Learning optimization methods and systems based on cognitive style fusion memory half-life prediction
By constructing a learner multimodal physiological-behavioral rhythm map and a knowledge resonance field model, the problem of incomplete perception of learner state in existing learning optimization systems is solved, achieving precise and emotionally adaptive optimization of the learning process, and improving learning effectiveness and participation.
Patent Information
- Application Number
- CN202610397359.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-30
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2046-03-30
AI Technical Summary
Existing technologies in learning optimization systems fail to fully perceive learners' states, especially their emotional and motivational states, and fail to effectively simulate the dynamic memory characteristics of knowledge. Furthermore, the synergistic effect between different technology modules is insufficient, resulting in inadequate intelligence and flexibility in intervention strategies, making it difficult to achieve in-depth optimization.
By acquiring learners' multimodal physiological-behavioral rhythms, constructing a dynamic rhythm map, and combining it with the memory decay rhythm in the knowledge graph, we can simulate the energy transmission and interference between knowledge nodes, generate rhythm tuning intervention strategies, and achieve a learning optimization closed loop.
It improves the accuracy and foresight of learning optimization, enables continuous and in-depth insights into learners' psychological states, proactively intervenes in knowledge content at the highest risk of forgetting, promotes the cohesion and anti-interference ability of cognitive networks, and enhances the learning experience and long-term engagement.
Smart Images

Figure CN122264216B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of information technology, and in particular to a learning optimization method and system based on cognitive style fusion memory half-life prediction. Background Technology
[0002] With the rapid development of information technology, the education field has gradually introduced intelligent and personalized learning systems to adjust teaching content and methods according to individual learners' differences. These systems typically utilize big data analytics to construct knowledge graphs representing the relationships between learning content and track learners' learning progress, answer results, and other behavioral data. Some more advanced systems are beginning to incorporate physiological signals, such as electroencephalography (EEG) and eye movements, to assess learners' attention levels or cognitive load, thereby dynamically adjusting learning paths or providing immediate feedback. These technologies have, to some extent, improved learning efficiency and personalization, providing a preliminary technological framework for adaptive learning.
[0003] Chinese Patent Publication No. CN119476624A discloses an adaptive learning optimization method and system based on big data analysis, relating to the field of machine learning technology. The method includes the following steps: deploying computing nodes on the learner's device to acquire learning data, and configuring an edge computing gateway to transmit the processed learning data to a cloud computing center; analyzing the learning data through the cloud computing center to obtain the learner's learning status; and providing a strategy optimization scheme based on the learning status. This invention, through personalized learning path recommendations, allows learners to master knowledge points more quickly, reducing exploration time in areas they are not proficient in. Targeted learning content pushes and exercises can help learners improve in courses where they perform poorly, thereby improving overall academic performance. Intermittent transmission of learning data reduces the computational load on the entire device, and the judgment of analysis results ensures the accuracy of the learning data and the precision of the analysis results.
[0004] Existing technologies typically employ the following solutions: First, behavioral data analysis, which infers learners' learning status and knowledge mastery by recording their operation logs, practice results, and learning duration on learning platforms; second, knowledge graph construction and recommendation, which organizes subject knowledge into a graph structure and uses recommendation algorithms to provide learners with relevant learning resources and practice questions; and third, physiological signal monitoring, which uses devices such as EEG and eye-tracking to acquire learners' physiological data and extract simple physiological indicators such as attention and cognitive load. These technologies usually operate independently or only perform basic information integration, forming a linear "perception-processing-feedback" process.
[0005] However, existing technical solutions have significant technical shortcomings. First, their perception of learners' states is not comprehensive or in-depth enough, particularly failing to incorporate learners' emotional and motivational states into core considerations. This means intervention strategies may not be effective in addressing learning inefficiencies caused by psychological factors. For example, simply perceiving high cognitive load cannot distinguish between positive challenges due to the difficulty of the content and negative pressure due to excessive fatigue. Second, knowledge models generally lack simulation of the dynamic memory characteristics of knowledge itself, failing to fully utilize cognitive psychology principles such as the forgetting curve for accurate prediction, and failing to effectively simulate the complex reinforcement and interference relationships between knowledge points. This results in a certain lag and crudeness in the system's review reminders and knowledge reinforcement. Third, the synergistic effect between different technical modules is insufficient. Data integration remains at the level of simple information aggregation, failing to form a unified mechanism that can drive the interaction and coordination of information across various dimensions. This leads to insufficient intelligence and flexibility in intervention strategies, making it difficult to achieve deep optimization of the learning process. Summary of the Invention
[0006] To address the aforementioned issues, this invention provides a learning optimization method and system based on cognitive style fusion memory half-life prediction. It dynamically couples and tunes the learner's multimodal physiological-behavioral rhythms with the memory decay rhythms of the knowledge resonance field, and drives predictive intervention strategies through the resonance rhythm center. This significantly improves the accuracy, emotional adaptability, and long-term effects of learning optimization.
[0007] The above objectives can be achieved through the following approach: Learning optimization methods based on cognitive style fusion for predicting memory half-life include: Acquire learners' real-time physiological and behavioral time-series signals; Rhythm coupling analysis is performed on real-time physiological time-series signals and real-time behavioral time-series signals to generate learner dynamic rhythm maps; Obtain the inherent memory decay rhythm of each knowledge node in the knowledge graph; Based on the global rhythm stability index in the learner's dynamic rhythm map, the intrinsic memory decay rhythm is dynamically adjusted to generate personalized memory decay rhythms for each knowledge node. Based on the personalized memory decay rhythm and the semantic relationship between each knowledge node, the energy propagation and interference between knowledge nodes are simulated to generate the rhythmic state of the knowledge resonance field. Based on learners’ dynamic rhythm map and knowledge resonance field rhythm state, rhythm matching diagnosis is performed through the resonance rhythm center to generate rhythm tuning intervention strategies. Implement rhythm tuning intervention strategies and collect learner feedback data after implementation; By updating learner dynamic rhythm maps and knowledge resonance field rhythm states using learner feedback data, a learning optimization loop is completed.
[0008] Optionally, the step of performing rhythm coupling analysis on real-time physiological time-series signals and real-time behavioral time-series signals to generate a learner's dynamic rhythm map includes: extracting at least one physiological rhythm sub-signal from the real-time physiological time-series signals; extracting at least one behavioral rhythm sub-signal from the real-time behavioral time-series signals; calculating the real-time phase synchronization between different physiological rhythm sub-signals and between physiological rhythm sub-signals and behavioral rhythm sub-signals; constructing dynamic connection weights based on the real-time phase synchronization, using each rhythm sub-signal as a node, and generating the learner's dynamic rhythm map using the dynamic connection weights.
[0009] Optionally, the step of dynamically adjusting the intrinsic memory decay rhythm based on the global rhythm stability index in the learner's dynamic rhythm map to generate a personalized memory decay rhythm for each knowledge node includes: analyzing the phase synchronization patterns between nodes in the learner's dynamic rhythm map, identifying pressure coupling patterns or disordered coupling patterns; calculating the global rhythm stability index based on the identified coupling patterns; and injecting an accelerated decay factor or a decay inhibition factor into the intrinsic memory decay rhythm of all knowledge nodes according to the global rhythm stability index to generate the personalized memory decay rhythm.
[0010] Optionally, the step of simulating energy propagation and interference between knowledge nodes based on personalized memory decay rhythm and semantic relationships between knowledge nodes to generate the knowledge resonance field rhythm state includes: using the personalized memory decay rhythm of each knowledge node as the fundamental frequency, and combining it with its current memory energy amplitude to generate a knowledge memory oscillation wave; propagating the knowledge memory oscillation wave in the knowledge graph according to the strength of semantic relationships between knowledge nodes; when the knowledge memory oscillation waves of different knowledge nodes meet on the associated path, calculating the positive resonance enhancement effect or negative resonance interference effect based on the similarity of their fundamental frequencies, updating the current memory energy amplitude of each knowledge node, and comprehensively obtaining the knowledge resonance field rhythm state.
[0011] Optionally, the step of generating a rhythm tuning intervention strategy based on the learner's dynamic rhythm map and the rhythm state of the knowledge resonance field, through the resonance rhythm center, includes: identifying a set of key knowledge nodes whose memory energy amplitude is below a threshold and is in a rapidly decaying phase from the rhythm state of the knowledge resonance field; diagnosing the learner's current cognitive load rhythm pattern and emotional rhythm pattern from the learner's dynamic rhythm map; calculating the matching degree between the inherent knowledge rhythm characteristics of the key knowledge node set and the learner's current cognitive load rhythm pattern and emotional rhythm pattern in the resonance rhythm center; and generating the rhythm tuning intervention strategy, which includes a pre-learning rhythm guidance task and a main learning task, wherein the pre-learning rhythm guidance task is used to adjust the learner's emotional rhythm pattern to a state that matches the rhythm required by the main learning task.
[0012] Optionally, the steps of implementing the rhythm tuning intervention strategy and collecting learner feedback data after implementation include: presenting the pre-implementation rhythm guidance task to the learner and monitoring the learner's rhythm changes in real time; when the learner's emotional rhythm pattern is detected to reach the expected matching state, presenting the main learning task to the learner; and recording the learner's interaction data and subsequent physiological time-series signals during the completion of the main learning task as the learner feedback data.
[0013] Optionally, the step of updating the learner dynamic rhythm map and the rhythm state of the knowledge resonance field using learner feedback data includes: updating the current memory energy amplitude of nodes in the key knowledge node set using the interaction data in the learner feedback data; analyzing the physiological time-series signals in the learner feedback data to obtain new rhythm sub-signals and updating the dynamic connection weights in the learner dynamic rhythm map; and recalculating the positive or negative resonance effect between the key knowledge node set and its associated nodes based on the updated current memory energy amplitude, thereby updating the rhythm state of the knowledge resonance field.
[0014] Optionally, after updating the knowledge resonance field rhythm state, the method further includes: generating a rhythm tuning effect evaluation based on a comparison of the knowledge resonance field rhythm state before and after the update, and a comparison of the learner's dynamic rhythm map before and after implementing the intervention strategy; using the rhythm tuning effect evaluation as a reward signal, inputting it into a reinforcement learning model to optimize the rhythm matching diagnosis rules in the resonance rhythm center and the generation parameters of the rhythm tuning intervention strategy.
[0015] Optionally, obtaining the intrinsic memory decay rhythm of each knowledge node in the knowledge graph includes: obtaining the static attributes of each knowledge node in the knowledge graph, wherein the static attributes include knowledge type, abstraction level, and historical mastery difficulty coefficient; and calculating the intrinsic memory decay rhythm of each knowledge node based on the static attributes through a rhythm mapping model.
[0016] This invention also provides a learning optimization system based on cognitive style fusion memory half-life prediction, used to implement the above-mentioned learning optimization method, the system comprising: The data acquisition module is used to acquire the learner's real-time physiological and behavioral time-series signals. The learner rhythm analysis module is used to perform rhythm coupling analysis on the real-time physiological time-series signal and the real-time behavioral time-series signal, and to generate and maintain the learner's dynamic rhythm map. The knowledge rhythm management module is used to manage the knowledge graph, store and dynamically adjust the inherent memory decay rhythm and personalized memory decay rhythm of each knowledge node; The knowledge resonance field simulation module is used to simulate energy propagation and interference based on the personalized memory decay rhythm and inter-node relationships of knowledge nodes, and to generate and update the rhythm state of the knowledge resonance field. The resonant rhythm central decision-making module is used to perform rhythm matching diagnosis based on the learner's dynamic rhythm map and the rhythm state of the knowledge resonance field, and generate rhythm tuning intervention strategies. The intervention execution and feedback module is used to execute the rhythm tuning intervention strategy, collect learner feedback data after execution, and send the learner feedback data to the learner rhythm analysis module and the knowledge resonance field simulation module to complete the closed-loop update.
[0017] Compared with the prior art, the present invention has the following advantages: This invention improves the accuracy and foresight of the learning optimization process by constructing a dual-closed-loop dynamic coupling model. Unlike traditional systems that only passively respond to discrete events, this invention can continuously and deeply understand the learner's physiological, emotional, and cognitive load states, and dynamically integrate them with the personalized memory decay patterns of knowledge points. This enables proactive intervention on knowledge content at the highest risk of forgetting when the learner is at their optimal psychological readiness and most receptive learning time. This rhythm-matching-based collaborative decision-making mechanism ensures the effectiveness and relevance of teaching interventions, reducing ineffective learning and knowledge forgetting.
[0018] This invention achieves intelligent diagnosis and resolution of cognitive conflicts by simulating the resonance and interference mechanism of knowledge memory. Traditional knowledge models only focus on the logical relationship of knowledge and are difficult to reveal potential confusion caused by conceptual similarity or ambiguity of representation. This invention introduces dynamic simulation of knowledge resonance field, which can warn and locate knowledge points that may produce negative resonance (mutual interference). On this basis, combined with the learner's real-time cognitive elasticity, it actively designs and guides learners to conduct dialectical thinking and comparative analysis, prompting learners to actively reconstruct at the knowledge structure level. This not only eliminates existing cognitive obstacles, but also fundamentally strengthens the correct connection between knowledge points and builds a cognitive network with greater cohesion and anti-interference ability.
[0019] This invention achieves a high degree of personalization and emotional support in the learning process, enhancing the learning experience and learners' long-term engagement. Traditional adaptive learning often focuses on the delivery of knowledge content and adjustment of difficulty, neglecting learners' emotional fluctuations and psychological needs. This invention incorporates learners' emotional and motivational states into a cognitive coupling network, enabling it to perceive learners' fatigue, anxiety, or burnout, and integrates elements of emotion regulation and interest stimulation into intervention strategies. This humanized intervention not only alleviates learning pressure and stimulates learning interest but also cultivates learners' intrinsic learning motivation, making learners feel understood and supported, thereby promoting the continuous improvement of learning effectiveness and the formation of good learning habits. Attached Figure Description
[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0021] Figure 1 This is a flowchart illustrating the learning optimization method based on cognitive style fusion memory half-life prediction in this invention. Figure 2 This is a schematic diagram of the 3D model of the knowledge resonance field rhythm state of the learning optimization method based on cognitive style fusion memory half-life prediction in this invention; Figure 3 This is a schematic diagram comparing the rhythm tuning intervention effects of the learning optimization method based on cognitive style fusion memory half-life prediction in this invention. Detailed Implementation
[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.
[0023] Reference Figure 1 As shown, one embodiment of the present invention proposes a learning optimization method based on cognitive style fusion memory half-life prediction, including: Acquire learners' real-time physiological and behavioral time-series signals; Rhythm coupling analysis is performed on real-time physiological time-series signals and real-time behavioral time-series signals to generate learner dynamic rhythm maps; Obtain the inherent memory decay rhythm of each knowledge node in the knowledge graph; Based on the global rhythm stability index in the learner's dynamic rhythm map, the intrinsic memory decay rhythm is dynamically adjusted to generate personalized memory decay rhythms for each knowledge node. Based on the personalized memory decay rhythm and the semantic relationship between each knowledge node, the energy propagation and interference between knowledge nodes are simulated to generate the rhythmic state of the knowledge resonance field. Based on learners’ dynamic rhythm map and knowledge resonance field rhythm state, rhythm matching diagnosis is performed through the resonance rhythm center to generate rhythm tuning intervention strategies. Implement rhythm tuning intervention strategies and collect learner feedback data after implementation; By updating learner dynamic rhythm maps and knowledge resonance field rhythm states using learner feedback data, a learning optimization loop is completed.
[0024] Specifically, a dual-loop dynamic information processing mechanism for learners was constructed and linked, namely, the dynamic coupling of learner cognition and emotion in the outer loop and the oscillation law of knowledge memory in the inner loop. By collecting and analyzing learners' physiological and behavioral signals in real time, multi-dimensional rhythmic sub-signals were abstracted, and the real-time phase synchronization between these rhythmic sub-signals was analyzed to characterize the learner's unique coupling pattern of cognitive load and emotional state, forming a learner dynamic rhythm map that dynamically evolves with task progress. At the same time, each knowledge point in the knowledge map is regarded as an oscillating unit with memory energy and inherent decay characteristics. By using the global state information reflected in the learner dynamic rhythm map as adjustment parameters, the inherent memory decay rhythm of knowledge points is dynamically and personalizedly corrected, thereby generating a more accurate personalized memory decay rhythm. This personalized decay rhythm guides the simulation of energy propagation and interference between knowledge points, and depicts the activity and phase of each knowledge point in the knowledge network in real time. The mutual influence relationship forms a knowledge resonance field rhythm state. Through the resonance rhythm center, the learner's dynamic rhythm map is continuously matched and diagnosed with the knowledge resonance field rhythm state to identify the most suitable timing, method and content for intervention, and generate corresponding rhythm tuning intervention strategies. Instead of treating cognition, emotion and memory as independent modules, they are abstracted into mutually influential rhythms, and learning is optimized by simulating their resonance and tuning. It should be noted that the cognitive style of this invention does not refer only to static personality traits in traditional psychology (such as field-independent / field-dependent), but to a real-time cognitive processing mode based on dynamic physiological rhythm characteristics. It refers to the specific combination pattern of cognitive load and emotional arousal (e.g., high load low anxiety mode, or low load high boredom mode) exhibited by the learner's EEG rhythm, heart rate variability and eye movement characteristics within a specific time window. By capturing this dynamic style, the real-time prediction of memory half-life can be achieved.
[0025] This enhances the accuracy and adaptability of personalized learning. By deeply coupling learners' cognitive and emotional dynamics with the memory decay patterns of knowledge, it achieves a shift from traditional post-event response to proactive intervention. It can observe learners' subtle psychological fluctuations in different situations in real time, ensuring that knowledge points that need to be consolidated or expanded are recommended when learners are in the most suitable cognitive and emotional state. This improves the effective encoding efficiency of learning content. The method no longer focuses solely on the correctness of knowledge points, but intervenes more deeply in the learner's internal physiological and psychological mechanisms during the learning process. This helps prevent learning fatigue, stimulates intrinsic learning motivation, and makes the overall learning effect far exceed the simple superposition of single technical means, providing a breakthrough optimization path for personalized learning.
[0026] Optionally, performing rhythm coupling analysis on real-time physiological time-series signals and real-time behavioral time-series signals to generate a learner's dynamic rhythm map includes: extracting at least one physiological rhythm sub-signal from the real-time physiological time-series signal; extracting at least one behavioral rhythm sub-signal from the real-time behavioral time-series signal; calculating the real-time phase synchronization between different physiological rhythm sub-signals and between physiological rhythm sub-signals and behavioral rhythm sub-signals; constructing dynamic connection weights based on the real-time phase synchronization, using each rhythm sub-signal as a node, and generating the learner's dynamic rhythm map using the dynamic connection weights.
[0027] Specifically, to identify the core "rhythmic" components representing the learner's internal state fluctuations from the raw, multimodal, continuous input data stream, physiological and emotional rhythm extraction is performed. Real-time physiological temporal signals include one or more combinations of EEG signals, heart rate variability (HRV) signals, skin conductance (GSR) signals, and facial expression signals. This supports the construction of a dimensionality-reduced dynamic rhythm map using only real-time behavioral temporal signals (such as eye movement trajectories and interaction logs) even when physiological sensor data is missing, ensuring usability in non-wearable device environments. This process uses specific digital signal processing algorithms to identify periodic or quasi-periodic patterns from the raw waveform data of real-time acquired EEG, HRV, and GSR. For example, for EEG... For EG data, bandpass filtering and power spectral density analysis were used to extract the continuous temporal variation curves of the relative power of the Alpha band and the Theta / Alpha ratio, representing fluctuations in attention, and these were used as a physiological rhythm sub-signal. For HRV data, Fourier transform analysis was used to analyze the heartbeat interval sequence to obtain the dynamic changes of low-frequency (LF) and high-frequency (HF) power components within a specific time window, and the fluctuation of the LF / HF ratio was used as a physiological rhythm sub-signal reflecting the balance of the sympathetic / parasympathetic nervous system. For GSR data, wavelet analysis and other methods were used to identify the slowly changing baseline fluctuations in skin conductance and the rapidly changing skin conductance response, and these were synthesized into a physiological rhythm sub-signal reflecting fluctuations in arousal or stress levels. These extracted physiological rhythm sub-signals are essentially continuous time series reflecting the fluctuations in the learner's internal psychological state.
[0028] Next, to capture the explicit rhythmic characteristics of learners' interactions with learning content, behavioral interaction rhythm extraction is performed. This process extracts behavioral rhythmic sub-signals representing fluctuations in cognitive input or information processing speed from real-time collected eye movement trajectory data (such as saccades, gaze duration, and pupil diameter change rate) and interaction log data (such as mouse click frequency and keyboard input intervals) through statistical analysis and time series pattern recognition. For example, the continuous fluctuations in pupil diameter within a specific time window can serve as a behavioral rhythmic sub-signal reflecting cognitive load or information reception rate; the dynamic change in the ratio of eye movement trajectory scanning rate to gaze duration per unit time can serve as a behavioral rhythmic sub-signal reflecting the rhythm of visual attention allocation. These behavioral rhythmic sub-signals are quantitative descriptions of periodic or trend changes in learners' external behavioral patterns.
[0029] Furthermore, to quantify the mutual influence and synergistic relationship between different intrinsic physiological and overt behavioral rhythmic components, real-time phase synchronization calculations are performed. These calculations employ time-series analysis techniques, such as sliding window cross-correlation or the Phase Locking Value (PLV), to assess the consistency or synchronicity of the fluctuation patterns of any two extracted physiological or behavioral rhythmic sub-signals within a specific time window. For any two rhythmic sub-signal sequences P1 and P2, they are first converted into analytic signals using a Hilbert transform H to obtain their respective instantaneous phase angles. and Then, within the preset sliding time window... Within, calculate its phase difference. Real-time phase synchronization C is defined as:
[0030] in, The value is an imaginary unit, and N is the number of sampling points within the sliding window. The calculated result C is a value between 0 and 1. A higher value indicates stronger phase synchronization between the two rhythmic sub-signals within that time window, suggesting stronger coordinated or consistent fluctuations between them in the current learner state. This real-time phase synchronization value C changes with the time window. The dynamic updates of the sliding rhythm reflect the real-time evolution of relationships between rhythms. To map these real-time synchronization quantification results into a structured network model, dynamic connection weights are constructed. This process uses the previously calculated real-time phase synchronization C as the weight of the connection edges in graph theory. Each physiological rhythm sub-signal and behavioral rhythm sub-signal is treated as an independent node, representing different functional modules or state dimensions in the learner's cognitive-emotional system. The weight W of the connection edges between nodes is real-time equivalent to the corresponding real-time phase synchronization C. For example, if the real-time phase synchronization C between the "pupil diameter change rate rhythm" and the "Alpha band relative power rhythm" is high, then the connection weight W between these two nodes in the constructed graph will also be high, indicating that the two are currently working closely together.
[0031] By aggregating all nodes and their dynamic connection weights, a learner dynamic rhythm graph is generated. The learner dynamic rhythm graph G is represented in a computable graph structure as follows:
[0032] V represents the set of rhythm nodes, each node corresponding to a physiological or behavioral rhythm sub-signal; E represents the set of connections between nodes; W represents the connection weight matrix. T represents the node phase state vector; T represents the time index sequence. The graph structure is dynamically updated with a sliding window over time, allowing for continuous calculation and state prediction of the rhythm network structure within different time periods. This is a weighted graph model with rhythm sub-signals as nodes and real-time phase synchronization C as dynamic connection weights. This graph reflects the complex dynamic coupling relationships of the learner's current cognitive load, emotional state, and attention in real time. Its topology and connection weights are continuously updated over time. For example, if the connection weight between the "emotional tension rhythm" and the "cognitive load rhythm" suddenly increases, it means that the learner is in a state of tension accompanied by high load. If the connection weight between the "focus rhythm" and other rhythms is at a medium to high level, it reflects that the learner is engaged in effective and calm learning. This dynamic rhythm graph serves as the core input for the next stage of decision-making.
[0033] Optionally, based on the global rhythm stability index in the learner's dynamic rhythm map, dynamically adjusting the intrinsic memory decay rhythm to generate personalized memory decay rhythms for each knowledge node includes: analyzing the phase synchronization patterns between nodes in the learner's dynamic rhythm map, identifying pressure coupling patterns or disordered coupling patterns; calculating the global rhythm stability index based on the identified coupling patterns; and injecting accelerated decay factors or decay inhibition factors into the intrinsic memory decay rhythms of all knowledge nodes according to the global rhythm stability index to generate personalized memory decay rhythms.
[0034] Specifically, to quantify the overall robustness and internal rhythmic harmony of learners' current cognitive and emotional states, coupling pattern analysis and identification will be performed. This process first continuously receives and parses the learner's dynamic rhythmic map generated in the previous step. This map reflects the real-time phase synchronization between the nodes of each rhythmic sub-signal. Graph theory algorithms, such as community detection or cluster analysis, combined with preset pattern recognition rules, are used to scan the topological structure and connection weight distribution of the map. For example, when nodes related to "emotional tension rhythm" (such as the decrease in the relative power rhythm of the Alpha band and the rapid fluctuation of the GSR rhythm) and nodes related to "cognitive load rhythm" (such as the continuous increase in the pupil diameter change rate rhythm) show a phase synchronization that is consistently higher than the preset threshold K1 and is maintained for more than the preset duration T1 seconds, it is identified as a "stress coupling pattern". Conversely, if the phase synchronization C between most rhythmic sub-signals in the map is generally lower than the preset threshold K2, or there is no obvious dominant synchronization pattern, it is identified as a "disordered coupling pattern", indicating that the learner's internal rhythm lacks consistency and order. The results of these pattern recognitions serve as a basis for judging the overall stability of the learner's state. Based on this, a global rhythm stability index is calculated to provide a unified and quantifiable indicator of the learner's current rhythmic state. This index is a comprehensive measure of the current "orderliness" or "harmony" of the learner's dynamic rhythmic map, and is quantitatively calculated based on the coupling patterns identified in the previous step and the overall connection weight distribution of the map.
[0035] The global rhythm stability index S can be calculated by representing the degree of disorder through the reciprocal of the phase synchronicity C among all nodes in the weighted average spectrum, or by mapping it through a specific neural network model. Where P is a binary variable indicating the existence of "pressure coupling mode" (P=1 if pressure mode is detected, otherwise P=0); D is a binary variable indicating the existence of "disorder coupling mode" (D=1 if disorder mode is detected, otherwise D=0); Avg(C) is the average phase synchronization among all valid node pairs in the current graph; For example, weighting coefficients calibrated based on experimental data. The value range is [0.3, 0.6]. The value range is [0.2, 0.4]. The value range is [0.1, 0.3]. This index S is a floating-point number between 0 and 1. The closer the value is to 1, the higher the learner's global rhythm stability and the more harmonious and orderly their internal cognitive-emotional state; the closer the value is to 0, the lower the stability, the more disordered the state, or the more pressure they are under. To achieve personalized and dynamic adaptation in memory prediction, personalized memory decay rate injection and adjustment are performed. This process uses the global rhythm stability index S obtained in the previous step as a core regulatory factor, acting on the inherent memory decay rhythm of each knowledge node in the preset knowledge graph. When the S value is lower than the preset stability threshold... (For example =0.6), indicating that the learner's current state is unstable or under pressure, will automatically inject an accelerated decay factor into the intrinsic memory decay rhythm of all knowledge nodes. Conversely, if the S value is higher than the preset promotion threshold... (For example =0.8), indicating that the learner is in a highly focused and positive emotional state, and a decay inhibition factor will be injected. The personalized memory decay rhythm of each knowledge node k Updated to: in, It is the inherent memory decay rhythm of knowledge node k, that is, the rate of memory forgetting of this knowledge point under ideal conditions; It is the basic acceleration factor, for example, the range of values [0.1, 0.3], which indicates that the more unstable the learner's state, the more obvious the forgetting acceleration effect; This is the basic inhibitory factor, for example, with a value range of [0.05, 0.15]. It indicates that the better the learner's state, the better the memory retention. Ensure that only when the stability index S is above the promotion threshold Only when the inhibitory factor begins to take effect does it come into play. Through this dynamic adjustment, the personalized memory decay rhythm of each knowledge node can reflect the influence of the learner's current psychological state on their memory retention ability in real time and in accordance with cognitive laws.
[0036] Optionally, based on the personalized memory decay rhythm and the semantic relationship between each knowledge node, the energy propagation and interference between knowledge nodes are simulated to generate the rhythm state of the knowledge resonance field. This includes: using the personalized memory decay rhythm of each knowledge node as the fundamental frequency and combining it with its current memory energy amplitude to generate a knowledge memory oscillation wave; propagating the knowledge memory oscillation wave in the knowledge graph according to the strength of the semantic relationship between knowledge nodes; when the knowledge memory oscillation waves of different knowledge nodes meet on the associated path, calculating the positive resonance enhancement effect or negative resonance interference effect based on the similarity of their fundamental frequencies, updating the current memory energy amplitude of each knowledge node, and comprehensively obtaining the rhythm state of the knowledge resonance field.
[0037] Specifically, refer to Figure 2 As shown, in order to transform the knowledge graph from a static conceptual network into a dynamic memory behavior model, the first step is to generate a knowledge memory oscillation wave. This process abstracts each knowledge node as an oscillator with periodic behavior, using the personalized memory decay rhythm of the knowledge node as its fundamental frequency. This fundamental frequency F defines the rate at which the memory energy of knowledge decays naturally without intervention. Combined with the current memory energy amplitude M of the knowledge point, that is, the learner's current mastery of the knowledge point, a "knowledge memory oscillation wave" simulating the fluctuation of the memory state of the knowledge point over time is generated.
[0038] Knowledge memory oscillation wave for each knowledge node k Defined as: in, It is the instantaneous value of the knowledge memory oscillation wave of knowledge node k at time t; It is the memory energy amplitude of the knowledge node at time t, with the initial value being the learner's mastery score of the knowledge point after learning it; It represents the personalized memory decay rhythm of knowledge node k, indicating the forgetting rate; t is a time variable. This is the initial phase when the knowledge point is activated or learned. In this embodiment, Based on the timestamp of the last time this knowledge point was learned or reviewed The calculation is performed using the following formula: in, The calculation uses a preset memory consolidation period constant (e.g., a value of 24 hours or a dynamic period set according to the Ebbinghaus curve) to ensure that knowledge points learned at similar time points have similar initial phases, thus making it more likely to produce resonance enhancement or interference in the same phase during subsequent propagation. This simulates the interaction between new and old knowledge within a specific time window, with the amplitude representing the degree of mastery and the frequency representing the rate of forgetting.
[0039] Next, to simulate the characteristic that knowledge does not exist in isolation in the learner's brain, but is interconnected and mutually influential, the propagation of knowledge memory oscillation waves is performed. This process refers to the semantic relationship strength between nodes in the preset knowledge graph to determine the propagation path and intensity of the oscillation waves. When a knowledge node is activated, its memory oscillation wave will be centered on that knowledge node and spread outwards towards the node with the strongest semantic relationship. Greater than the preset threshold The semantic relationship is propagated to adjacent knowledge nodes j, and the semantic relationship strength is determined. It can be the weights of connections between nodes in a predefined knowledge graph, reflecting the conceptual similarity, prerequisite relationships, or application relevance of knowledge points, as well as the intensity of propagation. Defined as: in, It is the memory energy amplitude of knowledge node k at the moment of propagation; This is the propagation coefficient, for example, ranging from [0.05, 0.2], used to control the proportion of energy propagation. The memory energy amplitude of each knowledge node j that receives the fluctuation will undergo an initial gain due to propagation.
[0040] To simulate the complex effects of mutual reinforcement or interference between knowledge points during the learning process, interference and energy updates of knowledge memory oscillation waves are performed. When knowledge memory oscillation waves from multiple different knowledge nodes converge at a specific knowledge point j through a shared semantic association path, they interfere. This is based on the fundamental frequency of these converging oscillation waves. Given its current phase difference, calculate the positive resonance enhancement effect or negative resonance interference effect generated at knowledge point j. For all oscillation waves converging at knowledge point j... ( (For the source knowledge node that converges to j), the memory energy amplitude of knowledge point j The update rules are as follows: in, In time The change in memory energy within the body is expressed as: Here It is the coupling coefficient, for example, with a value range of [0.01, 0.05], which adjusts the intensity of the interference effect; It is the source knowledge node The instantaneous value of the memory oscillation wave; It is the instantaneous value of the memory oscillation wave of knowledge point j; It is the difference between the personalized memory decay rhythm (fundamental frequency) of the two nodes. To prevent smoothing constants with zero denominators (e.g., 1e-5), the denominator ensures that the interference effect of nodes with closer frequencies is more significant, thus avoiding the problem of subtracting different units; and These are the source knowledge nodes. And the instantaneous phase of knowledge point j at time t; It is a knowledge point The strength of the semantic relationship between knowledge point j and knowledge point j. When When the value approaches 1 (in phase), positive resonance occurs, and memory energy is enhanced; when... When it is negative, It becomes negative, directly from the total energy. Subtraction from the middle achieves the mathematical effect of inhibiting or forgetting. Through this mechanism, the current memory energy amplitude and oscillation state of all knowledge nodes are comprehensively updated, forming a real-time evolving knowledge resonance field rhythm state, accurately reflecting the dynamic changes and mutual influences in the learner's knowledge network. Optionally, based on the learner's dynamic rhythm map and the knowledge resonance field rhythm state, rhythm matching diagnosis is performed through the resonance rhythm center to generate rhythm tuning intervention strategies, including: identifying a set of key knowledge nodes with memory energy amplitudes below a threshold and in a rapidly decaying phase from the knowledge resonance field rhythm state; diagnosing the learner's current cognitive load rhythm pattern and emotional rhythm pattern from the learner's dynamic rhythm map; calculating the matching degree between the inherent knowledge rhythm characteristics of the key knowledge node set and the learner's current cognitive load rhythm pattern and emotional rhythm pattern in the resonance rhythm center; and generating a rhythm tuning intervention strategy that includes a pre-learning rhythm guidance task and a main learning task, wherein the pre-learning rhythm guidance task is used to adjust the learner's emotional rhythm pattern to a state that matches the rhythm required by the main learning task.
[0041] Specifically, to identify opportunities for precise intervention on key knowledge points that learners are about to forget, the system first performs key knowledge node set identification. This process continuously monitors the rhythmic state of the knowledge resonance field updated in the previous step. For each knowledge node in the knowledge resonance field, its current memory energy amplitude M and personalized memory decay rhythm F are evaluated. If it is found that the memory energy amplitude M of a certain knowledge node has dropped to the preset forgetting threshold, the system will take action. (For example =0.4, representing a mastery level below 40%, and its personalized memory decay rhythm F predicts its future time window. (For example If the knowledge node shows significant decay within 15 minutes, it is marked as a potential key forgetting point. All knowledge nodes meeting this condition, along with their strongly associated nodes (semantic relationship strength R greater than a preset value), will be considered as key forgetting points. ,For example, =0.7), which are aggregated to form one or more sets of key knowledge nodes, which are the knowledge content carriers for subsequent interventions.
[0042] Next, to understand the learner's current psychological readiness, a cognitive load and emotional pattern diagnosis is performed. This process extracts and analyzes two core pieces of information from the continuously updated learner dynamic rhythm map from the previous step: the learner's cognitive load rhythm pattern and emotional rhythm pattern. The cognitive load rhythm pattern is analyzed by examining rhythmic sub-signals related to focus and information processing, such as the fluctuation pattern of the Alpha / Theta band power ratio and the persistence of the pupil diameter change rate rhythm. The emotional rhythm pattern is analyzed by examining rhythmic sub-signals related to stress, excitement, and pleasure, such as the LF / HF ratio rhythm in HRV, the GSR rhythm, and the fluctuation characteristics of facial expression rhythm. Based on the combination of these rhythmic sub-signals, the diagnostic algorithm outputs discrete pattern labels such as "high cognitive load with anxiety," "low cognitive load with fatigue," or "moderate cognitive load with calmness," along with the confidence level for each pattern. To determine the optimal timing and method of intervention, a rhythm matching degree calculation was performed using the resonant rhythm center. This calculation aimed to assess the best window for intervention on the key knowledge node set under the current learner state. The resonant rhythm center assessed the matching degree between the learner's cognitive load rhythmic pattern and emotional rhythmic pattern diagnosed in the previous step and the personalized memory decay rhythm (as its inherent "knowledge rhythmic feature") of the identified key knowledge node set. The resonant rhythm center calculates its match degree (Match) with the current learner state, and the calculation formula is as follows: in, Represents the current learner's state, determined by their cognitive load rhythm pattern. and emotional rhythm patterns composition; It is a collection of key knowledge nodes The inherent knowledge rhythm characteristics can be represented by the average or dominant frequency of the personalized memory decay rhythms across all nodes; it is and Pre-set emotional expectations related to the difficulty and complexity of the learning content; It is a function that assesses the fit between knowledge rhythm and learner cognitive load. For example, for high-frequency (fast decay) knowledge, learners have a high fit when their cognitive load is low. It is a function that assesses the degree of match between the emotions required for knowledge learning and the learner's current emotions. For example, knowledge that requires deep thinking has a high degree of match when the learner is in a calm emotional state. and These are weighting coefficients, for example, =0.6, =0.4, thus identifying the set of knowledge points most suitable for intervention in the current learner state, avoiding blind intervention. Based on the rhythm matching degree calculation results and the preset intervention strategy library, a rhythm tuning intervention strategy is generated. This strategy includes a pre-learning rhythm guidance task and a main learning task, aiming to proactively tune the learner's state to optimize subsequent learning effects. For example, if the diagnosis result is "the set of key knowledge nodes requires high-focus learning, but the learner's current mood is slightly anxious", the strategy will first select "relaxation meditation guidance" or "soothing music relaxation" as the pre-learning rhythm guidance task. The duration of this task is typically between 30 seconds and 3 minutes, designed to help learners shift their emotional rhythm pattern towards a calmer and more focused state. Once continuous monitoring confirms that the learner's emotional rhythm pattern has adjusted in the expected direction (e.g., the GSR rhythm tends to stabilize and the HF component of HRV increases), the main learning task is triggered. This task can be an interactive exercise, a scenario-based application problem, or a high-quality explanation provided by generative artificial intelligence, targeting a set of key knowledge nodes. Its content, format, and difficulty are strictly matched to the inherent knowledge rhythm characteristics of the set of key knowledge nodes and the learner's current tuned cognitive load rhythm pattern.
[0043] Optionally, implementing rhythm tuning intervention strategies and collecting learner feedback data after implementation includes: presenting learners with a pre-learning rhythm guidance task and monitoring learners' rhythm changes in real time; presenting learners with the main learning task when the learners' emotional rhythm pattern reaches the expected matching state; and recording learners' interaction data and subsequent physiological time-series signals during the completion of the main learning task as learner feedback data.
[0044] Specifically, to proactively guide learners into a more suitable psychological state for subsequent learning, a pre-learning rhythm guidance task is first presented. This process instantiates the "pre-learning rhythm guidance task" from the rhythm tuning intervention strategy and displays it to learners through the learning interface. This task can be a short interactive game, such as a concentration exercise requiring users to click synchronously according to a slow beat; or it can be a passive relaxation experience, such as playing a video of natural scenery with breathing guidance prompts. The core objective of this task is to tune specific emotional rhythm patterns. While the task is presented, the background data collection module collects data at high frequency. (For example, EEG sampling rate 256Hz, eye-tracking sampling rate 120Hz) Real-time monitoring of learners' physiological and behavioral temporal signals is conducted. These data streams are immediately input into the learner's rhythm analysis module. To ensure the guided task achieves the expected tuning effect, real-time rhythm change monitoring and evaluation are performed. During the learner's execution of the pre-guided rhythm task, a dynamic rhythm map of the learner is continuously constructed, and their current cognitive load rhythm pattern and emotional rhythm pattern are compared in real time with the target rhythm pattern of the task. For example, if the goal is to reduce anxiety, rhythmic sub-signals related to GSR and HRV will be monitored in detail. The tuning effect is evaluated by calculating the distance D between the current emotional rhythm pattern and the target emotional rhythm pattern. in This is the currently monitored pattern of emotional rhythms. The target emotional rhythm pattern is preset by the guided task. V() is a function that maps the rhythm pattern to a predefined multi-dimensional feature space. Its feature dimensions may include quantitative indicators such as rhythm frequency, amplitude stability, and multi-rhythm synchronization. This represents the Euclidean distance between two vectors. When this distance D is less than a preset convergence threshold... Or the task execution time exceeds the preset maximum duration. (For example =180 seconds), the tuning is completed. This process ensures that subsequent learning tasks are carried out in a verified and reasonable state. Once the learner's emotional rhythm pattern has reached the expected matching state, the presentation of the main learning task is triggered and executed. This task is the core content of the rhythm tuning intervention strategy, which directly targets the previously identified set of key knowledge nodes. Based on the inherent knowledge rhythm characteristics of the set and the learner's cognitive load rhythm pattern after tuning, the most suitable learning materials and interaction methods are dynamically selected or generated. For example, for a set of knowledge points that require deep logical thinking and the learner is in a calm and focused state, a complex project-based problem-solving task may be presented; while for a set of knowledge points that require rapid memorization, a card-style memorization exercise with rhythmic prompts may be presented. To create a closed-loop feedback loop, learner feedback data is continuously recorded throughout the entire learning task and for a period after its completion. This includes not only recording learners' direct responses, such as correctness of answers, response time, and submitted code or text—these are explicit interactive data—but also continuously collecting physiological and behavioral temporal signals (such as EEG and heart rate) and behavioral temporal signals (such as eye movements and mouse click patterns) during the learning task. These data collectively constitute complete learner feedback data. This data is timestamped and stored, serving as crucial input for subsequent model self-optimization and updates, reflecting the comprehensive effects of the intervention strategy at both cognitive and physiological levels.
[0045] Optionally, learner feedback data is used to update the learner dynamic rhythm map and the rhythm state of the knowledge resonance field, including: using the interaction data in the learner feedback data to update the current memory energy amplitude of the nodes in the key knowledge node set; analyzing the physiological time-series signals in the learner feedback data to obtain new rhythm sub-signals and update the dynamic connection weights in the learner dynamic rhythm map; and recalculating the positive or negative resonance effect between the key knowledge node set and its associated nodes based on the updated current memory energy amplitude, thereby updating the rhythm state of the knowledge resonance field.
[0046] Specifically, to directly assess the impact of intervention on knowledge mastery and revise the memory model, the memory energy amplitude of knowledge nodes is first updated. This process analyzes the interaction data obtained from the main learning task, especially the learners' responses to the set of key knowledge nodes. For each key knowledge node k that was intervened, its memory energy amplitude is... The updates will be adjusted based on the responses: If a learner answers a question related to knowledge node k correctly, their memory energy amplitude will be [value missing]. Increase: in The memory energy magnitude of the updated knowledge node k. The memory energy amplitude of knowledge node k before the update. It is the positive gain coefficient, for example, the value range [0.1, 0.3], which represents the memory consolidation effect of correct answers; This is the difficulty coefficient of the knowledge node k corresponding to the question, for example, the value range is [0.5, 1.0]. Answering difficult questions correctly has a more significant reinforcement effect. If the learner answers the question related to knowledge node k incorrectly, then their memory energy amplitude will decrease. Reduced, and will be subject to additional penalties: in, This is a penalty coefficient, for example, ranging from [0.05, 0.2], indicating that incorrect answers may lead to knowledge confusion or accelerated forgetting. This update ensures that the current memory energy amplitude of each knowledge node in the knowledge resonance field accurately reflects the learner's latest mastery after the intervention. Simultaneously, to reflect the impact of the intervention on the learner's internal state, the learner's dynamic rhythm map is updated. This process analyzes the new physiological and behavioral time-series signals collected after the intervention, which directly reflect changes in the learner's cognitive load and emotional state after the main learning task. The latest physiological and behavioral rhythm sub-signals are extracted from these new data streams, and their real-time phase synchronization C is recalculated. Based on these latest C values, all connection weights in the learner's dynamic rhythm map are reconstructed. This process is similar to previous rhythm coupling analysis and map construction methods. For example, if the learner is in a "stress coupling mode" before the intervention, but after the intervention, the phase synchronization between the learner's HRV rhythm and EEG focus rhythm increases, and the GSR rhythm fluctuation weakens, the new map will show a more "calm and focused" topology and connection weights, reflecting that the intervention has successfully improved the learner's internal state.
[0047] To update the interrelationships between knowledge points, the rhythmic state of the knowledge resonance field is recalculated and updated. After updating the memory energy amplitude M of individual knowledge nodes, these new amplitudes M immediately affect the generation of knowledge memory oscillation waves in the knowledge resonance field. The updated memory energy amplitudes of all knowledge nodes are then used. and their respective personalized memory decay rhythms. Regenerate the knowledge memory oscillation wave of each knowledge node Then, based on the semantic relationship strength between nodes in the preset knowledge graph, and considering whether certain knowledge points were particularly reinforced in the main learning task (for example, the negative resonance weight of two confused knowledge points reinforced through comparison and differentiation questions may be permanently reduced), the oscillation wave propagation and interference effects between all knowledge nodes are recalculated. This step updates the positive resonance enhancement effect and negative resonance interference effect between all knowledge nodes. Through this process, the rhythm state of the knowledge resonance field is comprehensively refreshed, reflecting not only the learner's latest mastery of individual knowledge points, but also the overall activation pattern and interaction force in the knowledge network, providing the most real-time and accurate basis for subsequent intervention decisions. Optionally, after updating the rhythm state of the knowledge resonance field, the following steps are also included: generating a rhythm tuning effect evaluation based on the comparison of the rhythm state of the knowledge resonance field before and after the update, and the comparison of the learner's dynamic rhythm graph before and after implementing the intervention strategy; using the rhythm tuning effect evaluation as a reward signal, inputting it into the reinforcement learning model, and optimizing the generation parameters of the rhythm matching diagnostic rules and rhythm tuning intervention strategies in the resonance rhythm center.
[0048] Specifically, to objectively quantify the comprehensive benefits of a learning intervention, an evaluation of rhythmic tuning effects is conducted. This process simultaneously considers changes in learners' dynamic rhythmic patterns and the rhythmic state of the knowledge resonance field before and after the intervention. First, the learners' dynamic rhythmic patterns are compared before and after the intervention. For example, it quantifies whether the learners' emotional rhythmic patterns have successfully shifted from an anxious mode to a calm mode, and whether the cognitive load rhythmic patterns have reached the target optimization range. Simultaneously, it assesses whether the memory energy amplitude M of the key knowledge node set in the knowledge resonance field has significantly increased, whether negative resonance interference within it has decreased, or whether positive resonance between key related knowledge points has strengthened. For example, it calculates whether the global rhythmic stability index S in the learners' dynamic rhythmic patterns has increased by at least the preset value after the intervention. (For example =0.1), and the average memory energy amplitude of the key knowledge node set. Has the minimum preset value been increased? (For example =0.15), the combination of these multi-dimensional indicators forms a comprehensive quantitative assessment of the intervention's effectiveness. Based on this comprehensive assessment, in order to continuously optimize its decision-making ability and intervention effect, a reward signal R is generated and output. This reward signal directly quantifies the degree of success of this intervention. The formula for calculating the reward signal R is: in, It is the change in the learner's global rhythm stability index S before and after the intervention, representing the improvement in emotional / cognitive state; It is the average memory energy amplitude of the set of key knowledge nodes before and after the intervention. The change in indicates the improvement in knowledge acquisition; This is the total time spent on this intervention, used to penalize interventions that are too long or inefficient; It is a preset time penalty coefficient, for example, the value range is [0.001, 0.005]; and These are weighting coefficients; This represents the average difficulty coefficient of this intervention task (normalized to between 0 and 1). The introduction of a difficulty factor, with a difficulty reward weight (e.g., 0.3), aims to prevent attempts to shorten time in an attempt to achieve higher scores. Or simply improve accuracy (impact) Instead, it tends to push overly simple tasks, guiding users to find the optimal balance between 'challenge' and 'completion efficiency'. This reward signal R can be positive (successful intervention) or negative (failed intervention or ineffective intervention), serving as external feedback to drive iterative optimization.
[0049] The reward signal R, along with its corresponding rhythm matching diagnostic rule selected for this intervention (e.g., "if diagnosed as mode A, then execute strategy B") and parameters used to generate the intervention content (e.g., "the number and difficulty of generating interesting cases"), are input into a pre-defined reinforcement learning model. Through continuous trial and error and learning, this reinforcement learning model automatically adjusts the decision-making logic and parameter configuration of the resonant rhythm center based on the obtained reward signal R. For example, if a rhythm matching diagnostic rule repeatedly results in low or even negative rewards, the reinforcement learning model will reduce the probability of that rule being selected or modify its triggering conditions. Conversely, for rules and parameter combinations that consistently generate high rewards, the model will strengthen their weights, making them preferred for future use. This process ensures that the resonant rhythm center can continuously "learn" and adapt to the personalized needs of different learners, making the overall intervention effect increasingly intelligent and precise, thereby achieving continuous evolution and optimization.
[0050] Optionally, the intrinsic memory decay rhythm of each knowledge node in the knowledge graph can be obtained, including: obtaining the static attributes of each knowledge node in the knowledge graph, including knowledge type, abstraction level, and historical mastery difficulty coefficient; and calculating the intrinsic memory decay rhythm of each knowledge node based on the static attributes through a rhythm mapping model.
[0051] Specifically, to establish a basic model of the initial memory characteristics of knowledge nodes, the static attributes of knowledge nodes are acquired. This process reads the static attributes attached to each knowledge node from the pre-constructed knowledge graph. Static attributes refer to non-dynamic data that are associated with the inherent characteristics of the knowledge content itself. For example, for the concept of "superposition state in quantum mechanics", its knowledge type can be labeled as "conceptual understanding type"; its level of abstraction can be labeled as "high level of abstraction"; its historical mastery difficulty coefficient can be statistically analyzed based on the learning records of a large number of learners or manually labeled by experts, for example, 0.8 (out of 1.0). These static attributes objectively describe the inherent difficulty or confusion of knowledge points, laying the foundation for predicting their intrinsic memory decay rhythm. To transform these static descriptions into rhythmic parameters that can be used for dynamic simulation, an intrinsic memory decay rhythm is calculated using a rhythmic mapping model. This model is a pre-trained mapping function that takes multiple static attributes of knowledge nodes as input and outputs a quantified intrinsic memory decay rhythm. The pre-trained mapping function f is constructed based on a large-scale historical learning dataset, collecting historical learning logs from tens of thousands of learners. Static attributes (type, abstraction, difficulty) of knowledge points are extracted as feature vectors X, and the learner's backtesting forgetting rate at different time intervals is fitted as a decay parameter Y. The mapping relationship is obtained through training with multiple linear regression or backpropagation neural networks, enabling the model to predict the intrinsic memory decay rhythm based on the static attributes of new knowledge points. It predicts the inherent decay rhythm of the population average. This intrinsic memory decay rhythm, in which the intrinsic memory decay rhythm With memory half-life The following relationship must be satisfied: The computable prediction of the memory half-life of a knowledge node through this mathematical relationship essentially represents the inherent rate of natural decay of memory energy for that knowledge point without the influence of learner-specific factors. For example, highly abstract and difficult concepts typically have a faster decay rate (faster forgetting), while simple and concrete knowledge points may have a slower decay rate. For knowledge node k, its inherent memory decay rate... It can be calculated using a linear or nonlinear mapping function f: in It is the knowledge type of knowledge node k, which is usually converted into numerical features through one-hot encoding or word embedding; It represents the level of abstraction of knowledge node k, usually quantified as a value between 0 and 1; This is the historical mastery difficulty coefficient of knowledge node k, also quantified as a value between 0 and 1. The function f can be a multiple linear regression model or a small neural network. For example, when... It is a "conceptual understanding type". It is 0.9. When it is 0.8, the calculated It may be 0.15 times / day (i.e., memory energy decays by 15% per day). This inherent memory decay rhythm serves as the fundamental frequency or decay rate of each knowledge point in its memory model, providing a basic parameter for subsequent dynamic adjustments considering learner-specific factors.
[0052] This invention also provides a learning optimization system based on cognitive style fusion memory half-life prediction, used to implement the above-mentioned learning optimization method. The system includes: The data acquisition module is used to acquire the learner's real-time physiological and behavioral time-series signals. The learner rhythm analysis module is used to perform rhythm coupling analysis on real-time physiological time-series signals and real-time behavioral time-series signals, and to generate and maintain learner dynamic rhythm maps. The knowledge rhythm management module is used to manage the knowledge graph, store and dynamically adjust the inherent memory decay rhythm and personalized memory decay rhythm of each knowledge node; The knowledge resonance field simulation module is used to simulate energy propagation and interference based on the personalized memory decay rhythm and inter-node relationships of knowledge nodes, and to generate and update the rhythm state of the knowledge resonance field. The resonant rhythm central decision-making module is used to perform rhythm matching diagnosis based on the learner's dynamic rhythm map and the rhythm state of the knowledge resonance field, and generate rhythm tuning intervention strategies. The intervention execution and feedback module is used to execute rhythm tuning intervention strategies and collect learner feedback data after execution. The feedback data is then sent to the learner rhythm analysis module and the knowledge resonance field simulation module to complete the closed-loop update.
[0053] In this embodiment, the technology of this invention is applied to the online education platform "ZhiXue.com" to provide personalized and adaptive subject learning services to 300 high school students. This platform focuses on students' cognitive load and emotional state when learning complex knowledge points to optimize memory efficiency; traditional systems only recommend content based on answer accuracy and learning time, making it difficult to understand the learner's internal state and effectively predict memory decay. This experiment uses the method of this invention to learn high school mathematics topics in an experimental group of 100 students, while another 100 students serve as a control group using traditional adaptive learning methods. The experiment lasts for 4 weeks to demonstrate the parameter calculation process and learning optimization effect of this technology.
[0054] The system collects learners' multimodal data in real time, including physiological signals such as electroencephalography (EEG), heart rate variability (HRV), and skin conductance (GSR) when completing math and physics exercises, as well as behavioral signals such as eye movement, mouse click frequency, and keyboard input intervals. These time-series signals are converted into multiple "rhythmic sub-signals" and matched with the attributes of knowledge points in the knowledge graph. Rhythmic coupling analysis is performed on the aforementioned physiological and behavioral signals every 10 seconds. For example, the continuous change curves of the relative power of the alpha band and the Theta / Alpha ratio are extracted from the EEG (256Hz sampling) as the "attention fluctuation rhythm." If the relative alpha power increases from 0.3 to 0.6 and the Theta / Alpha ratio decreases from 0.8 to 0.4 within a 30-second window, it can be determined that the attention rhythm is showing an upward trend. The dynamic changes in the LF / HF ratio in HRV (100Hz sampling) reflect the "sympathetic / parasympathetic balance rhythm." For example, if the LF / HF fluctuates between 2.0 and 3.5 in a sliding 5-minute window spectral analysis, it indicates moderate sympathetic activation; if the LF / HF rapidly rises above 4.5, it is identified as an enhanced stress rhythm. The baseline and rapid changes in skin conductance in GSR (50Hz sampling) constitute the "arousal / stress rhythm": for example, if the peak frequency of GSR increases from 1 to 5 times per minute within 30 seconds, and the baseline conductance continues to rise, it indicates a significant enhancement of the arousal / stress rhythm.
[0055] Regarding behavioral signals, pupil diameter fluctuations were extracted from eye movements (120Hz sampling) as the "cognitive load rhythm." For example, when solving complex problems, if the mean pupil diameter of students increases from 3mm to 4.5mm and the fluctuation variance increases by 20%, it can be considered an enhancement of the cognitive load rhythm. Interaction logs (mouse click frequency, keyboard input interval) reflect the "information processing speed rhythm." For instance, when encountering difficult problems, if the mouse click frequency decreases from 1.5 times / second to 0.3 times / second and the keyboard input interval increases from 2 seconds to 5 seconds, it indicates a slowdown in the information processing speed rhythm. A sliding method with a 10-second window and a 2-second step size was used to calculate the cross-correlation and phase lock value (PLV) between any two rhythmic sub-signals to evaluate real-time phase synchronization. For example, when a student is solving a problem, the PLV value of the pupil diameter change rate rhythm and the EEGAlpha fluctuation rhythm may rapidly increase from 0.3 to 0.75, indicating an enhancement of their coordinated fluctuation. The phase synchronization C was used as the connection weight W between the nodes of each rhythmic sub-signal to construct a dynamic rhythm map. In the example above, the connection weight between the "pupil rate of change rhythm" and the "Alpha relative power rhythm" is updated to 0.75. The learner's overall state is assessed based on the dynamic rhythm map, and the global rhythm stability index S is updated every minute. Specifically, coupling modes are first identified: if the phase synchronization between the GSR rhythm and the LF / HF rhythm remains above the threshold K1=0.75 for more than one minute, and the pupil rate of change rhythm exhibits high-amplitude fluctuations, the learner is considered to be in a pressure coupling mode (P=1); if the phase synchronization between most sub-signals is generally below the threshold K2=0.25 and there is no obvious dominant mode, it is identified as a disordered coupling mode (D=1). Then, the formula is used... Calculate the stability index S. For example, if a pressure coupling mode is identified within 3 minutes (P=1, D=0) and the average phase synchronization between all node pairs is at this time... =0.4, then S=1-(0.5+0.12)=0.38; if no pressure or disturbance is identified (P=0, D=0) and =0.7. Therefore, S = 1 - 0.06 = 0.94. Thus, the S value reflects the learner's overall rhythmic stability in real time. By using the stability index S, the learner's personalized memory decay rate for corresponding knowledge points can be adjusted. Set a stability threshold. =0.6, promoting threshold =0.8, the inherent decay rhythm of knowledge point k From prior knowledge (such as an abstract concept) =0.12 times / day). Let the base acceleration factor be... =0.2, basic inhibitory factor =0.1. Calculate the new decay rhythm based on the S value: For example, when S = 0.38 (< 0.6), This means that the decay rate of this knowledge point increases by 12.4%; when S=0.94 (>0.8), That is, the decay rate is slightly slower by 0.2%, so the same knowledge point has different personalized decay rhythms under different states.
[0056] It also constructs a knowledge resonance field, setting a memory energy amplitude for each knowledge point k. (For example, after learning) =0.8), personalized memory decay rhythm and propagation coefficient =0.1 This generates a memory oscillation wave. , Calculated based on the timestamp of the last review of this knowledge point, when the knowledge point is activated (e.g. When the value is increased to 0.9, its oscillation wave will propagate to the relevant knowledge nodes according to the propagation coefficient α, and the propagation intensity... ,in This represents the semantic correlation between knowledge points k and j (range 0~1). For example, if... =0.12 times / day, upon activation =0.9, correlation with knowledge point j =0.7, then =0.1*0.7*0.9=0.063, this value will be added to the memory energy of knowledge point j. Interference between oscillating waves is also considered: let the coupling coefficient be... =0.03. When the difference in oscillation frequencies between knowledge point A and knowledge point C is less than 0.02 (at which point the denominator in the formula is extremely small, resulting in a significant increase in the weight of the resonance term) and they are in phase, the oscillation of A can resonate positively with C, significantly enhancing the energy of C. For example, if the current energy of C... =0.6, and its energy can be renewed to 0.65 after positive resonance.
[0057] Simultaneously, it identifies key knowledge nodes and makes rhythm-matching decisions, scanning the knowledge resonance field every minute, and when a certain knowledge point... Drop to the forgetting threshold When the Theta / Alpha ratio is below 0.4 and is predicted to continue decreasing by more than 0.1 within 15 minutes, it is marked as a "critical forgetting point," and all nodes with a semantic relevance of ≥0.7 are included in this "critical knowledge node set." Simultaneously, the learner's current cognitive load and emotional pattern are analyzed: for example, if the Theta / Alpha rhythm is consistently above 1.0, pupil diameter is continuously enlarging, and GSR fluctuates rapidly, a diagnosis of "high cognitive load accompanied by anxiety" is made. For each critical knowledge node set... The matching degree Match is calculated based on the learning needs of the knowledge set and the current learner status. , For example, if a knowledge set has an average decay rate of 0.18 times / day (indicating high difficulty) and requires high concentration and low anxiety, while the learner is in a state of "high workload + anxiety," then let the cognitive fit be... =0.3, Emotional Match =0.2, with weight =0.6、 =0.4 The calculated Match = 0.6 * 0.3 + 0.4 * 0.2 = 0.26, which is very low. Since it is below the threshold of 0.4, a targeted rhythm tuning intervention strategy is generated: In this example, when a match of 0.26 is identified and the learner is anxious, a "3-minute soothing music relaxation" activity will be arranged as a pre-guided task. For this knowledge set (such as the topic of "derivative solving"), a context-based derivative application problem will be selected as the main learning task, allowing the student to learn in a relaxed and calm state. Rhythm tuning intervention will then be implemented, first presenting the aforementioned "3-minute soothing music" guided task, and monitoring signals such as EEG (256Hz), HRV (100Hz), and GSR (50Hz) in real time, with the goal of reducing the student's stress level. For example, the goal is to reduce the peak frequency of GSR from 5 times / minute to 2 times / minute, and the high-frequency component (HF) of HRV from 0.3... Increased to 0.5 This aims to calm emotional rhythms and predefines the emotional feature vector of the target calm state. (e.g., GSR component 0.1, HF component 0.8), and calculate the current state. The effectiveness is evaluated using the Euclidean distance from the target. Assuming the student's state after the guided task is GSR=0.3 and HF=0.6, then the distance... The distance was greater than the threshold of 0.25, so the guidance continued; after about 2 minutes and 30 seconds, the distance dropped below the threshold, and the tuning was considered complete. Then the main learning task "Context-based derivative application problems" was presented, and the students' answers were recorded (e.g., 2 out of 3 questions were answered correctly, and 15 minutes were taken) and EEG, HRV, GSR, eye movement and other data were collected.
[0058] Based on students' answers, update the memory capacity of relevant knowledge points: for example, if a student correctly answers a question of a certain difficulty level... The derivative problem with a value of 0.7 corresponds to the old memory energy of the knowledge point k. =0.4, let the correct answer be the gain coefficient. =0.2, then calculate =0.4+(1-0.4)*0.2*0.7=0.484, making The value increased from 0.4 to 0.484. Simultaneously, the dynamic rhythmic map was updated using new signals collected during guided and learning tasks: for example, after completing derivative problems, students' alpha power significantly increased, the LF / HF ratio of HRV decreased, and the GSR baseline stabilized. At this point, the PLV value between the recalculated "attention rhythm" and "sympathetic / parasympathetic balance rhythm" increased from 0.3 before intervention to 0.65, and the map connection weights were updated accordingly. The knowledge resonance field was also updated based on the updated values. Upon recalculation, the increased amplitude of knowledge points due to correct answers will further strengthen the positive resonance effect between them and their associated nodes.
[0059] To evaluate the effectiveness of a rhythm tuning intervention and generate reward signals, assume the learner's stability index before a certain intervention. =0.38 (stress mode), after intervention =0.82; Average memory energy amplitude of the key knowledge point set It increased from 0.35 to 0.48. =0.44, =0.13; Average difficulty coefficient of this intervention task (situational derivative problem) =0.7. Let the weight be... =0.4, =0.4, =0.3, total time spent =20 minutes, time penalty coefficient =0.002. Calculate the reward signal R: R = 0.4x0.44 + 0.4x0.13 + 0.3x0.7 - 0.02x20, R=0.176+0.052+0.21-0.04=0.398, A positive result indicates that the intervention received positive feedback. The reward signal R=0.398, along with the rhythm matching diagnostic and control strategy used (e.g., "high load + anxiety and matching degree <0.4, 3 minutes of soothing music + situational derivative question"), was input into the reinforcement learning model. The model, based on positive feedback, enhanced the selection probability of this strategy in similar situations and fine-tuned relevant parameters (e.g., music duration, question difficulty). Finally, the static attributes of the knowledge nodes were obtained to calculate their intrinsic memory decay rhythm. For example, knowledge point k "limited..." The definition is a concept comprehension question with an abstraction level of 0.9 and a historical difficulty level of 0.85. Knowledge point j, "the derivative of the sine function," is a formula application question with an abstraction level of 0.6 and a historical difficulty level of 0.65. A pre-trained linear mapping model is used. (Where IsConcept=1 represents a conceptual type) Calculations can be performed to obtain the inherent decay rate of knowledge point k. =0.05 + 0.08 * 0.9 + 0.04 * 0.85 + 0.03 = 0.186 times / day; The inherent decay rate of knowledge point j. =0.05 + 0.08 * 0.6 + 0.04 * 0.65 + 0 = 0.124 times / day. Through the complete technical process described above, refer to... Figure 3 As shown, this embodiment demonstrates the entire chain from raw data collection to optimized learning output. At the end of the 4-week experiment, the learning effects of the experimental group and the control group were significantly different: the average retention rate of key knowledge points in the experimental group was 78%, while that in the control group was 62% (an increase of approximately 25.8%); students' self-rated learning efficiency satisfaction increased from 3.5 to 4.2 (an increase of 20%); the average weekly time to complete learning tasks decreased from 145 minutes to 120 minutes (a decrease of 17.2%); and students' active participation increased from 60% to 85% (an increase of 41.7%). In a typical intervention, the dynamic changes in students' state were significant: before the intervention, the global stability index S was 0.38 (stress-coupled mode), which rose to 0.75 after the guided task (tending towards calm), and further increased to 0.82 (calm and focused) after the main learning. At the same time, the average memory energy of key knowledge points increased from 0.35 to 0.48. This indicates that the rhythmic tuning intervention effectively improved learners' cognitive load and emotional state, and strengthened knowledge consolidation. Multiple intervention records show that the strategy adaptively optimizes: for example, in scenario one, the initial S=0.38, =0.35 (“High workload + anxiety”), after adopting the strategy of 3 minutes of music + situational questions =0.44、 =0.13, Time taken 20 minutes, Reward R=0.245; Scenario 2 Initial S=0.52, =0.42 (“moderate workload + fatigue”), after adopting the interactive challenge + basic practice strategy =0.28、 =0.08, time taken 15 minutes, R=0.160; Scenario 3 initial S=0.90, =0.60 (“Calm and Focused”), no guidance needed, directly to higher-level problems. =0.05、 =0.06, time taken 10 minutes, R=0.080. These data demonstrate the significant improvement in learning effectiveness achieved by the method of this invention, and the effectiveness of continuously optimizing the intervention strategy through positive feedback. In summary, this embodiment verifies the engineering feasibility and significant advantages of this invention in personalized learning optimization.
[0060] It should be noted that the electrical connections between the various units described above do not necessarily represent direct or indirect connections. Any indirect connection method can be applied to the embodiments of the present invention as long as it achieves the purpose of the present invention. The above are merely exemplary embodiments of the present invention and should not be construed as limiting the scope of the present invention.
[0061] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.
Claims
1. A learning optimization method based on cognitive style fusion memory half-life prediction, characterized in that, include: Acquire learners' real-time physiological and behavioral time-series signals; Rhythm coupling analysis is performed on real-time physiological time-series signals and real-time behavioral time-series signals to generate learner dynamic rhythm maps; The step of performing rhythm coupling analysis on real-time physiological time-series signals and real-time behavioral time-series signals to generate a learner's dynamic rhythm map includes: extracting at least one physiological rhythm sub-signal from the real-time physiological time-series signals; extracting at least one behavioral rhythm sub-signal from the real-time behavioral time-series signals; calculating the real-time phase synchronization between different physiological rhythm sub-signals and between physiological rhythm sub-signals and behavioral rhythm sub-signals; constructing dynamic connection weights based on the real-time phase synchronization, using each rhythm sub-signal as a node, and generating the learner's dynamic rhythm map using the dynamic connection weights. Obtain the inherent memory decay rhythm of each knowledge node in the knowledge graph; The process of obtaining the intrinsic memory decay rhythm of each knowledge node in the knowledge graph includes: obtaining the static attributes of each knowledge node in the knowledge graph, wherein the static attributes include knowledge type, level of abstraction, and historical mastery difficulty coefficient; and calculating the intrinsic memory decay rhythm of each knowledge node based on the static attributes through a rhythm mapping model. Based on the global rhythm stability index in the learner's dynamic rhythm map, the intrinsic memory decay rhythm is dynamically adjusted to generate personalized memory decay rhythms for each knowledge node. The method of dynamically adjusting the intrinsic memory decay rhythm based on the global rhythm stability index in the learner's dynamic rhythm map to generate personalized memory decay rhythms for each knowledge node includes: analyzing the phase synchronization patterns between nodes in the learner's dynamic rhythm map, identifying pressure coupling patterns or disordered coupling patterns; calculating the global rhythm stability index based on the identified coupling patterns; and injecting acceleration decay factors or decay inhibition factors into the intrinsic memory decay rhythms of all knowledge nodes according to the global rhythm stability index to generate personalized memory decay rhythms. Based on the personalized memory decay rhythm and the semantic relationship between each knowledge node, the energy propagation and interference between knowledge nodes are simulated to generate the rhythmic state of the knowledge resonance field. The process of generating a knowledge resonance field rhythm state by simulating energy propagation and interference between knowledge nodes based on personalized memory decay rhythms and semantic relationships between knowledge nodes includes: using the personalized memory decay rhythm of each knowledge node as the fundamental frequency and combining it with its current memory energy amplitude to generate a knowledge memory oscillation wave; propagating the knowledge memory oscillation wave in the knowledge graph according to the strength of semantic relationships between knowledge nodes; when the knowledge memory oscillation waves of different knowledge nodes meet on the associated path, calculating the positive resonance enhancement effect or negative resonance interference effect based on the similarity of their fundamental frequencies, updating the current memory energy amplitude of each knowledge node, and comprehensively obtaining the knowledge resonance field rhythm state. Based on learners’ dynamic rhythm map and knowledge resonance field rhythm state, rhythm matching diagnosis is performed through the resonance rhythm center to generate rhythm tuning intervention strategies. Implement rhythm tuning intervention strategies and collect learner feedback data after implementation; By updating learner dynamic rhythm maps and knowledge resonance field rhythm states using learner feedback data, a learning optimization loop is completed.
2. The learning optimization method based on cognitive style fusion memory half-life prediction according to claim 1, characterized in that, The method of generating a rhythm tuning intervention strategy based on learner dynamic rhythm map and knowledge resonance field rhythm state, through rhythm matching diagnosis via resonance rhythm center, includes: identifying a set of key knowledge nodes whose memory energy amplitude is below a threshold and is in a rapidly decaying phase from the knowledge resonance field rhythm state; diagnosing the learner's current cognitive load rhythm pattern and emotional rhythm pattern from the learner dynamic rhythm map; calculating the matching degree between the inherent knowledge rhythm characteristics of the key knowledge node set and the learner's current cognitive load rhythm pattern and emotional rhythm pattern in the resonance rhythm center; and generating the rhythm tuning intervention strategy including a pre-learning rhythm guidance task and a main learning task, wherein the pre-learning rhythm guidance task is used to adjust the learner's emotional rhythm pattern to a state that matches the rhythm required by the main learning task.
3. The learning optimization method based on cognitive style fusion memory half-life prediction according to claim 2, characterized in that, The steps of implementing the rhythm tuning intervention strategy and collecting learner feedback data after implementation include: presenting the pre-implementation rhythm guidance task to the learner and monitoring the learner's rhythm changes in real time; when the learner's emotional rhythm pattern is detected to reach the expected matching state, presenting the main learning task to the learner; and recording the learner's interaction data and subsequent physiological time-series signals during the completion of the main learning task as the learner feedback data.
4. The learning optimization method based on cognitive style fusion memory half-life prediction according to claim 3, characterized in that, The steps of updating the learner dynamic rhythm map and the rhythm state of the knowledge resonance field using learner feedback data include: updating the current memory energy amplitude of nodes in the key knowledge node set using the interaction data in the learner feedback data; analyzing the physiological time-series signals in the learner feedback data to obtain new rhythm sub-signals and updating the dynamic connection weights in the learner dynamic rhythm map; and recalculating the positive or negative resonance effect between the key knowledge node set and its associated nodes based on the updated current memory energy amplitude, thereby updating the rhythm state of the knowledge resonance field.
5. The learning optimization method based on cognitive style fusion memory half-life prediction according to claim 4, characterized in that, After updating the knowledge resonance field rhythm state, the process also includes: generating a rhythm tuning effect evaluation based on the comparison of the knowledge resonance field rhythm state before and after the update, and the comparison of the learner's dynamic rhythm map before and after the implementation of the intervention strategy; using the rhythm tuning effect evaluation as a reward signal, inputting it into a reinforcement learning model to optimize the rhythm matching diagnosis rules in the resonance rhythm center and the generation parameters of the rhythm tuning intervention strategy.
Citation Information
Patent Citations
Adaptive learning optimization method and system based on big data analysis
CN119476624A
Online teaching optimization method and system based on emotion recognition
CN120355539A
Knowledge tracking model fusing dynamic edge weight and forgetting gating
CN120706470A
