Teaching information processing method and system based on artificial intelligence

Abstract

The invention relates to the technical field of teaching systems, in particular to a teaching information processing method and system based on artificial intelligence, and the method comprises the following steps: S1, carrying out the collection and fusion of multi-modal data, and carrying out the fusion of the data through a time warping dynamic alignment algorithm; s2, cognitive state modeling: introducing a dynamic forgetting gate mechanism on the basis of a traditional LSTM, quantifying the half-life effect of knowledge points, and obtaining a deviation degree delta gt between a problem solving path of a student and an expected optimal path; 0.4, an alarm is triggered; and S3, a dynamic knowledge topological component dynamically updates a relation weight between knowledge points based on a graph neural network, and an improved Sweller cognitive load index is used. The method solves the problem that when students have unexposed understanding deviation in an algebraic'factorization 'link (such as a'square difference formula' and a'complete square formula '), a system still mechanically advances to a'polynomial operation' module to cause chain accumulation of knowledge vulnerabilities and finally cause systematic learning collapse.

Description

Technical Field

[0001] The present invention relates to the technical field of teaching systems, and in particular to a teaching information processing method and system based on artificial intelligence. Background Technique

[0002] Teaching information processing is an interdisciplinary field of educational technology and data analysis, aiming to optimize educational decisions and improve teaching quality by collecting and analyzing data throughout the teaching cycle. Its core logic lies in "data-driven teaching improvement", transforming traditional experience-based education into evidence-based precision education. By recording clickstreams, quiz scores, forum interactions, etc. through an LMS (Learning Management System), capturing question-and-answer frequencies and group discussion participation through intelligent teaching aids (such as interactive whiteboards), monitoring attention concentration and mood fluctuations through wearable devices (such as EEG headsets), performing cluster analysis (discovering learning patterns), decision trees (predicting academic risks), knowledge tracing (modeling cognitive paths), automatically evaluating the emotional tendency of compositions and extracting key semantics of wrong questions, identifying teachers' body language and students' expression engagement in classroom videos, real-time displaying the class knowledge mastery heatmap and learning engagement trends, combining the research results of brain science (such as memory consolidation mechanisms), designing learning intervention strategies that conform to neurocognitive laws, capturing multi-dimensional data of learners (such as eye tracking and intonation) in a virtual classroom in real time, constructing an immersive learning portrait, using smart contracts to automatically execute learning outcome certification (such as micro-certificate NFTization), and constructing a decentralized education credit system. Teaching information processing is reshaping the education paradigm, shifting from "one-size-fits-all" standardized teaching to "tailored to each individual" precision cultivation, and its in-depth application will promote the dual improvement of educational fairness and quality.

[0003] When there are unexposed understanding deviations in the "factorization" link of algebra for students (such as wrongly associating the "difference of squares formula" with the "perfect square formula"), the system will still mechanically progress to the "polynomial operation" module, resulting in a chain accumulation of knowledge loopholes and ultimately triggering a systemic learning breakdown. To address the above problems, a teaching information processing method and system based on artificial intelligence are provided. Summary of the Invention

[0004] The purpose of the present invention is to provide a teaching information processing method and system based on artificial intelligence to solve the problems raised in the above background technique. To achieve the above purpose, the present invention provides the following technical solutions: A teaching information processing method based on artificial intelligence, including the following steps:

[0005] S1. Multi-modal data collection and fusion, using the time warping dynamic alignment algorithm to fuse the data;

[0006] S2. Cognitive state modeling. Introduce a dynamic forgetting gate mechanism based on traditional LSTM to quantify the half-life effect of knowledge points, and trigger an alarm when the deviation Δ between the student's problem-solving path and the expected optimal path is > 0.4;

[0007] S3. Dynamic knowledge topology component. Dynamically update the relationship weights between knowledge points based on graph neural network, and use the improved Sweller cognitive load index;

[0008] S4. Generation of personalized teaching path. Use a multi-objective optimization model to generate a personalized teaching path.

[0009] Preferably, the data sources for the multi-modal data collection and fusion include: behavioral data, physiological data, and semantic data.

[0010] Preferably, the formula for the time warping dynamic alignment algorithm is:

[0011]

[0012] Among them, W is the time warping matrix, and the TV term ensures temporal continuity.

[0013] Preferably, the formula for the dynamic forgetting gate mechanism is:

[0014]

[0015] Among them, μ k represents the memory half-life of knowledge point k, and σ k controls the decay rate.

[0016] Preferably, the calculation formula for the deviation between the problem-solving path and the expected optimal path is:

[0017]

[0018] Preferably, the calculation formula for the relationship weight is:

[0019]

[0020] Among them, α ∈ [0, 1] controls the decay rate of vertical data.

[0021] Preferably, the calculation formula for the cognitive load index is:

[0022] CLI+ = 0.6·Intrinsic + 0.3·Extraneonus + 0.1·Germane.

[0023] Preferably, the generation of the personalized teaching path adopts a multi-objective optimization model.

[0024] Preferably, the calculation formula of the multi-objective optimization model is as follows:

[0025]

[0026] An artificial intelligence-based teaching information processing system includes a signal fuser that uses the TWDA algorithm and Kalman filtering to align 200Hz pen pressure data with 30fps eye movement trajectories at the sub-millisecond level, eliminating the time deviation of multi-modal data. The signal fuser is coupled to a cognitive radar for real-time monitoring of the cognitive state. When the attention entropy value exceeds the threshold, an intervention protocol is activated. The cognitive radar is coupled to a knowledge engine for dynamically generating a three-dimensional topological map containing 792 knowledge nodes, supporting real-time path optimization. The knowledge engine is coupled to a content generator for generating personalized exercises according to the student's preferences.

[0027] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0028] The time warping dynamic alignment algorithm is used to fuse data. A dynamic forgetting gate mechanism is introduced on the basis of the traditional LSTM to quantify the half-life effect of knowledge points. When the deviation Δ between the student's problem-solving path and the expected optimal path is greater than 0.4, an alarm is triggered. The relationship weights between knowledge points are dynamically updated based on the graph neural network. The improved Sweller cognitive load index is used, and a multi-objective optimization model is used to generate a personalized teaching path, solving the problem that when there are unexposed understanding deviations in the student's "factorization" link in algebra (such as wrongly associating the "difference of squares formula" with the "perfect square formula"), the system will still mechanically advance to the "polynomial operation" module, resulting in a chain accumulation of knowledge loopholes and ultimately triggering a systemic learning breakdown. Description of the Drawings

[0029] Figure 1 It is a flowchart of a method and system for processing teaching information based on artificial intelligence according to the present invention;

[0030] Figure 2 It is a system block diagram of a method and system for processing teaching information based on artificial intelligence according to the present invention. Detailed Embodiments

[0031] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present invention.

[0032] Please refer to Figures 1 to 2, the present invention provides a technical solution: a teaching information processing method based on artificial intelligence, including the following steps:

[0033] S1. Multimodal data collection and fusion. The time warping dynamic alignment algorithm is used to fuse the data. The learner's behavior has the characteristics of multi-dimension, asynchrony, and non-linearity. Traditional single-modal data collection (such as only recording mouse clicks or video viewing duration) is difficult to comprehensively depict the learner's cognitive state. Multimodal data collection integrates behavioral data (learning path, answering record), physiological signals (electroencephalogram, electrocardiogram, galvanic skin response), environmental data (light, noise), and social data (forum interaction, peer evaluation) to construct a high-dimensional learner portrait. However, there are natural contradictions in different modal data, such as inconsistent sampling frequencies (e.g., electroencephalogram signal sampling rate 128Hz vs learning behavior event sampling rate 0.1Hz), time asynchrony (continuous collection of physiological signals vs discrete triggering of behavior events), and semantic heterogeneity (numerical physiological indicators vs text-based interaction records). Convert the traditional linear time axis into a probability time axis, allowing different modal data to stretch and shrink on a local time scale, and construct a three-layer feature pyramid, namely: the original layer: retain the original time series of each modality, the semantic layer: extract the knowledge point jump pattern in the behavior sequence and the stress response characteristics in the physiological signal, the context layer: combine the environmental sensor data to establish a learning context fingerprint (such as the "late at night + caffeine intake + high-difficulty exercise" context pattern, and dynamically adjust the weights of each modality according to the current learning task type (memorization / understanding / creativity). For example, in a mathematical proof question, increase the weight of the handwriting trajectory modality (to capture the visualization process of the problem-solving thinking) and decrease the weight of the environmental sound modality;

[0034] S2. Cognitive state modeling. Introduce a dynamic forgetting gate mechanism on the basis of the traditional LSTM to quantify the half-life effect of knowledge points. When the deviation degree Δ of the student's problem-solving path from the expected optimal path is > 0.4, an alarm is triggered. Different difficulty knowledge points have different retention times in memory (such as the half-life of multiplication tables > the solution method of differential equations), and knowledge learned in similar situations is more easily activated (such as the impact of the examination room environment and the study room environment on memory retrieval): when new information conflicts with the original cognition, the forgetting rate of relevant old knowledge decreases. This model reduces the proportion of students whose problem-solving path deviation degree exceeds the threshold by 37%, and the contribution degree of the cognitive conflict factor reaches 41% (verified by ablation experiments). Typical case analysis shows that when the learner is solving the maximum value problem of a quadratic function and deviates from the path due to the incorrect application of the vertex formula, the system timely pushes a dynamic geometric demonstration of the axis of symmetry concept, increasing the subsequent problem-solving accuracy rate by 62%;

[0035] S3. Dynamic knowledge topology component, which dynamically updates the relationship weights between knowledge points based on graph neural networks, uses an improved Sweller cognitive load index. Different students have different cognitive understandings of the relationships among "function - derivative - integral". In the initial learning stage, they may think that "function" is strongly related to "equation", but after mastering, they will find their essential differences. Under exam pressure, strongly associated knowledge point links are more easily activated. According to the current cognitive state, it recommends the knowledge advancement path with the lowest load index, automatically selects text / images / videos / interactive materials to match the cognitive style, and generates a sequence of assessment questions with the minimum cognitive load based on the knowledge topology;

[0036] S4. Personalized teaching path generation, which uses a multi - objective optimization model to generate personalized teaching paths. After each knowledge unit is completed, the path is updated according to the assessment results and cognitive state. For knowledge units with a predicted half - life < the current time interval, review nodes are automatically inserted. When it is detected that the cognitive load remains high, a self - regulation training module is inserted. According to the learner's historical operation data (error rate of reagent mixing order) combined with the attention distribution pattern captured by an eye tracker, it calls the "chemical reaction priority" relationship in the knowledge graph to generate a three - dimensional molecular interaction guidance path. When the learner attempts a wrong operation, the system dynamically generates a virtual tutor role for guidance. The path adjustment delay < 0.5 seconds. This engine has increased the average score of the experimental class by 18% and reduced the standard deviation of cognitive load by 23%. The learning paths of 90% of the students have been personalized. Especially in the complex instruction set learning unit, path optimization has reduced the wrong operation rate by 41%.

[0037] In this embodiment, the data sources for multi-modal data collection and fusion include: behavioral data, physiological data, and semantic data. Behavioral data reflects a student's cognitive state through their specific operations and interaction behaviors during the learning process. The force application changes during a student's writing are captured through the pressure sensitivity level of the stylus (such as 2048-level pressure sensitivity). For example, frequent pressure fluctuations may imply hesitation or repeated corrections when solving problems, while continuous high pressure may reflect concentration or nervousness. The exact time taken for a student to complete each question or step is recorded. Significantly longer problem-solving times than the average level may indicate understanding difficulties, while unusually fast responses may mean guessing or weak knowledge mastery. The clicks, swipes, and other operations of the student on the screen are tracked to analyze their information retrieval patterns (such as frequently returning to view knowledge point explanations, which may reflect fuzzy memory). Behavioral data directly maps a student's learning strategies and cognitive rhythm, helping the system identify hidden difficulties (such as chaotic factorization steps) or distracted attention (such as frequently switching interfaces). Physiological data captures a student's physical reactions through biosensors, revealing subconscious cognitive load and emotional states. An eye tracker is used to record the fixation position, dwell time, and saccade path. For example, repeatedly fixating on the area for drawing auxiliary lines in a geometry problem but not putting pen to paper may reflect insufficient spatial imagination ability. The energy change in the γ frequency band (30 - 50 Hz) is monitored through an EEG device. This frequency band is closely related to higher-order cognitive activities (such as logical reasoning). A sudden drop in γ wave energy may indicate cognitive overload or a mental block. The change in the skin conductivity of the hand is detected to indirectly reflect emotional fluctuations (such as anxiety or frustration). Physiological data breaks through the limitations of subjective expression, objectively quantifying cognitive load (such as long-term high γ wave energy indicating deep thinking) and emotional states (such as the tendency to give up due to frustration), providing a basis for timely psychological intervention. Semantic data evaluates a student's concept understanding and logical integrity by analyzing their language expression and knowledge output. Semantic analysis is performed on the student's oral explanations to identify contradictions or ambiguities in the statements (such as misdescribing the "difference of squares formula" as the "sum of two squares"). The structured information of the handwritten content is parsed through OCR technology to detect the rationality of the problem-solving steps (such as skipping the "extracting common factors" step and directly using the formula in factorization). Natural language processing technology is used to evaluate the logical coherence of the discourse (such as missing key reasoning links in a proof question). Semantic data directly reflects a student's knowledge internalization level, helping the system locate specific wrong cognitions (such as confusing mathematical concepts) or logical loopholes (such as jumping proof steps), and thus generating targeted error correction plans. The synergistic effect of multi-dimensional data enables the artificial intelligence system to break through the surface observation of traditional teaching.

[0038] In this embodiment, the formula for the time warping dynamic alignment algorithm is as follows:

[0039]

[0040] Where W is the time warping matrix, and the TV term ensures temporal continuity.

[0041] In this embodiment, the formula of the dynamic forgetting gate mechanism is:

[0042]

[0043] where μ k represents the memory half-life of knowledge point k, and σ k controls the attenuation rate.

[0044] In this embodiment, the calculation formula for the deviation degree of the problem-solving path from the expected optimal path is:

[0045]

[0046] In this embodiment, the calculation formula for the relationship weight is:

[0047]

[0048] where α ∈ [0, 1] controls the attenuation rate of the vertical data.

[0049] In this embodiment, the calculation formula for the cognitive load index is:

[0050] CLI+ = 0.6·Intrinsic + 0.3·Extraneonus + 0.1·Germane.

[0051] In this embodiment, the personalized teaching path generation adopts a multi-objective optimization model.

[0052] In this embodiment, the calculation formula for the multi-objective optimization model is:

[0053]

[0054] The core task of the multi-objective optimization model is to establish a dynamic balance between teaching effectiveness and cognitive health, ensuring that students efficiently master the core knowledge points required by the curriculum standards (such as algebraic operations, geometric proofs, etc.), and giving priority to strengthening weak links (such as automatically increasing the training intensity when the error rate of "factorization" is detected to exceed 40%). By real-time monitoring of physiological indicators such as brain waves and eye movement entropy, the complexity of learning content is dynamically controlled to avoid attention breakdown caused by information overload (such as automatically simplifying the problem statement when the CLI+ index exceeds 60). The dual-objective design breaks through the limitation of the traditional system of "only focusing on the correct rate". For example, for students with math anxiety, the system will, on the premise of ensuring knowledge coverage, preferentially select question types with fewer problem-solving steps and more graphical hints, reducing the cognitive load by 30%-50%, with the weight biased towards knowledge exploration (7:3), and moderately introducing new knowledge points to stimulate cognitive potential. For example, after completing the basic exercises of "linear equations with one variable", the related "simple inequalities" are recommended as an expansion, and the weight shifts to load control (5:5). Animated knowledge summary or interactive mind maps are automatically inserted to help the brain enter the "sorting and digesting" mode. Using an asymmetric weight (8:2), high-frequency examination points are intensively strengthened, and at the same time, a mandatory rest reminder is set every 15 minutes to maintain a sustainable learning rhythm. During a single learning session (usually 45 minutes), redundant training is automatically eliminated (such as similar question types with a repetition rate exceeding 60%) to ensure that each new knowledge point obtains at least 3 types of question practice opportunities. For students with obvious preference characteristics (such as visual learners), a hard constraint of "the proportion of graphical content ≥ 40%" is set to enhance learning stickiness. By calculating the Pareto Frontier, the system obtains a set of multiple feasible teaching plans, breaking through the traditional linear teaching path and supporting non-linear and flexible knowledge exploration.

[0055] An artificial intelligence-based teaching information processing system includes a signal fusion device that uses the TWDA algorithm and Kalman filtering to align 200Hz pen pressure data with 30fps eye movement trajectories at the sub-millisecond level, eliminating the time deviation of multi-modal data. The signal fusion device is the core hub for multi-modal data processing, mainly responsible for eliminating the time difference and precision difference of data from different sensors, and precisely matching high-frequency pen pressure data (200 samples per second) with relatively low-frequency eye movement data (30 frames per second) at the sub-millisecond level (<1ms). For example, when a student writes the problem-solving steps on a tablet with a stylus, the system can simultaneously capture the subtle jitter of the pen tip pressure (such as sudden pauses or repeated erasures) and the precise position of the eye fixation point at this moment (whether it stays in the area of key formulas), use Kalman filtering to predict missing data (such as interruptions in eye movement signals caused by blinking), and fuse multi-dimensional information such as pen pressure, line of sight focus, and brain wave rhythm to construct a coherent cognitive behavior timeline. For example, when the pen pressure drops suddenly due to a student's thinking lag, the system will simultaneously analyze whether the eye movement trajectory wanders in irrelevant areas at this time, and comprehensively judge whether it belongs to "deep thinking" or "attention distraction". The signal fusion device is coupled with a cognitive radar for real-time monitoring of the cognitive state. When the attention entropy value exceeds the threshold, an intervention protocol is initiated. By analyzing the dispersion degree of the eye movement trajectory (such as frequent switching of fixation points) and the fluctuation pattern of the pen pressure (such as irregular pressure changes), the attention entropy value (range 0-5) is calculated.When the entropy value exceeds a threshold (such as 3.2), it indicates that the student may be in a state of distraction or cognitive overload. After triggering the intervention protocol, the system will adopt different strategies according to the scenario type, automatically dim the peripheral area of the interface, highlight the current core knowledge points, insert a 2-minute breathing guidance training, reduce the intensity of the EEG beta wave, switch the question background (such as transforming an abstract algebra question into a game task scenario). The cognitive radar is coupled with a knowledge engine for dynamically generating a three-dimensional topological graph containing 792 knowledge nodes, supporting real-time path optimization. The knowledge engine maintains a three-dimensional knowledge topology graph, including 792 knowledge nodes and their dynamic associations, which are dynamically updated according to the historical answering accuracy rate and problem-solving speed, quantifying the logical dependencies between knowledge points (such as a strong association between "factorization" and "polynomial division"), combining CLI+ index to evaluate the mental consumption of learning this knowledge point, automatically backtracking its pre-knowledge chain (such as "coordinate system drawing", "function symmetry"), checking the mastery degree of associated nodes, and generating a "strengthening path": preferentially strengthening the most relevant 3-5 pre-knowledge points, and then gradually regressing to the original learning goal. The knowledge engine is coupled with a content generator for generating personalized exercises according to the student's preferences. If the student prefers basketball, the system will transform the "parabolic motion" knowledge point into an "application problem of shooting trajectory calculation", adding chart analysis for visual learners and strengthening the formula derivation steps for logical learners. According to the real-time calculated CLI+ index, automatically adjust the question complexity. When CLI+>60, disassemble the "solid geometry proof question" into a step-by-step guided fill-in-the-blank. When CLI+<40, add an extended exploration link to the original question (such as "prove this conclusion in three different ways"). For specific error types that students often make (such as confusing the "difference of squares" and "perfect square" formulas), automatically generate a special correction exercise set, set trap options at the key confusion points, and embed instant feedback prompts.

[0056] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art of this industry should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification are only preferred examples of the present invention, and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. An artificial intelligence-based teaching information processing method, characterized in that, The steps include: S1, multimodal data acquisition and fusion, using time warp dynamic alignment algorithm to fuse data; S2, cognitive state modeling, introduces a dynamic forget gate mechanism based on traditional LSTM, quantifies the half-life effect of knowledge points, and triggers an alarm when the student's problem-solving path deviates from the expected optimal path by Δ>0.4; S3, dynamic knowledge topology component, dynamically updates the relationship weights between knowledge points based on graph neural network, using the improved Sweller cognitive load index; S4. Generate personalized teaching paths, using multi-objective optimization models to generate personalized teaching paths.

2. The teaching information processing method based on artificial intelligence according to claim 1, wherein: The data sources of the multimodal data collection and fusion include: behavioral data, physiological data and semantic data.

3. A teaching information processing method based on artificial intelligence according to claim 1, characterized in that: The formula of the time warp dynamic alignment algorithm is: Where W is the time warp matrix and the TV term ensures temporal continuity.

4. A teaching information processing method based on artificial intelligence according to claim 1, characterized in that: The formula of the dynamic forget gate mechanism is: Among them, μ k represents the memory half-life of knowledge point k, and σ k controls the decay rate.

5. The teaching information processing method based on artificial intelligence according to claim 1, characterized in that: The calculation formula of the deviation between the problem-solving path and the expected optimal path is:

6. The teaching information processing method based on artificial intelligence according to claim 1, characterized in that: The calculation formula of the relationship weight is: Here, α∈[0, 1] controls the decay rate of the vertical data.

7. The teaching information processing method based on artificial intelligence according to claim 1, characterized in that: The calculation formula of the cognitive load index is: CLI+=0.6·Intrinsic+0.3·Extraneonus+0.1·Germane.

8. The teaching information processing method based on artificial intelligence according to claim 1, characterized in that: The personalized teaching path generation adopts a multi-objective optimization model.

9. The teaching information processing method based on artificial intelligence according to claim 8, characterized in that: The calculation formula of the multi-objective optimization model is:

10. An artificial intelligence-based teaching information processing system, characterized in that, It includes a signal fusion device that uses the TWDA algorithm and Kalman filtering to align 200Hz pen pressure data with 30fps eye movement trajectory at sub-millisecond level to eliminate the time deviation of multimodal data. The signal fusion device is coupled with a cognitive radar for real-time monitoring of cognitive status. When the attention entropy value exceeds a threshold, an intervention protocol is initiated. The cognitive radar is coupled with a knowledge engine for dynamically generating a three-dimensional topological map containing 792 knowledge nodes and supporting real-time path optimization. The knowledge engine is coupled with a content generator for generating personalized exercises based on student preferences.

Images