Mobile terminal online learning optimization method and system

Abstract

The invention provides a mobile terminal online learning optimization method and system, and belongs to the technical field of mobile terminal learning platform optimization, and the method comprises the steps: 1, collecting the original behavior data of a terminal user, and outputting a learner state packet containing a real-time knowledge attention vector, a cognitive load index and an equipment efficiency score; 2, outputting a knowledge blind area mark set, and meanwhile, generating learning state description containing blind area marks and load levels; 3, generating a primary content optimization strategy based on the blind area mark set, carrying out strategy feasibility verification in combination with a load grade and an equipment efficiency score, and outputting a final executable strategy set; 4, executing the executable strategy set and controlling interface rendering and resource allocation; and 5, collecting eye movement track change data and system performance indexes, and outputting a feedback data stream used for updating the knowledge attention vector. The learning content and the user experience are optimized, the cognitive load is reduced, and the high efficiency and individuation of the learning process are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of optimizing mobile learning platforms, and particularly to a method and system for optimizing mobile online learning. Background Art

[0002] With the popularization of mobile learning, personalized learning and optimizing user experience have become research hotspots. Especially in online learning, how to adjust learning content and interfaces in real time according to learners' behavioral data to improve learning efficiency and effectiveness has become a key requirement for technological development.

[0003] Current online learning systems mainly rely on learners' operation history and simple behavioral data analysis, such as click times and completion progress, to optimize the learning process. However, most of these systems lack in-depth perception of learners' real-time status and cannot accurately capture multi-dimensional information such as learners' cognitive load, knowledge attention, and device efficiency. Therefore, dynamic and personalized learning content optimization cannot be achieved. Existing technical solutions cannot comprehensively consider learners' multi-dimensional real-time data for personalized learning optimization, resulting in content recommendations and resource allocations that cannot accurately match learners' cognitive states and device performances. As a result, learning efficiency may be limited, and the system's optimization strategies often cannot be adjusted in a timely manner according to learners' actual needs.

[0004] Therefore, the present invention provides a method and system for optimizing mobile online learning. Summary of the Invention

[0005] The present invention provides a method for optimizing mobile online learning, which is used to provide a personalized optimization strategy for mobile online learning by collecting multi-dimensional behavioral data of end users in real time, combining knowledge attention, cognitive load, and device efficiency. Compared with the prior art, this method realizes dynamic adjustment based on learners' states, improves learning efficiency by generating content optimization strategies, controlling interface rendering, and resource allocation. The closed-loop feedback mechanism further optimizes learning content and user experience, reduces cognitive load, and ensures the efficiency and personalization of the learning process.

[0006] According to a method for optimizing mobile online learning provided by the present invention, it includes: Step 1: Collect the original behavioral data of end users through a preset sensor array, where the original behavioral data includes: eye movement data, touch data, and environmental data, and output a learner state packet containing a real-time knowledge attention vector, a cognitive load index, and a device efficiency score; Step 2: Perform node-level matching between the knowledge attention vector and a pre-constructed knowledge graph, and output a knowledge blind spot marker set. At the same time, divide the load level according to the cognitive load index, and generate a learning state description containing the blind spot marker and the load level; Step 3: Generate a primary content optimization strategy based on the blind area tag set, verify the feasibility of the strategy by combining the load level and the device efficiency score, and output the final executable strategy set; Step 4: Execute the executable strategy set and control the interface rendering and resource allocation; Step 5: Collect the eye movement trajectory change data and system performance metrics after executing the final executable strategy set, and output the feedback data stream for updating the knowledge attention vector.

[0007] Preferably, collect the original behavior data of the end user through a preset sensor array, and output a learner status package including a real-time knowledge attention vector, a cognitive load index, and a device efficiency score, including: Collect the original behavior data of the end user through a preset sensor array; Determine the real-time knowledge attention vector based on the eye movement data, determine the cognitive load index based on the touch data, and determine the device efficiency score based on the environmental data, and then output the learner status package.

[0008] Preferably, determining the real-time knowledge attention vector based on the eye movement data includes: Determine the eye movement fixation points and the corresponding fixation durations based on the eye movement data; Match the eye movement fixation points with the regions of the interface display content, and generate a distribution vector reflecting the relative attention intensity of each content block of the interface display content in combination with the fixation duration corresponding to each eye movement fixation point; Determine the distribution vector reflecting the relative attention intensity of each content block of the interface display content as the real-time knowledge attention vector.

[0009] Preferably, determining the cognitive load index based on the touch data includes: Analyze the touch data to obtain the touch pressure volatility, touch trajectory data, and operation time data; Determine the cognitive load index based on the touch pressure volatility, touch trajectory data, and operation time data: Wherein, is the touch pressure volatility, is the preset user baseline pressure volatility, d is the actual touch trajectory length, is the ideal touch trajectory length, is the preset acceptable deviation threshold, is the total touch pause duration, is the total touch operation duration, the preset minimum detectable pause time threshold, is the preset reference pause ratio, is the individual operation style adaptation factor, and , is a preset extremely small positive number.

[0010] Preferably, perform node-level matching on the knowledge attention vector and the pre-constructed knowledge graph, output a knowledge blind spot marking set, and at the same time, divide the load level according to the cognitive load index, and generate a learning state description including the blind spot marking and the load level, including: Traverse the knowledge attention vector, and mark the knowledge points with all attention values lower than the dynamic threshold as the first type of nodes; Retrieve the associated knowledge points in the pre-constructed knowledge graph that have a direct cognitive dependence relationship with the first type of nodes, and mark them as the second type of nodes after screening by the cognitive dependence strength; Determine the blind spot repair priority based on the attention deviation degree of the first type of nodes, the cognitive dependence strength between the second type of nodes and the core blind spot, and the topological importance of the first type of nodes and the second type of nodes in the knowledge graph, and then output a knowledge blind spot marking set; Divide the user's real-time cognitive load level according to the numerical range of the cognitive load index; Integrate the knowledge blind spot marking set and the cognitive load level to generate a learning state description.

[0011] Preferably, generate a primary content optimization strategy based on the blind spot marking set, verify the strategy feasibility in combination with the load level and the device efficiency score, and output a final executable strategy set, including: Load the basic strategy framework from the predefined strategy template library according to the node type in the blind spot marking set, and then generate the corresponding content optimization strategy; Verify the strategy feasibility of the content optimization strategy based on the load level and the device efficiency score, and output a final executable strategy set.

[0012] Preferably, execute the executable strategy set and control the interface rendering and resource allocation, including: When executing the executable strategy set, adjust the rendering priority of the interface elements according to the content optimization strategy; obtain the environmental state prediction result based on the preset algorithm and environmental data; Allocate network resources based on the environmental state prediction result.

[0013] Preferably, collect the eye movement trajectory change data and system performance indicators after the execution of the final executable strategy set, and output a feedback data stream for updating the knowledge attention vector, including: Monitor the eye movement trajectory change data of the user for the optimized content after the strategy execution, and extract the visual attention transfer characteristics; Record the device resource consumption during the strategy execution; Extract the core examination points of the target nodes from the pre-constructed knowledge graph to generate micro-tests and obtain the micro-test results; Determine the difference in knowledge attention vectors before and after strategy execution based on visual attention transfer features, and combine the micro-test results and device resource consumption to determine the reasons for the change in the actual knowledge retention rate; Generate corresponding feedback data streams based on the reasons for the change in the actual knowledge retention rate.

[0014] A mobile online learning optimization system, comprising: A data collection module, configured to collect the original behavior data of the terminal user through a preset sensor array; A status analysis module, configured to perform node-level matching between the knowledge attention vector and a pre-constructed knowledge graph, output a knowledge blind area mark set, and at the same time, divide the load level according to the cognitive load index, and generate a learning status description including the blind area mark and the load level; A strategy generation module: generate a primary content optimization strategy based on the blind area mark set, perform strategy feasibility verification by combining the load level and the device efficiency score, and output a final executable strategy set; An execution control module, which executes the executable strategy set and controls the interface rendering and resource allocation feedback learning module; A feedback learning module, which collects the eye movement trajectory change data and system performance indicators after the execution of the final executable strategy set, and outputs a feedback data stream for updating the knowledge attention vector.

[0015] Compared with the prior art, the beneficial effects of the present application are as follows: By collecting multi-dimensional behavior data of terminal users in real time, combining knowledge attention, cognitive load, and device efficiency, personalized optimization strategies are provided for mobile online learning. Compared with the prior art, this method realizes dynamic adjustment based on the learner's state. By generating content optimization strategies, controlling interface rendering and resource allocation, the learning efficiency is improved. The closed-loop feedback mechanism further optimizes the learning content and user experience, reduces the cognitive load, and ensures the efficiency and personalization of the learning process. Description of the Drawings

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0017] Figure 1 It is a schematic flow chart of a mobile online learning optimization method provided by an embodiment of the present invention.

[0018] Figure 2 It is a schematic structural diagram of a mobile online learning optimization system provided by an embodiment of the present invention. Detailed implementation manners

[0019] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without creative efforts shall fall within the protection scope of the present invention. Embodiment 1

[0020] The embodiment of the present invention provides a method for optimizing mobile online learning, as Figure 1 shown, including: Step 1: Collect the original behavior data of the terminal user through a preset sensor array, where the original behavior data includes: eye movement data, touch data, and environmental data, and output a learner status packet including a real-time knowledge attention vector, a cognitive load index, and a device efficiency score; Step 2: Perform node-level matching between the knowledge attention vector and a pre-constructed knowledge graph, and output a knowledge blind area marking set. At the same time, divide the load level according to the cognitive load index, and generate a learning status description including the blind area marking and the load level; Step 3: Generate a primary content optimization strategy based on the blind area marking set, and perform strategy feasibility verification by combining the load level and the device efficiency score, and output a final executable strategy set; Step 4: Execute the executable strategy set and control interface rendering and resource allocation; Step 5: Collect the eye movement trajectory change data and system performance indicators after executing based on the final executable strategy set, and output a feedback data stream for updating the knowledge attention vector.

[0021] In this embodiment, the preset sensor array is a multi-modal sensor combination integrated in the mobile terminal for real-time collection of user learning behavior data, including: an eye tracker: tracking the fixation point coordinates and dwell time (reflecting the attention distribution); a touch screen sensor: recording the operation force and sliding trajectory (evaluating the cognitive load); an environmental sensor: detecting network latency and device temperature (judging the operating environment). Example: When the user is learning a math formula, the eye tracker finds that he repeatedly scans a certain step, the touch screen data shows that the answer is frequently modified, and the environmental sensor detects network jitter.

[0022] In this embodiment, the real-time knowledge attention vector represents the dynamic distribution of the attention intensity of the user to each node in the knowledge graph, and is generated through the following steps.

[0023] Beneficial effects of the above technical solution: By collecting multi-dimensional behavior data of end-users in real time and combining knowledge attention, cognitive load, and device efficiency, personalized optimization strategies are provided for mobile online learning. Compared with the prior art, this method realizes dynamic adjustment based on the learner's state. By generating content optimization strategies, controlling interface rendering, and resource allocation, the learning efficiency is improved. The closed-loop feedback mechanism further optimizes the learning content and user experience, reduces the cognitive load, and ensures the efficiency and personalization of the learning process. Example 2

[0024] An embodiment of the present invention provides a method for optimizing mobile online learning. By presetting a sensor array to collect the original behavior data of end-users, a learner status packet including a real-time knowledge attention vector, a cognitive load index, and a device efficiency score is output, including: Collect the original behavior data of end-users through a preset sensor array; Determine the real-time knowledge attention vector based on eye movement data, determine the cognitive load index based on touch data, determine the device efficiency score based on environmental data, and then output the learner status packet.

[0025] In this embodiment, the device efficiency score is obtained by analyzing the real-time environmental data of the mobile terminal and comprehensively evaluating the device's current ability to execute learning tasks, mainly including: Network status: latency, jitter, bandwidth fluctuation (affecting content loading); Hardware status: CPU / GPU load, memory margin, battery life (affecting calculation and rendering); Environmental interference: light intensity, background noise (affecting concentration). Generation logic: Data normalization: Convert each index into a standardized value between 0 and 1; Dynamic weighting: Adjust the weight according to the type of learning task (e.g., video courses focus on the network, interactive exercises focus on the CPU); Threshold warning: Trigger a downgrading strategy when the score is below the critical value (e.g., turn off animations). Example: Scenario: The user uses the mobile phone to study on the subway, and the environmental data detects: Network latency 200ms (poor); Remaining battery power 15% (low); Strong light causes screen reflection (interference). Scoring result: The device efficiency score is 0.3 (full score 1.0), and the system automatically: Switches to pure text mode; Preloads the basic content of the next chapter; Pauses non-essential background processes.

[0026] Beneficial effects of the above technical solution: By presetting a sensor array to collect multi-dimensional user behavior data, comprehensive monitoring of the learner's real-time state is realized. Compared with the prior art, the knowledge attention, cognitive load, and device efficiency score of the learner are extracted, and a comprehensive learner status packet is output, providing data support for subsequent learning content optimization. Through real-time feedback and adjustment, the personalized learning experience is improved, the cognitive load is effectively reduced, and the learning efficiency is optimized. Example 3

[0027] An embodiment of the present invention provides a method for optimizing mobile online learning. Based on eye movement data, a real-time knowledge attention vector is determined, including: Determine the eye movement fixation points and the corresponding fixation durations based on the eye movement data; Match the eye movement fixation points with the regions of the interface display content, and generate a distribution vector reflecting the relative attention intensity of each content block of the interface display content in combination with the fixation duration corresponding to each eye movement fixation point; Determine the distribution vector reflecting the relative attention intensity of each content block of the interface display content as the real-time knowledge attention vector.

[0028] In this embodiment, the eye movement fixation point refers to the real-time focus position coordinates (x, y) of the user's eyeball on the screen, and is sampled by an infrared eye tracker at a frequency of ≥60Hz. Example: When learning a mathematical formula, the fixation points are densely distributed around the "integral sign".

[0029] In this embodiment, the region of the interface display content divides the screen into several logical blocks, and each block is bound to a specific knowledge point: Block coding rule: Text area: Divided by paragraphs Chart area: Divided by graphic elements. Example: The chemical courseware interface is divided into a "molecular formula area" (A), a "reaction animation area" (B), and a "text description area" (C).

[0030] In this embodiment, the distribution vector of the relative attention intensity represents the proportion of the attention resources obtained by each content block. The determination method: Statistically calculate the total residence duration of all fixation points in the block, and calculate the percentage of this duration in the total fixation duration of the interface. Example: It is measured that area A accounts for 65%, area B 25%, and area C 10% → vector [0.65, 0.25, 0.1]. Technical linkage example, when it is detected that: 80% of the fixation points are concentrated in the "trigonometric function image" block, and the fixation duration ratio of this block reaches 70%, the system determines that this knowledge point is a high attention area and automatically increases its rendering priority.

[0031] The beneficial effects of the above technical solutions: By accurately determining the fixation points and fixation durations based on the eye movement data, and generating a distribution vector of attention intensity in combination with the region of the interface display content, the accurate measurement of the real-time knowledge attention is realized. Compared with the prior art, it can capture the attention dynamics of learners more carefully, thereby providing more accurate data support for personalized learning content optimization. By dynamically adjusting the learning content, the learning efficiency is improved, the cognitive load is reduced, and the user experience is optimized. Embodiment 4

[0032] An embodiment of the present invention provides a method for optimizing mobile online learning. Based on touch data, a cognitive load index is determined, including: Analyze the touch data to obtain the touch pressure volatility, touch trajectory data, and operation time data; Determine the cognitive load index based on the touch pressure volatility, touch trajectory data, and operation time data: where, is the touch pressure volatility, is the preset user baseline pressure volatility, d is the actual touch trajectory length, is the ideal touch trajectory length, is the preset acceptable deviation threshold, is the total touch pause duration, is the total touch operation duration is the preset minimum detectable pause time threshold, is the preset reference pause ratio, is the individual operation style adaptation factor, and , is the preset extremely small positive number.

[0033] In this embodiment, the minimum detectable pause time threshold is the shortest pause time threshold that has been experimentally verified and reliably recognized in human operation behavior, and is used to filter out invalid pauses caused by non-cognitive factors (such as touch sampling noise, muscle micro-vibrations, etc.). Its typical value is 200 - 300 milliseconds (based on research conclusions in the field of HCI); In this embodiment, the individual operation style adaptation factor is a dimensionless personalized calibration coefficient, which is used to quantify the deviation degree of the user's operation habit from the standard cognitive load model. Its characteristics are as follows: physical meaning, the user's sensitivity to trajectory deviation (low κ≈0.3: error-tolerant user; high κ≈1: precise user) the following characteristics of the user's historical operation data: average trajectory deviation degree, error correction behavior frequency, and the k factor is calibrated and obtained through the data of the user's initial N (N≥5) learning sessions.

[0034] In this embodiment, during the mobile online learning process, the interaction behavior between the learner and the device is reflected by the touch data, and these touch data are closely related to the learner's cognitive load. The following elaborates on the principle of determining the cognitive load index from three aspects: touch pressure volatility, touch trajectory data, and operation time data. The correlation between touch pressure volatility and cognitive load, when the learner is in different cognitive load states, their physiological state will change accordingly. When operating a mobile device for learning, this change will be reflected as a change in the hand muscle tension. The higher the cognitive load, the stronger the psychological tension of the learner, and the more unstable the control of the hand muscles, which will in turn cause fluctuations in the pressure applied during the touch screen process. The touch pressure volatility reflects the degree of pressure change per unit time. Comparing it with the preset user baseline pressure volatility comparatively, relatively The larger it is, the higher the current cognitive load borne by the learner, because a larger stress volatility means an increased degree of instability of the hand muscles due to cognitive activities; in the formula this term, when increases, the value of this term approaches 1, representing an increase in cognitive load; cognitive load will affect the learner's attention allocation and action control ability. When the cognitive load is low, the learner can control the operation trajectory of the finger on the screen more precisely, and the actual touch trajectory length d is relatively close to the ideal touch trajectory length . However, as the cognitive load increases, the learner needs to allocate more attention resources to the understanding and thinking of the learning content, resulting in a decline in the control ability of the touch operation. At this time, the actual touch trajectory will deviate from the ideal trajectory, and the degree of deviation can be measured by ( is the preset acceptable deviation threshold). In the formula , the farther the actual touch trajectory deviates from the ideal trajectory, that is, the larger the value of , the larger the value of the logarithmic term, and the larger the overall value of this term, reflecting a higher cognitive load. The individual operation style adaptation factor takes into account the influence of different learners' own operation habit differences on the trajectory deviation; the operation time data includes the total touch pause duration and the total touch operation duration. During the learning process, when encountering difficult-to-understand knowledge points or complex learning tasks, the learner needs more time for thinking, understanding, and digestion, which will lead to pauses in the touch operation. The proportional relationship between the total touch pause duration and the total touch operation duration can reflect the situation where the learner interrupts the operation due to cognitive processing during the learning process. The preset minimum detectable pause time threshold is used to filter out some extremely short pauses caused by non-cognitive factors such as device response delay, and the preset reference pause ratio is used as a comparison reference. In the formula this term, when the total touch pause duration is relatively long, that is, the learner spends more time thinking, the value of this term will increase accordingly, reflecting a higher state of cognitive load.

[0035] In this embodiment, , this sub-term measures the change of the current touch pressure volatility relative to the baseline pressure volatility through an exponential function. When is close to 0, that is, when the pressure volatility is very low, approaches 1, then approaches 0, indicating a lower cognitive load; when increases, that is, when the pressure fluctuates violently, approaches 0, approaches 1, reflecting a higher cognitive load. It characterizes the cognitive load from the perspective of pressure fluctuation, and there is a positive correlation between pressure fluctuation and cognitive load. This relationship is incorporated into the calculation of the entire cognitive load index through this sub-term; In this embodiment, it reflects the cognitive load from the perspective of the accuracy of the touch trajectory. The higher the cognitive load, the lower the operation accuracy and the greater the trajectory deviation. This association is incorporated into the formula through this sub-item; In this embodiment, , the numerator part represents the relative relationship between the total touch pause duration and the reference pause ratio after deducting the minimum detectable pause time threshold; the denominator part comprehensively considers the relationship between the total operation duration and the reference pause ratio. This sub-item reflects the pause situation during the touch operation through the operation of the two. When the total touch pause duration is relatively long, that is, the learner has more thinking time, the value of this item will increase. Function: Measure the cognitive load from the dimension of operation time, especially from the perspective of touch pause time. A long pause time means that the learner is thinking about the learning content and the cognitive load is relatively high. This sub-item incorporates this factor into the calculation of the cognitive load index; In this embodiment, is a preset extremely small positive number. During the actual mobile touch operation data collection process, due to factors such as the accuracy limitation of device sensors and data measurement errors, the situation where d is exactly equal to almost never occurs. However, from the perspectives of theoretical rigor and preventing program errors, it is still necessary to avoid such extreme situations. Selecting such an extremely small positive number is on the one hand because it is small enough that its numerical impact on the final calculation result can be ignored and it will not change the representation logic and magnitude of the formula for the cognitive load index; on the other hand, in computer science and mathematical calculations, is a commonly used extremely small value. Similar extremely small values also include etc., which have been widely used in algorithms and formulas that need to handle similar boundary situations and have universality and practicality. Therefore, considering the rigor of the comprehensive mathematical operation logic and the actual application scenario, an extremely small positive number such as is selected to avoid the situation where the logarithmic function is meaningless.

[0036] The beneficial effects of the above technical solutions: By analyzing the pressure volatility, trajectory data, and operation time in the touch data, the cognitive load index is accurately calculated. Compared with the prior art, this method can comprehensively consider all aspects of the touch operation, more accurately reflect the learner's cognitive load, timely adjust the learning content and interface, and optimize the learning experience. Through personalized cognitive load assessment, it can effectively reduce excessive or insufficient load, improve learning efficiency, and ensure that the learning process is more efficient and meets individual needs. Embodiment 5

[0037] An embodiment of the present invention provides a method for optimizing mobile online learning, which performs node-level matching between a knowledge attention vector and a pre-constructed knowledge graph to output a knowledge blind spot marking set. At the same time, according to the cognitive load index, the load level is divided to generate a learning state description including the blind spot marking and the load level, including: Traverse the knowledge attention vector, and mark the knowledge points with all attention values lower than the dynamic threshold as the first type of nodes; Retrieve the associated knowledge points in the pre-constructed knowledge graph that have a direct cognitive dependence relationship with the first type of nodes, and mark them as the second type of nodes after screening by the cognitive dependence intensity; Based on the attention deviation degree of the first type of nodes, the cognitive dependence intensity between the second type of nodes and the core blind spots, and the topological importance of the first type of nodes and the second type of nodes in the knowledge graph, determine the blind spot repair priority, and then output the knowledge blind spot marking set; According to the numerical range of the cognitive load index, divide the user's real-time cognitive load level; Integrate the knowledge blind spot marking set and the cognitive load level to generate a learning state description.

[0038] In this embodiment, for the knowledge points with all attention values lower than the dynamic threshold, the dynamic threshold is a determination criterion automatically adjusted according to the real-time distribution characteristics of the knowledge attention vector, and is used to identify low-attention knowledge points. Its calculation comprehensively considers the overall attention level fluctuation (such as mean, dispersion) and learning stage characteristics (preview / review) to ensure accurate identification of blind spots in different learning scenarios. For example, the threshold is automatically increased in the review stage to strictly screen weak points.

[0039] In this embodiment, the output of the knowledge blind spot marking set includes a marking set containing the following information: node classification identifier (first type / second type); attention deviation level; repair priority rating; associated node reference list.

[0040] In this embodiment, the real-time cognitive load level is to establish a mapping relationship between the cognitive load index and the interaction quality index; when high-frequency misoperations accompanied by eye movement dispersion characteristics are detected, the load level is upgraded. Low load (CLI ≤ threshold 1): The user has sufficient cognitive resources and can accept high-complexity learning content; Medium load (threshold 1 < CLI ≤ threshold 2): The user is in a normal learning state and needs to balance the content difficulty and interaction intensity; High load (CLI > threshold 2): The user's cognition is overloaded and needs to reduce the complexity of the learning content or provide auxiliary strategies.

[0041] In this embodiment, the learning state description: knowledge blind spot distribution (core blind spot, associated influence area); repair priority ranking (knowledge points that need to be strengthened first); current load level (low / medium / high); recommended intervention strategies (such as simplifying content, increasing guidance, adjusting rhythm).

[0042] Beneficial effects of the above technical solution: By combining the knowledge attention vector with the node-level matching of the knowledge graph, the knowledge blind spots of learners can be accurately identified, and the learning status description can be dynamically adjusted according to the cognitive load index. Compared with the prior art, this method can effectively mark the knowledge blind spots, and preferentially repair the blind spots according to the cognitive dependence relationship and the cognitive load of the learners, and optimize the presentation order of the learning content. This method improves the personalization and pertinence of the learning content, ensures a moderate cognitive load during the learning process, and enhances the learning efficiency and experience. Embodiment 6

[0043] The embodiment of the present invention provides a mobile online learning optimization method, which generates a primary content optimization strategy based on a blind spot marking set, combines the load level and the device efficiency score to verify the feasibility of the strategy, and outputs a final executable strategy set, including: From a predefined strategy template library, load the basic strategy framework according to the node type in the blind spot marking set, and then generate the corresponding content optimization strategy; Verify the feasibility of the content optimization strategy based on the load level and the device efficiency score, and output the final executable strategy set.

[0044] In this embodiment, loading the basic strategy framework according to the node type in the blind spot marking set includes: knowledge blind spot distribution (core blind spot, associated influence area); repair priority ranking (knowledge points that need to be strengthened preferentially); current load level (low / medium / high); recommended intervention strategies (such as simplifying content, adding guidance, adjusting rhythm).

[0045] In this embodiment, the corresponding content optimization strategy is generated based on the knowledge points corresponding to the first type of nodes in the blind area marker set to generate a deep reinforcement intervention plan; (1. Content reconstruction strategy: Multimodal content replacement: Replace the original text / formula with dynamic visual content (such as interactive charts, concept animations). Example: Mathematical formula → Step-by-step derivation animation + Real-time variable adjustment tool. Cognitive scaffold construction, insert a guiding question chain, and guide users to independently derive conclusions through progressive questioning. Example: When learning physical laws, gradually prompt "What will the result be if the XX parameter is changed?" 2. Interaction enhancement strategy: Instant feedback training: Embed a micro-test unit, which is automatically triggered after answering wrong: Error cause analysis (such as "Confused concept A with B"); Targeted remedial content (such as playing a 30-second intensive lecture on the corresponding knowledge point). Operational simulation, provide a virtual experiment environment (such as chemical molecule assembly, programming code debugging sandbox), and strengthen understanding through operations. 3. Path guidance strategy, Make the dependency relationship explicit: Dynamically display the knowledge dependency graph beside the content, and highlight the association between the current node and the mastered / unmastered nodes. Example: When learning "Derivative application", the sidebar shows its dependency relationship with "Limit calculation"). Based on the second type of nodes (associated influence area) in the blind area marker set, generate an auxiliary learning plan; (1. Lightweight content hint: Contextual association hint: When the user touches the core blind area, automatically floatingly display the key conclusion card of the associated knowledge point. Example: When learning "Multivariable calculus", floatingly prompt "Definition of partial derivative (associated prerequisite knowledge)". Comparison learning module, juxtapose and display the differences between easily confused concepts (such as a comparison table of "Chemical reaction VS Physical change"). 2. Adaptive review mechanism: Triggered review: When it is detected that the user's operation is hesitant (such as long pause, repeated modification), push a 60-second quick review video of the associated knowledge; Error question traceability reinforcement: When the user answers a question wrong in the core blind area, automatically insert a minimized test question of the associated knowledge point (such as 1-2 multiple-choice questions); 3. Load optimization: Information is loaded in chunks, disassemble the associated knowledge into collapsible content chunks, and the user can expand them as needed (such as "Click to view the proof process"). Non-intrusive annotation, use colors / icons to mark the reference points of the associated knowledge in the text, and display the summary when hovering (to avoid interrupting the main process)).

[0046] The beneficial effects of the above technical solution: By generating a primary content optimization strategy based on the blind area marker set and combining the cognitive load level and device efficiency score to verify the feasibility of the strategy. Compared with the prior art, this method optimizes through personalized strategies, ensuring that the content presentation order conforms to the learner's cognitive state and device conditions, effectively avoiding excessive load or device performance bottlenecks, and improving learning efficiency and experience. Through dynamic adjustment of the strategy, more efficient learning resource allocation and personalized learning support are achieved. Embodiment 7

[0047] An embodiment of the present invention provides a method for optimizing mobile online learning, which executes an executable policy set and controls interface rendering and resource allocation, including: When executing the executable policy set, adjust the rendering priority of interface elements according to the content optimization policy; obtain the environmental state prediction result based on a preset algorithm and environmental data; Allocate network resources based on the environmental state prediction result.

[0048] When executing the content policy, adjust the rendering priority of interface elements according to the content optimization policy; obtain the environmental state prediction result based on a preset algorithm and environmental data; Allocate network resources based on the environmental state prediction result.

[0049] In this embodiment, to adjust the rendering priority of interface elements: Step 1: Visual salience grading, core content enhancement: For the knowledge points marked as high priority in the policy set, implement the following optimizations: Spatial dominance: Expand the content area to more than 40% of the screen visible area; Dynamic focus: Add visual guidance animations (such as pulse highlighting, progressive expansion); Color contrast: Use complementary color combinations (such as dark blue background + bright yellow text) to improve readability. Auxiliary content degradation: For low-priority content: Fold and hide: By default, only the title is displayed, and the details are expanded after clicking; Transparency adjustment: Reduce the saturation of non-core areas (such as grayscale processing).

[0050] In this embodiment, allocating network resources based on the environmental state prediction result includes: Network state modeling: Analyze historical RTT (round-trip time) and packet loss rate data to predict the available bandwidth range in the next 10 seconds: Stable and high speed (>5Mbps): Enable high-definition resource preloading; Fluctuating medium speed (1-5Mbps): Switch to adaptive bitrate streaming; Low speed / unstable (<1Mbps): Start the offline cache priority mode. Continuing learning offline: When a network interruption is detected: Automatically switch to the copies of the last 3 knowledge points in the local cache; Limit new content requests and only maintain core interaction functions. Resource competition arbitration: When multiple policies compete for bandwidth: Prioritize ensuring data transmission of the answer verification interface; Pause non-real-time updates (such as learning progress synchronization).

[0051] The beneficial effects of the above technical solutions: By collecting the eye movement trajectory change data and system performance metrics after policy execution, accurately analyze the actual effects of the optimization policies. Compared with the prior art, this method can not only evaluate the optimization effect of learning content through visual attention transfer features and micro-test results, but also comprehensively understand the influencing factors of the policies in combination with the device resource consumption situation. Through the feedback data stream, timely update the knowledge attention vector, continuously optimize the learning content, and improve the knowledge retention rate and learning efficiency of learners. Embodiment 8

[0052] An embodiment of the present invention provides a method for optimizing mobile online learning, which collects eye movement trajectory change data and system performance metrics after the execution of the final executable policy set, and outputs a feedback data stream for updating the knowledge attention vector, including: Extract visual attention transfer features from the eye movement trajectory change data of the user's attention to the optimized content after the execution of the monitoring policy; Record the device resource consumption during the execution of the policy; Extract the core examination points of the target node from the pre-constructed knowledge graph to generate a micro-test and obtain the micro-test results; Determine the difference in the knowledge attention vectors before and after the execution of the policy based on the visual attention transfer features, and combine the micro-test results and the device resource consumption to determine the reasons for the change in the actual knowledge retention rate; Generate a corresponding feedback data stream based on the reasons for the change in the actual knowledge retention rate.

[0053] In this embodiment, the generation of dynamic test questions includes: Step 1: Extract the core examination points of the target node from the knowledge graph; automatically generate micro-test questions that meet the following principles: Option interference: The wrong options contain typical cognitive misunderstandings (such as confusing "derivative" with "differential"); Answer interpretability: Each question is accompanied by an instant analysis of ≤50 words. Step 2: Non-intrusive trigger trigger conditions (start when any one is met): The user stays at the current knowledge point for more than the threshold; Detect repeated content label switching behavior; Mandatory detection points preset by the policy (such as every 3 learning units completed). Policy effectiveness verification: Correct rate ≥80% → Mark the policy as effective and strengthen the corresponding content form weight; Correct rate ≤50% → Trigger the policy rework and optimization process. Cognitive load monitoring: Sudden increase in response time by 200% + error → Indicate cognitive overload and need to simplify the content; Frequent answer modification + low correct rate → Prompt the lack of knowledge point association. For example: Embodiment (mathematics learning scenario) Test question generation: Core knowledge point: Trigonometric sum formula Micro-test question: "What is the correct expansion of sin(A + B)? ①sinAcosB + cosAsinB ②sinA + sinB..." Result application: If the wrong option ② is selected, the feedback system automatically strengthens the comparative explanation of "function superposition vs linear superposition"; If the answering time > 15 seconds but correct, reduce the appearance frequency of subsequent similar questions.

[0054] In this embodiment, the difference in the knowledge attention vectors before and after the execution of the policy is the improvement amplitude of the attention to the core blind area nodes and the attention diffusion degree of the associated influence area; In this embodiment, by combining the micro-test results and the device resource consumption, the reasons for the change in the actual knowledge retention rate are determined, including: Attribution of strategy effectiveness: If the attention level increases and the test accuracy rate rises, mark the strategy as effective; Attribution of device performance: If the resource consumption is too high resulting in interaction latency, mark the strategy that needs to be degraded; Attribution of load adaptability: If the strategy causes an increase in the operation error rate under high load, mark the strategy that needs to be simplified.

[0055] In this embodiment, a corresponding feedback data stream is generated based on the reasons for the change in the actual knowledge retention rate, including: Cause classification and tagging, Classification of strategy effectiveness: Effective strategy: If the knowledge retention rate increases and the eye movement trajectory shows concentrated attention, mark it as a "success case"; Ineffective strategy: If the retention rate remains unchanged but the system load is normal, mark it as "need to optimize content matching degree"; Failed strategy: If the retention rate decreases or causes high load, mark it as "need to be eliminated or degraded". Attachment of attribution tags: Add an execution environment tag to each strategy (such as "applicable to high device efficiency" "use with caution under high load"); Record the characteristics of the associated knowledge points (such as "applicable to abstract concepts" "rely on visual memory"). Structuring of the feedback data stream: Generation of model update instructions: For the knowledge attention vector model: Successful strategy: Reinforce the initial weight assignment of the relevant nodes; Failed strategy: Reduce the default weight of the corresponding content form (such as animation / text). Update rule: Only adjust the model parameters directly associated with the strategy to avoid global perturbations. Iteration instructions for the strategy library: Store the successful strategies and their environment tags in the preferred strategy library for subsequent priority calls; For ineffective / failed strategies: Add restrictive conditions (such as "only enable when CLI < 0.5"); or move them to the library to be verified and wait for further testing.

[0056] Beneficial effects of the above technical solution: By executing the executable policy set, the interface rendering and resource allocation are optimized to ensure that the presentation priority of learning content meets the needs of learners. Compared with the prior art, this method dynamically adjusts the network resource allocation in combination with the environmental state prediction results, avoids resource bottlenecks, and ensures a smooth learning process. By adjusting the interface rendering priority, the user experience and learning efficiency are improved, while the use of device resources is optimized, enhancing the intelligence and adaptability of the learning platform. Embodiment 9

[0057] An embodiment of the present invention provides a mobile online learning optimization system, as Figure 2 shown, including: A data collection module for collecting the original behavior data of terminal users through a preset sensor array; A state analysis module for performing node-level matching between the knowledge attention vector and a pre-constructed knowledge graph, outputting a knowledge blind area mark set, and at the same time, dividing the load level according to the cognitive load index and generating a learning state description including the blind area mark and the load level; Policy Generation Module: Generates primary content optimization policies based on the blind spot tag set, verifies the feasibility of the policies by combining the load level and device efficiency score, and outputs the final executable policy set; Execution Control Module, executes the executable policy set and controls the interface rendering and resource allocation feedback learning module; Feedback Learning Module, collects the eye movement trajectory change data and system performance metrics after the execution of the final executable policy set, and outputs the feedback data stream for updating the knowledge attention vector.

[0058] In this embodiment, the data acquisition module is used to collect the original behavior data of the end user through a preset sensor array, including: an eye movement tracking unit configured to obtain eye movement fixation points and fixation duration data; a touch perception unit configured to record touch pressure, trajectory, and operation time data; and an environment monitoring unit configured to detect network status, device temperature, and power data.

[0059] In this embodiment, the state analysis module includes: a knowledge attention calculation engine configured to map the eye movement data to a real-time knowledge attention vector; a cognitive load evaluator configured to generate a cognitive load index based on the touch data; and a device efficiency analyzer configured to evaluate the terminal operating state and output a device efficiency score.

[0060] In this embodiment, the policy generation module includes: a knowledge graph matching component configured to perform node-level matching of the knowledge attention vector with a pre-constructed knowledge graph; a blind spot tag generator configured to output a knowledge blind spot tag set and a repair priority; and a policy optimization decision maker configured to generate an executable policy set by combining the load level and device efficiency score.

[0061] In this embodiment, the execution control module includes: a dynamic rendering controller configured to adjust the display priority of interface elements and the visual enhancement scheme; and a resource allocation scheduler configured to manage network and computing resources based on the environment prediction result.

[0062] In this embodiment, the feedback learning module includes: an effect monitor configured to collect eye movement trajectory changes and system performance metrics; and a model update engine configured to optimize the knowledge attention vector generation rule according to the feedback data stream.

[0063] In this embodiment, it further includes: a system bus that connects each module in a unidirectional data flow manner to ensure that: the output of the data acquisition module is only transmitted to the state analysis module; the input of the policy generation module strictly depends on the output of the state analysis module; the instruction generation of the execution control module is only based on the decision result of the policy generation module; and the data input of the feedback learning module only comes from the monitoring output of the execution control module.

[0064] Beneficial effects of the above technical solution: By collecting multi-dimensional behavior data of end-users in real time and combining knowledge attention, cognitive load, and device efficiency, personalized optimization strategies are provided for mobile online learning. Compared with the prior art, this method realizes dynamic adjustment based on the learner's state. By generating content optimization strategies, controlling interface rendering, and resource allocation, learning efficiency is improved. The closed-loop feedback mechanism further optimizes learning content and user experience, reduces cognitive load, and ensures the efficiency and personalization of the learning process.

[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features. However, such modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An optimization method for mobile online learning, characterized in that, Including: Step 1: Collect the original behavior data of the end-user through a preset sensor array. The original behavior data includes eye movement data, touch data, and environmental data, and output a learner status package containing a real-time knowledge attention vector, a cognitive load index, and a device efficiency score; Step 2: Perform node-level matching between the knowledge attention vector and a pre-constructed knowledge graph, and output a knowledge blind spot marker set. At the same time, divide the load level according to the cognitive load index, and generate a learning status description containing the blind spot marker and the load level; Step 3: Generate a primary content optimization strategy based on the blind spot marker set, verify the feasibility of the strategy by combining the load level and the device efficiency score, and output a final executable strategy set; Step 4: Execute the executable strategy set and control interface rendering and resource allocation; Step 5: Collect the eye movement trajectory change data and system performance metrics after executing the final executable strategy set, and output a feedback data stream for updating the knowledge attention vector.

2. The mobile - side online learning optimization method according to claim 1, wherein, Collect the original behavior data of the end-user through a preset sensor array, and output a learner status package containing a real-time knowledge attention vector, a cognitive load index, and a device efficiency score, including: Collect the original behavior data of the end-user through a preset sensor array; Determine the real-time knowledge attention vector based on the eye movement data, determine the cognitive load index based on the touch data, determine the device efficiency score based on the environmental data, and then output the learner status package.

3. The mobile terminal online learning optimization method according to claim 2, wherein Determine the real-time knowledge attention vector based on the eye movement data, including: Determine the eye movement fixation points and the corresponding fixation durations based on the eye movement data; Match the eye movement fixation points with the regions of the interface display content, and generate a distribution vector reflecting the relative attention intensity of each content block of the interface display content in combination with the fixation duration corresponding to each eye movement fixation point; Determine the distribution vector reflecting the relative attention intensity of each content block of the interface display content as the real-time knowledge attention vector.

4. The mobile terminal online learning optimization method according to claim 2, characterized in that Determine the cognitive load index based on the touch data, including: Analyze the touch data to obtain the touch pressure volatility, touch trajectory data, and operation time data; Determine the cognitive load index based on the touch pressure volatility, touch trajectory data, and operation time data: Among them, is the touch pressure volatility, is the preset user baseline pressure volatility, d is the actual touch trajectory length, is the ideal touch trajectory length, is the preset acceptable deviation threshold, is the total touch pause duration, is the total touch operation duration, is the preset minimum detectable pause time threshold, is the preset reference pause ratio, is the individual operation style adaptation factor, and , is a preset extremely small positive number.

5. The mobile terminal online learning optimization method according to claim 1, characterized in that Perform node-level matching between the knowledge attention vector and a pre-constructed knowledge graph, and output a knowledge blind spot marker set. At the same time, divide the load level according to the cognitive load index, and generate a learning status description containing the blind spot marker and the load level, including: Traverse the knowledge attention vector, and mark the knowledge points with all attention values lower than the dynamic threshold as the first type of nodes; Retrieve the associated knowledge points in the pre-constructed knowledge graph that have a direct cognitive dependence relationship with the first type of nodes, and mark them as the second type of nodes after screening by the cognitive dependence intensity; Determine the blind spot repair priority based on the attention deviation degree of the first type of nodes, the cognitive dependence intensity between the second type of nodes and the core blind spot, and the topological importance of the first type of nodes and the second type of nodes in the knowledge graph, and then output the knowledge blind spot marker set; Divide the real-time cognitive load level of the user according to the numerical range of the cognitive load index; Integrate the knowledge blind spot marker set and the cognitive load level to generate a learning status description.

6. The mobile terminal online learning optimization method according to claim 1, wherein Generate a primary content optimization strategy based on the blind spot tag set, verify the feasibility of the strategy by combining the load level and the device efficiency score, and output the final executable strategy set, including: From the predefined strategy template library, load the basic strategy framework according to the node types in the blind spot tag set, and then generate the corresponding content optimization strategy; Verify the feasibility of the content optimization strategy based on the load level and the device efficiency score, and output the final executable strategy set.

7. An optimization method for mobile online learning according to claim 1, characterized in that Execute the executable strategy set and control the interface rendering and resource allocation, including: When executing the executable strategy set, adjust the rendering priority of the interface elements according to the content optimization strategy; obtain the environmental state prediction result based on the preset algorithm and environmental data; Allocate network resources based on the environmental state prediction result.

8. The mobile - end online learning optimization method according to claim 1, characterized in that, Collect the eye movement trajectory change data and system performance metrics after the execution of the final executable strategy set, and output the feedback data stream for updating the knowledge attention vector, including: Monitor the eye movement trajectory change data of the user for the optimized content after the strategy execution to extract the visual attention transfer features; Record the device resource consumption during the strategy execution; Extract the core key points of the target node from the pre-constructed knowledge graph to generate a micro-test and obtain the micro-test result; Determine the difference in the knowledge attention vectors before and after the strategy execution based on the visual attention transfer features, and combine the micro-test results and the device resource consumption to determine the reason for the change in the actual knowledge retention rate; Generate the corresponding feedback data stream based on the reason for the change in the actual knowledge retention rate.

9. A mobile online learning optimization system for implementing the method according to any one of claims 1-8, characterized in that, Including: The data acquisition module is used to collect the original behavior data of the end user through the preset sensor array; The status analysis module is used to perform node-level matching of the knowledge attention vector and the pre-constructed knowledge graph, output the knowledge blind spot tag set, and at the same time, divide the load level according to the cognitive load index, and generate a learning status description including the blind spot tag and the load level; The strategy generation module: generate a primary content optimization strategy based on the blind spot tag set, verify the feasibility of the strategy by combining the load level and the device efficiency score, and output the final executable strategy set; The execution control module executes the executable strategy set and controls the interface rendering and resource allocation feedback learning module; The feedback learning module collects the eye movement trajectory change data and system performance metrics after the execution of the final executable strategy set, and outputs the feedback data stream for updating the knowledge attention vector.

Images