task difficulty, knowledge point coverage, and training focus direction corresponding to each cognitive level; retrieves the corresponding basic task template according to the student's cognitive level; makes detailed adjustments to the task template by combining the weaknesses and learning style tags in the differentiated cognitive profile; supplements targeted reinforcement content, and combines them to form a personalized training task matrix.
[0021] Specifically, the process of locating vulnerability nodes and typical logical error paradigms through a rule engine and pattern matching algorithm is as follows: the rule engine loads preset knowledge vulnerability judgment rules and logical error identification standards; the pattern matching algorithm compares the students' answer data of completing training tasks with a preset error paradigm library, captures feature differences in the data, and determines the specific location of the vulnerability node and the error paradigm type to which it belongs.
[0022] Specifically, the metacognitive probing question chain library is used to generate guiding questions. The specific process is as follows: based on the identified error paradigm type, the corresponding basic probing template in the metacognitive probing question chain library is retrieved; the question expression is adjusted according to the preset progressive logic; and a guiding angle adapted to the student's cognitive profile is incorporated to form a progressive question sequence that promotes students' independent reflection.
[0023] Specifically, the process of generating scenario-based verification tasks is as follows: based on the learning style and life-related information in the cognitive profile, select familiar life scenarios, campus scenarios, or practical scenarios for students; design a verification task framework that includes problem situations, task objectives, operation steps, and verification standards; integrate prompts and knowledge application guidance into the task; and examine the students' error correction effects and practical application ability through task completion to form an error correction and knowledge reinforcement set that includes task data repair effect data.
[0024] Specifically, the process of constructing the group cognitive graph is as follows: summarize the error correction and knowledge reinforcement sets of all students; extract common vulnerability nodes and error paradigms; use knowledge points as graph nodes and error relationships as edges between nodes, and set the node size according to the frequency of error occurrence; set the edge weights according to the correlation strength between errors, and construct a hierarchical group cognitive graph in combination with the hierarchical relationship of knowledge points.
[0025] Specifically, the process of constructing the dynamic mastery assessment index system is as follows: setting multi-level evaluation standards and quantitative score ranges for each dimension, determining the weight coefficients of each dimension through the analytic hierarchy process; quantitatively scoring each dimension by combining students' answer data, behavioral data, and result data in training tasks; and comprehensively judging students' knowledge mastery status based on the scoring results.
[0026] Specifically, the process of generating a multi-dimensional capability feedback report is as follows: classifying and organizing data according to cognitive levels; presenting the performance of each dimension using visual charts; analyzing strengths and weaknesses in conjunction with textual descriptions, proposing targeted training and adjustment suggestions, and forming a structured feedback report that includes data visualization analysis conclusions and suggestions.
[0027] The beneficial effects of this invention are as follows:
[0028] (1) By setting up a multimodal learning data preprocessing mechanism, a pre-trained multimodal cognitive diagnostic model, and a dynamic division logic of cognitive levels, the system can first perform noise reduction and text transcription on students' speech data to extract semantic information and rhythmic features, remove redundant information and extract core viewpoints from text data to mark the relationships between knowledge points, and perform image enhancement processing on handwritten annotations and flowchart data to identify key markers and logical connections. After unifying the format, a standardized dataset is formed. Then, the model extracts and fuses features from each modality of data, and analyzes the knowledge mastery status and the integrity of thinking logic with reference to the cognitive feature dimension framework. The system outputs multi-dimensional cognitive feature vectors, including knowledge reserves and logical reasoning ability. Based on these vectors, core assessment dimensions are extracted, and cognitive levels are categorized according to standard thresholds. Data changes are tracked and adjusted in real-time. Simultaneously, information such as knowledge weaknesses and thinking tendencies is filtered and transformed into concrete labels to construct differentiated cognitive profiles. Finally, task templates are retrieved from the rule base based on the cognitive level, and targeted content is added to the profiles to generate personalized training task matrices. This avoids the limitations of traditional "one-size-fits-all" teaching, allowing each student to receive training content precisely matched to their cognitive level, reducing ineffective learning input, and enhancing students' acceptance and initiative in learning tasks.
[0029] (2) By setting up a rule engine, pattern matching algorithm, metacognitive questioning chain library and scenario-based verification task generation mechanism, the rule engine can load the preset knowledge vulnerability judgment rules and logical error identification standards. The pattern matching algorithm compares the student's answer data with the error paradigm library to capture feature differences and determine the specific location of the vulnerability node and the type of error paradigm to which it belongs. Then, based on the error paradigm, the basic template is retrieved from the questioning library. The expression is adjusted according to the progressive logic of "phenomenon identification - cause analysis - logical reconstruction - knowledge consolidation". A progressive question sequence is generated by integrating the guidance angle that matches the cognitive profile, guiding students to sort out the root cause of the error. Finally, referring to the learning style and life-related information in the profile, familiar scenarios such as campus and life are selected, and the knowledge points corresponding to the vulnerability are integrated into them. Scenario-based tasks containing problem context, task objectives, operation steps and verification standards are designed to test the error correction effect and knowledge application ability. It can accurately locate cognitive vulnerabilities and guide students to actively construct correct logic through guided reflection, rather than passively accepting answers, thus deepening their understanding of knowledge points. At the same time, scenario-based application can strengthen the connection between knowledge and real life and improve students' knowledge transfer ability and logical reconstruction ability. Attached Figure Description
[0030] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.
[0031] Figure 1 This is a system architecture diagram of a cognitive diagnosis and reverse thinking training education system based on dual AI collaboration according to the present invention;
[0032] Figure 2 This is a data flow diagram of a cognitive diagnosis and reverse thinking training education system based on dual AI collaboration according to the present invention. Detailed Implementation
[0033] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.
[0034] Please see Figure 1-2 A cognitive diagnosis and reverse thinking training education system based on dual AI collaboration;
[0035] Includes: cognitive hierarchical matching unit, error and omission targeted correction unit, pre-reinforcement unit, and dynamic capability iteration unit;
[0036] The cognitive hierarchical matching unit acquires and preprocesses multimodal learning data during the student's learning process; inputs the multimodal learning data into a pre-trained multimodal cognitive diagnostic model, and outputs a structured multidimensional cognitive feature vector; based on the multidimensional cognitive feature vector, dynamically divides students into different cognitive levels and constructs differentiated cognitive profiles; and generates a personalized training task matrix according to a preset cognitive level and task mapping rule base.
[0037] The error-targeting correction unit, based on the personalized training task matrix, uses a rule engine and pattern matching algorithm to locate vulnerability nodes and typical logical error paradigms; according to the identified error paradigms, it calls and generates a progressive question sequence from a preset metacognitive question chain library.
[0038] Based on the cognitive profile, scenario-based verification tasks are generated, forming a set of error correction and knowledge reinforcement.
[0039] The pre-reinforcement unit constructs a group cognitive graph based on the error correction and knowledge reinforcement set; based on the group cognitive graph, it dynamically reconstructs the logical order of the teaching content, puts the common cognition and related ability training modules in advance, and generates a teaching training sequence; based on the teaching training sequence, it builds a centralized special training module to gradually promote the extended learning of related knowledge.
[0040] The dynamic ability iteration unit constructs a dynamic mastery assessment index system; it quantitatively assesses students' knowledge mastery status based on the dynamic mastery assessment index system; it switches training modes according to the assessment results, and finally generates a multi-dimensional ability feedback report.
[0041] As a preferred technical solution of the present invention, the specific process of the cognitive hierarchical matching unit is as follows: First, multimodal learning data acquisition and preprocessing are carried out, collecting data such as students' speech, text answers, handwritten annotations, and flowchart drawing during learning; the speech data is denoised and transcribed, and semantic and rhythmic features are extracted; redundant information is removed from the text data, core viewpoints are extracted and knowledge point associations are marked; the handwritten and flowchart data are image-enhanced, key markers and logical connections are identified, and a standardized dataset is formed by unifying the format; the standardized data is input into a pre-trained multimodal cognitive diagnostic model, and the model integrates the core information of each modality through feature extraction and fusion, and solves the problem with reference to the preset cognitive dimension framework. The system analyzes knowledge mastery and logical thinking completeness to output structured multidimensional cognitive feature vectors. Based on these feature vectors, it extracts core evaluation dimensions such as knowledge reserves, logical reasoning ability, and knowledge transfer ability, sets judgment criteria for each dimension, and dynamically classifies cognitive levels by comparing feature vectors with standard thresholds. Simultaneously, it tracks changes in subsequent learning data to adjust level assignments in real time. Key information such as knowledge structure weaknesses and thinking style tendencies are selected from the feature vectors and transformed into concrete labels to construct differentiated cognitive profiles. Finally, based on the cognitive level, the system retrieves corresponding task templates from the mapping rule base, adjusts task details by combining weaknesses in the profile with learning style labels, supplements targeted reinforcement content, and generates a personalized training task matrix.
[0042] Specifically, the process of acquiring and preprocessing multimodal learning data during the student learning process is as follows: noise reduction and text transcription are performed on the speech data to extract semantic information and rhythmic features from the speech; redundant information is removed from the text data and the relationships between knowledge points are marked; image enhancement processing is performed on handwritten annotations and flowchart data to identify key markers and logical connections; and a standardized dataset that can be directly used as model input is formed after unifying the data format.
[0043] In this embodiment, the multimodal learning data includes various types of data generated during the student's learning process: First, audio data, covering recordings and real-time audio streams of classroom speeches, Q&A interactions, and oral explanations of problem-solving approaches; second, text data, including homework answers, test answers, electronic versions of study notes, and online discussion posts; third, handwritten data, involving scanned copies of paper notes, handwritten annotations, handwritten calculation process drafts, and draft mind maps of knowledge points; fourth, image data, including flowcharts, concept relationship diagrams, and experimental step diagrams drawn by students; and fifth, behavioral derivative data, such as learning duration distribution and task completion time records, which can assist in cognitive analysis.
[0044] In this embodiment, the "vulnerability node" specifically refers to a specific knowledge point or ability segment in a student's knowledge system that has deficiencies, including: knowledge point omission nodes (such as not mastering the derivation logic of a certain formula or a certain type of grammatical rule); concept confusion nodes (such as vague definition of similar concepts, such as confusion between the concepts of "factor" and "multiple"); weak application nodes (such as being able to recite theorems but not being able to apply them in the context of a problem); and logical connection nodes (such as the lack of reasoning basis for a certain step in a geometric proof).
[0045] In this embodiment, the logical error paradigm refers to typical logical deviations that repeatedly occur in students' thinking process, specifically including: conceptual confusion paradigm (confusing related but different knowledge points); logical discontinuity paradigm (missing key links in the reasoning chain, resulting in a lack of support for the conclusion); causal inversion paradigm (reversing the causal relationship between things to deduce); generalization paradigm (using a specific case to deduce a general conclusion); circular reasoning paradigm (using the conclusion itself as evidence to prove the conclusion); and contradictory reasoning paradigm (the logic in the reasoning process is self-contradictory).
[0046] In this embodiment, the knowledge mastery status specifically refers to the student's depth of understanding, accuracy of memory, and proficiency in application of knowledge points, encompassing the ability to accurately restate concept definitions, correctly derive formulas and theorems, flexibly apply knowledge points to different question types, and distinguish easily confused knowledge points; the logical integrity of thinking refers to the completeness of the logical chain when students think about problems, including whether key elements are covered when analyzing problems, whether the reasoning process follows causal and progressive relationships, and whether there are any breaks or jumps in the argumentation process; the coherence of expression refers to the logical connection when students express their views and problem-solving ideas through language or text, including whether there are reasonable logical connectors between sentences, whether paragraphs revolve around the core theme, whether the transition between different viewpoints or steps is natural, and whether the overall expression forms a clear logical structure.
[0047] In this embodiment, the knowledge vulnerability determination rules specifically include:
[0048] Knowledge Point Integrity Rule: For any learning content, if a student omits the core components, basic steps, or key details when explaining or applying the content, it is judged as a knowledge point integrity deficiency; Concept Definition Rule: If a student's description of the connotation and extension of a concept is inaccurate, or if the boundary between the concept and similar / easily confused concepts is unclear, it is judged as a concept comprehension deficiency; Knowledge Association Rule: If a student cannot establish a logical or application connection between the knowledge point and the preceding basic knowledge points or the subsequent extended knowledge points, it is judged as a knowledge connection deficiency; Scenario Adaptation Rule: If a student incorrectly applies a knowledge point to an inappropriate scenario, or omits the core application conditions of the knowledge point in the corresponding application scenario, it is judged as a knowledge application deficiency;
[0049] The specific criteria for identifying logical errors include:
[0050] The following criteria are used to determine logical breaks: 1. **Broken Reasoning Chain:** Students skip the necessary derivation steps between the conclusion and premises, jumping directly from premises to conclusion; 2. **Reversed Causality:** Students reverse the causal relationship between things, treating the result as the cause and the cause as the result; 3. **Overgeneralization:** Students derive a general conclusion covering the whole based on only a few cases or partial data; 4. **Conceptual Confusion:** Students mix the connotations of different concepts in their reasoning, using the confused concept as evidence to support their conclusion; 5. **Contradictory Reasoning:** Students' reasoning contains contradictions between their statements, evidence, and conclusion; 6. **Self-Contradictory Reasoning:** Students' reasoning contains contradictions between their statements, evidence, and conclusion.
[0051] Specifically, the pre-trained multimodal cognitive diagnostic model is used to output structured cognitive features. The specific process is as follows: feature extraction is performed on the standardized data of each modality; the core information of different modalities is integrated through feature fusion technology; according to the preset cognitive feature dimension framework, the knowledge mastery status, the integrity of thinking logic, and the coherence of expression are structured and analyzed, and finally a multidimensional cognitive feature vector reflecting the student's cognitive status is output.
[0052] Specifically, the process of dynamically classifying students into different cognitive levels is as follows: based on multidimensional cognitive feature vectors, core assessment dimensions are extracted, including knowledge reserves, logical reasoning ability, and knowledge transfer ability; judgment criteria for each dimension are set; by comparing feature vectors with preset standard thresholds, students are classified into corresponding cognitive levels; at the same time, changes in student learning data are tracked, and level assignments are adjusted in real time.
[0053] Specifically, the differentiated cognitive profile is used to visualize students' cognitive characteristics. The specific process is as follows: extract key information from the multidimensional cognitive feature vector, including weaknesses in knowledge structure, thinking tendencies, and learning style preferences; match the key information with a preset feature tag library and calculate the matching degree of each tag; select the matched tags to form a basic profile framework.
[0054] Specifically, the cognitive level and task mapping rule base is used to match training tasks. The specific process is as follows: the cognitive level and task mapping rule base presets the task difficulty, knowledge point coverage, and training focus direction corresponding to each cognitive level; retrieves the corresponding basic task template according to the student's cognitive level; makes detailed adjustments to the task template by combining the weaknesses and learning style tags in the differentiated cognitive profile; supplements targeted reinforcement content, and combines them to form a personalized training task matrix.
[0055] Specifically, the process of locating vulnerability nodes and typical logical error paradigms through a rule engine and pattern matching algorithm is as follows: the rule engine loads preset knowledge vulnerability judgment rules and logical error identification standards; the pattern matching algorithm compares the students' answer data of completing training tasks with a preset error paradigm library, captures feature differences in the data, and determines the specific location of the vulnerability node and the error paradigm type to which it belongs.
[0056] Specifically, the metacognitive probing question chain library is used to generate guiding questions. The specific process is as follows: based on the identified error paradigm type, the corresponding basic probing template in the metacognitive probing question chain library is retrieved; the question expression is adjusted according to the preset progressive logic; and a guiding angle adapted to the student's cognitive profile is incorporated to form a progressive question sequence that promotes students' independent reflection.
[0057] Specifically, the process of generating scenario-based verification tasks is as follows: based on the learning style and life-related information in the cognitive profile, select familiar life scenarios, campus scenarios, or practical scenarios for students; design a verification task framework that includes problem situations, task objectives, operation steps, and verification standards; integrate prompts and knowledge application guidance into the task; and examine the students' error correction effects and practical application ability through task completion to form an error correction and knowledge reinforcement set that includes task data repair effect data.
[0058] Specifically, the process of constructing the group cognitive graph is as follows: summarize the error correction and knowledge reinforcement sets of all students; extract common vulnerability nodes and error paradigms; use knowledge points as graph nodes and error relationships as edges between nodes, and set the node size according to the frequency of error occurrence; set the edge weights according to the correlation strength between errors, and construct a hierarchical group cognitive graph in combination with the hierarchical relationship of knowledge points.
[0059] Specifically, the process of constructing the dynamic mastery assessment index system is as follows: setting multi-level evaluation standards and quantitative score ranges for each dimension, determining the weight coefficients of each dimension through the analytic hierarchy process; quantitatively scoring each dimension by combining students' answer data, behavioral data, and result data in training tasks; and comprehensively judging students' knowledge mastery status based on the scoring results.
[0060] Specifically, the process of generating a multi-dimensional capability feedback report is as follows: classifying and organizing data according to cognitive levels; presenting the performance of each dimension using visual charts; analyzing strengths and weaknesses in conjunction with textual descriptions, proposing targeted training and adjustment suggestions, and forming a structured feedback report that includes data visualization analysis conclusions and suggestions.
[0061] In this embodiment, the experimental reverse thinking training of the primary school Chinese text "The Bee" is used as the application scenario. Relying on the dual AI collaborative cognitive diagnosis and reverse thinking training education system of the present invention (where "Xiao Fang" is an AI virtual character simulating real student interaction behavior, and "Xiao Fang Teacher" is an AI virtual character simulating teaching assistance behavior), the following implementation process is carried out:
[0062] First, knowledge input and multimodal data acquisition preprocessing are performed. Students input relevant knowledge about the "Bee" experiment by voice and simultaneously annotate key content such as "experimental steps" and "core objectives" with handwritten annotations. The system performs noise reduction and transcription of the voice input and extracts semantic information. It also performs image enhancement and key mark recognition on the handwritten annotations and forms a standardized dataset containing the key points of the experiment after unifying the format.
[0063] In the basic explanation section, the student verbally described the experimental steps of "Fabre catching 20 bees and putting them into a paper bag". "Student Xiao Fang" triggered a clarification of terminology: "Is the core purpose of putting them into the paper bag to prevent the bees from becoming familiar with the environment in advance?"; The system called the pre-trained multimodal cognitive diagnostic model to analyze the clarity dimension of the expression in this section. After comparing it with the preset logical judgment rules, it determined that there was no logical problem with this statement and simultaneously recorded the student's expression coherence feature data.
[0064] Moving into the in-depth simulation phase, a student explained the operation of "releasing bees four kilometers away." "Student Xiao Fang" questioned, "Is it feasible to conduct the experiment in my own backyard?" The student replied, "The backyard is too close and it's easy to rely on memory." The system used a pattern matching algorithm to compare the answer with the criteria for judging the logical integrity of the experiment, capturing the characteristic of "meeting the logical integrity standard." However, it also marked the flaw node—"failed to mention the connection between long distance and instinctive verification," clarifying that the student lacked understanding of the core logical connection of the experiment.
[0065] In the application phase, students used the analogy of "deliverymen remembering routes" to bees finding their way. "Xiao Fang" initiated an analogy test: "Deliverymen rely on maps, but what do bees rely on? You've omitted an instinctive factor in your analogy." The system evaluated the students based on the knowledge transfer ability dimension in the multidimensional cognitive feature vector, and diagnosed the conclusion that "knowledge transfer ability needs to be strengthened," while simultaneously recording the logical deviation characteristics in the analogy.
[0066] In the summary and reconstruction phase, students draw experimental framework diagrams. "Teacher Xiao Fang" uses the system to generate a comparison chart of "standard experimental framework VS student framework", marking the missing nodes of "experimental reproducibility" in the student framework. The system integrates the framework diagram data and comparison annotation information of this phase and adds them to the student's cognitive feature vector to improve their differentiated cognitive profile.
[0067] In the diagnosis and dynamic adaptation phase, the system summarizes the cognitive data from each phase and categorizes student problems into error patterns such as "incomplete knowledge transfer" and "missing key logical connections in experiments" through a rule engine. "Teacher Xiao Fang" combines the student's cognitive profile and error patterns to assist the system in retrieving adaptation content from the preset task mapping rule library and generating personalized training tasks that include "experimental core logical connection review questions," "instinctive factor analogy reinforcement exercises," and "experimental repeatability supplementary design tasks," thereby promoting the student's repair of cognitive gaps in experiments and strengthening of reverse thinking abilities.
[0068] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.
Claims
1. A cognitive diagnosis and reverse thinking training education system based on dual AI collaboration, characterized in that, include: Cognitive hierarchical matching unit, error and omission targeted correction unit, pre-reinforcement unit, dynamic capability iteration unit; The cognitive hierarchical matching unit acquires and preprocesses multimodal learning data during the student's learning process; inputs the multimodal learning data into a pre-trained multimodal cognitive diagnostic model, and outputs a structured multidimensional cognitive feature vector; based on the multidimensional cognitive feature vector, dynamically divides students into different cognitive levels and constructs differentiated cognitive profiles; and generates a personalized training task matrix according to a preset cognitive level and task mapping rule base. The error-targeting correction unit, based on the personalized training task matrix, uses a rule engine and pattern matching algorithm to locate vulnerability nodes and typical logical error patterns. Based on the identified error paradigms, a progressive question sequence is generated by calling from a pre-set metacognitive question chain library; Based on the cognitive profile, a scenario-based verification task is generated, forming a set of error correction and knowledge reinforcement; The specific process of locating vulnerability nodes and typical logical error paradigms through a rule engine and pattern matching algorithm is as follows: the rule engine loads preset knowledge vulnerability judgment rules and logical error identification standards; the pattern matching algorithm compares the students' answer data of completing training tasks with a preset error paradigm library, captures feature differences in the data, and determines the specific location of the vulnerability node and the error paradigm type to which it belongs. The pre-reinforcement unit constructs a group cognitive graph based on the error correction and knowledge reinforcement set; based on the group cognitive graph, it dynamically reconstructs the logical order of the teaching content, puts the common cognition and related ability training modules in advance, and generates a teaching training sequence; based on the teaching training sequence, it builds a centralized special training module to gradually promote the extended learning of related knowledge. The specific process of constructing the group cognitive graph is as follows: summarizing the error correction and knowledge reinforcement sets of all students; extracting common vulnerability nodes and error patterns; Using knowledge points as nodes in the graph and error relationships as edges between nodes, the node size is set according to the frequency of error occurrence; the edge weight is set according to the strength of the relationship between errors, and a hierarchical group cognitive graph is constructed by combining the hierarchical relationship of knowledge points. The dynamic ability iteration unit constructs a dynamic mastery assessment index system; it quantitatively assesses students' knowledge mastery status based on the dynamic mastery assessment index system; it switches training modes according to the assessment results, and finally generates a multi-dimensional ability feedback report.
2. The system according to claim 1, characterized in that, The specific process of acquiring and preprocessing multimodal learning data in the student learning process is as follows: noise reduction and text transcription of speech data, extraction of semantic information and rhythmic features from speech; Redundant information is removed from text data, and the relationships between knowledge points are marked; handwritten annotations and flowchart data are processed to improve image clarity and identify key markers and logical connections; after unifying the data format, a standardized dataset that can be directly used as model input is formed.
3. The system according to claim 1, characterized in that, The specific process of the pre-trained multimodal cognitive diagnostic model is as follows: feature extraction is performed on the standardized data of each modality; the core information of different modalities is integrated through feature fusion technology; according to the preset cognitive feature dimension framework, the knowledge mastery status, the integrity of thinking logic, and the coherence of expression are structured and analyzed, and finally a multidimensional cognitive feature vector reflecting the student's cognitive status is output.
4. The system according to claim 1, characterized in that, The specific process of dynamically classifying students into different cognitive levels is as follows: Based on multi-dimensional cognitive feature vectors, core assessment dimensions are extracted, including knowledge reserves, logical reasoning ability, and knowledge transfer ability; judgment criteria for each dimension are set; by comparing feature vectors with preset standard thresholds, students are classified into corresponding cognitive levels; at the same time, changes in student learning data are tracked, and level assignments are adjusted in real time.
5. The system according to claim 1, characterized in that, The specific process of the differentiated cognitive profile is as follows: extract key information from the multi-dimensional cognitive feature vector, including knowledge structure weaknesses, thinking tendencies, and learning style preferences; match the key information with a preset feature tag library and calculate the matching degree of each tag; select the matching tag combination to form a basic profile framework.
6. The system according to claim 1, characterized in that, The specific process of the cognitive level and task mapping rule base is as follows: The cognitive level and task mapping rule base presets the task difficulty, knowledge point coverage, and training focus for each cognitive level; retrieves the corresponding basic task template according to the student's cognitive level; makes detailed adjustments to the task template by combining the weaknesses and learning style tags in the differentiated cognitive profile; supplements targeted reinforcement content and combines them to form a personalized training task matrix.
7. The system according to claim 1, characterized in that, The specific process of the metacognitive question chain library is as follows: based on the identified error paradigm type, the corresponding basic question template in the metacognitive question chain library is retrieved; the question expression is adjusted according to the preset progressive logic; and a guiding angle that matches the student's cognitive profile is incorporated to form a progressive question sequence that promotes students' independent reflection.
8. The system according to claim 1, characterized in that, The specific process of generating scenario-based verification tasks is as follows: based on the learning style and life-related information in the cognitive profile, select familiar life scenarios, campus scenarios, or practical scenarios for students; design a verification task framework that includes problem context, task objectives, operation steps, and verification standards. The task incorporates prompts and knowledge application guidance; the completion of the task is used to test students' ability to correct errors and apply knowledge in practice, forming a set of error correction and knowledge reinforcement that includes task data repair effect data.
9. The system according to claim 1, characterized in that, The specific process of constructing the dynamic mastery assessment index system is as follows: setting multi-level evaluation standards and quantitative score ranges for each dimension, determining the weight coefficients of each dimension through the analytic hierarchy process; quantitatively scoring each dimension by combining students' answer data, behavioral data, and result data in training tasks; and comprehensively judging students' knowledge mastery status based on the scoring results.
10. The system according to claim 1, characterized in that, The specific process for generating a multi-dimensional capability feedback report is as follows: classifying and organizing according to cognitive levels; presenting the performance of each dimension using visual charts; combining textual descriptions to analyze strengths and weaknesses, proposing targeted training and adjustment suggestions, and forming a structured feedback report that includes data visualization analysis conclusions and suggestions.