Dynamic resource configuration and interaction system for language learning based on internet platform

Abstract

The application provides an internet platform-based language learning dynamic resource configuration and interaction system, relates to the technical field of data processing systems, and provides a double-layer cascade diagnosis mechanism; through a first calculation module, a state abnormality score is calculated based on the stability of user interaction behavior instead of simple right or wrong, so as to accurately distinguish between 'beneficial struggle' and 'harmful fatigue'; when the score exceeds a threshold value, a second calculation module is activated, specific learning segments are analyzed in combination with eye movement track data, and a cognitive bias value of quantitative specific cognitive impairment is calculated; on the one hand, through accurate state recognition, error intervention is avoided when the user is deeply thinking, and learning flow is effectively protected; on the other hand, through accurate cognitive attribution, the system can provide targeted and accurate assistance, improve tutoring efficiency and learning effect, and realize intelligent teaching upgrading.

Description

Technical Field

[0001] This invention relates to the field of data processing system technology, specifically to a dynamic resource allocation and interaction system for language learning based on an Internet platform. Background Technology

[0002] By capturing learners' facial expressions through webcams and using affective computing technology to determine their emotional state (such as joy, confusion, fatigue), the learning pace can be adjusted or incentive content can be pushed. These technologies enhance the personalization and intelligence of learning systems to some extent, allowing users to learn in an environment more suited to their abilities and states.

[0003] In the prior art, publication number CN112053020A, entitled "A Human-Computer Interactive Assisted Language Learning System," describes a method, system, and device for promoting language learning through human-computer interactive assisted language learning. The learning system and device implement an AI engine to learn each learner's preferred learning style and information about each learner. The preferred learning style influences the learning process.

[0004] Existing online language learning systems based on internet platforms typically adjust the difficulty of questions based on the user's correct answer rate or adjust the learning pace based on the emotions captured by the camera. This approach suffers from the following two major technical shortcomings:

[0005] Technical Deficiency 1: Existing systems suffer from insufficient resolution in diagnosing macroscopic learning states. They rely on outcome-oriented, coarse-grained indicators (such as correct or incorrect answers, time spent) or external emotional representations, making it difficult to effectively distinguish between two fundamentally different learning states: "Desirable-Difficulty" (simply put, beneficial struggle) and "Cognitive-Overload" (simply put, harmful fatigue). The former is the necessary cognitive load exhibited by learners when challenging their ability boundaries and engaging in deep information processing, a crucial process for building deep understanding and long-term memory; the latter is a negative state caused by learning content exceeding the learner's current cognitive capacity, leading to frustration, anxiety, or even abandonment of learning. Due to a lack of dynamic modeling of the stability of the user's internal cognitive processes, existing technologies often confuse these two states, resulting in incorrect intervention decisions: either intervening too early when learners are engaged in efficient deep thinking, disrupting the valuable "flow" experience; or reacting too slowly when learners are already in a cognitive predicament, missing the optimal intervention opportunity.

[0006] The second technical deficiency is the lack of diagnostic capability in attributing micro-level cognitive impairments to their causes. Even if the system can identify that a learner is encountering difficulties, it cannot accurately trace the root cause of their cognitive biases. For example, when the system detects that a learner spends too much time on a long and complex sentence, it cannot determine whether the root cause is a lack of understanding of the meaning of specific unfamiliar words, a failure to parse complex syntactic structures (such as nested clauses), or a lack of knowledge of a particular idiom or cultural background. This lack of diagnostic capability directly leads to the intervention measures provided by the system often being "one-size-fits-all" and non-targeted, such as providing whole-sentence translations or broad grammar point links. These interventions are inefficient, fail to truly address the learner's core problems, and cannot achieve precise and efficient personalized tutoring.

[0007] The information disclosed in the background section above is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0008] The purpose of this invention is to provide a dynamic resource allocation and interaction system for language learning based on an Internet platform, so as to solve the problems mentioned in the background art.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] A dynamic resource allocation and interaction system for language learning based on an internet platform, specifically including:

[0011] Data acquisition module: used to continuously acquire interactive behavior data that characterizes the user's macro-level learning status, as well as eye movement trajectory data collected by eye-tracking devices when the user reads learning materials;

[0012] First calculation module: used to calculate, in real time, a state anomaly score representing the stability of user behavior patterns based on the interactive behavior data;

[0013] Cognitive Diagnosis Activation Module: Used to activate the cognitive diagnosis process in response to the abnormal state score exceeding a preset diagnostic trigger threshold;

[0014] The second calculation module is used to calculate, in the cognitive diagnosis process, a cognitive deviation value that quantifies the degree of user comprehension deviation for the learning material segment corresponding to the time when the diagnostic trigger threshold is triggered, in combination with the eye movement trajectory data.

[0015] Intervention module: Based on the state abnormality score and the cognitive bias value, it selects and executes the corresponding adaptive intervention operation from a preset intervention strategy library.

[0016] Compared with existing technologies, the beneficial effects of this invention are as follows: Through the first calculation module, the system continuously analyzes the user's interactive behavior data stream, establishes a baseline model that characterizes the stability of individual user behavior patterns, and calculates the anomaly score of the current behavior relative to this baseline in real time. This anomaly score does not measure the speed or correctness of the behavior, but rather quantifies its deviation from the individual's normal state. This allows the system to accurately distinguish between two learning states: in "beneficial struggle," the user's behavior pattern, though tense, still maintains regularity, resulting in a low anomaly score; while in "harmful fatigue," the behavior pattern exhibits disorder and inconsistency, leading to a significant increase in the anomaly score.

[0017] A two-stage cascaded diagnostic process is constructed. First, the cognitive diagnostic process is activated only when the abnormal state score exceeds a preset threshold. Then, the second calculation module retrieves the learning material segment corresponding to the trigger moment and, combined with high-precision eye-tracking data, calculates a cognitive bias value that quantifies the root cause and degree of comprehension deviation. For example, prolonged fixation on a single word indicates semantic impairment, while repeated re-viewing of syntactic components indicates structural comprehension difficulties.

[0018] It achieves accurate identification of learning states and flow protection: the system can accurately determine whether the user is in a state of efficient deep thinking or inefficient cognitive overload, thereby avoiding erroneous intervention at critical moments, effectively protecting the user's learning flow, and improving the learning experience and immersion.

[0019] This approach achieves precise targeting and efficient tutoring: Based on accurate diagnosis of the root causes of cognitive biases, the intervention module can access auxiliary tools from the intervention strategy library, such as providing definitions for new words or highlighting the structure of complex clauses. This targeted intervention model greatly enhances the effectiveness of the assistance and learning efficiency, truly realizing intelligent and personalized language learning tutoring. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall system flowchart of the present invention.

[0021] Figure 2 This is a schematic diagram of the logic block of the first calculation module of the present invention.

[0022] Figure 3 This is a schematic diagram of the logic block of the cognitive diagnosis activation module of the present invention.

[0023] Figure 4 This is a schematic diagram of the logic block of the second calculation module of the present invention.

[0024] Figure 5 This is a schematic diagram of the logic block of the intervention module of the present invention. Detailed Implementation

[0025] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0026] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0027] Example 1:

[0028] Please see Figures 1 to 5 The present invention provides a technical solution:

[0029] A dynamic resource allocation and interaction system for language learning based on an internet platform, executed by computing devices, specifically includes:

[0030] Data acquisition module: used to continuously acquire interactive behavior data that characterizes the user's macro-level learning status, as well as eye movement trajectory data collected by eye-tracking devices when the user reads learning materials;

[0031] First calculation module: used to calculate, in real time, a state anomaly score representing the stability of user behavior patterns based on the interactive behavior data;

[0032] To further explain, in the first calculation module, calculating the state anomaly score includes:

[0033] The interaction data is processed into a time series;

[0034] Based on the time series, calculate the weighted interaction entropy that characterizes the degree of disorder in the user behavior pattern;

[0035] The time change rate of the weighted interactive entropy is used as the state anomaly score.

[0036] To further explain, the calculation steps for the state anomaly score specifically include:

[0037] Within a preset time window, the interactive behavior data is identified into multiple discrete interactive event types, and the probability of each discrete interactive event type occurring within the time window is determined.

[0038] Obtain a preset cognitive association weight that corresponds one-to-one with each discrete interaction event type. The cognitive association weight is used to characterize the degree of correlation between the discrete interaction event type and the user's cognitive load.

[0039] Based on the probability of occurrence of each discrete interaction event type and its corresponding cognitive association weight, the weighted interaction entropy is calculated as a quantitative representation of the degree of disorder in the user's current behavior pattern.

[0040] The state anomaly score is determined based on the change of the weighted interaction entropy within a continuous time window;

[0041] The beneficial effects of the first calculation module are explained as follows: different interactive behaviors have different importance in reflecting the user's cognitive state; the introduction of "cognitive association weight" makes the calculation of "weighted interaction entropy" related to cognitive psychology and meaningful "cognitive load disorder", which improves the accuracy and pertinence of state assessment, and is one of the innovative directions of this embodiment;

[0042] Furthermore, the discrete interactive event types include, but are not limited to: mouse click events, keyboard input events, interface switching events, content scrolling events, or text highlighting events; the system can configure and add or delete the event types that need to be monitored according to specific application scenarios.

[0043] The specific implementation process of the first calculation module in calculating the status anomaly score is broken down as follows:

[0044] 1.1) Determining the weighted interaction entropy: The goal is to calculate the "weighted interaction entropy" that accurately reflects the degree of disorder in the user's current cognitive load;

[0045] 1.11) Determine the probability of discrete interaction event types: Within a preset time window of length T, the system's front-end collector identifies and classifies all user interaction behavior data to obtain N discrete interaction event types;

[0046] "Discrete interactive event types" refer to specific user actions defined and identified by the program, including but not limited to "submitting an answer," "requesting a prompt," and "looking up a dictionary." These are captured in real-time by deploying event listeners at the software front end.

[0047] Use "Discrete Interaction Event Type" Characterization; the probability of the i-th event occurring is denoted as... The cognitive association weight of the i-th event is denoted as... ;

[0048] Subsequently, statistics were compiled for each type of discrete interactive event. Calculate the number of times it occurs within this time window and its probability of occurrence. Probability of occurrence This is the ratio of the number of occurrences of the i-th type of event to the total number of occurrences of all events within the time window, and is a value between 0 and 1;

[0049] 1.12) For obtaining cognitive association weights: The system retrieves the weights from a pre-defined configuration library for each discrete interaction event type. Obtain its unique corresponding cognitive association weight Cognitive association weights This is a dimensionless numerical value used to represent the degree of positive correlation between the i-th event and the user's cognitive difficulty or attention deficit. This cognitive association weight... The value is determined in two ways:

[0050] The first method involves language education and cognitive psychology experts pre-calibrating and storing the data in the system configuration table based on their experience.

[0051] The second approach involves training a machine learning model on historical user data to automatically generate and optimize the algorithm. For example, the weight of "frequent request prompts" would be significantly higher than the weight of "completing exercises sequentially".

[0052] 1.13) Calculation of weighted interaction entropy: Let the weighted interaction entropy be denoted as... The weighted cross-entropy The calculation method is as follows: calculate the probability of occurrence of each "discrete interactive event type". Its corresponding cognitive association weight Perform a product operation to obtain a weighted probability term; then, multiply this weighted probability term with the probability of the event occurring. Multiply the logarithmic values ​​again; finally, sum the results calculated for all event types and take their opposites.

[0053] 1.2) For the determination of the state anomaly score: The goal is to calculate the final "state anomaly score" based on the time series of "weighted cross-entropy".

[0054] 1.21) For the construction of time series: The system repeatedly executes "1.1)" with a fixed time window T as the period, thereby obtaining a weighted cross-entropy. The constructed time series is denoted as , , etc., where the subscript t represents the index of the current time window;

[0055] 1.22) Calculation of the state anomaly score: The state anomaly score is denoted as... The abnormal state score The initial source of the calculation method is the mathematical formula for the rate of change; specifically: the first step is to calculate the arithmetic mean of the "weighted interactive entropy" within the most recent k historical time windows to obtain the historical baseline entropy value; then, the weighted interactive entropy of the current time window is used... Subtracting the historical baseline entropy value yields the entropy increment; finally, this entropy increment is divided by the weighted interaction entropy of the current time window. This yields a standardized ratio value, which is called the "state anomaly score"; State Anomaly Score The higher the score, the greater the deviation of the user's state from its recent stable state.

[0056] Further explanation of the first calculation module: the weighted interactive entropy The calculation of this formula originates from the Shannon entropy formula in information theory. The improved calculation logic in this embodiment is as follows: for each type of discrete interaction event... The corresponding "probability of occurrence". Multiply by its cognitive association weight Multiply by the probability of occurrence The logarithm of all events is calculated, and finally the results of all events are summed and inverted.

[0057] In this calculation, cognitive association weights As a multiplier, it directly affects the probability of occurrence. Above, it occupies a core regulatory position. Cognitive association weight. The setting is based on the positive correlation between the event and cognitive difficulty: events that are directly related to the learning task and indicate that the user is encountering difficulties are given a higher cognitive association weight. Values ​​are assigned to routine or meaningless operations, while lower cognitive association weights are assigned to them. Value. This structure makes the weighted cross-entropy... The calculation results become more sensitive to events that better reflect cognitive struggles;

[0058] In this embodiment, "difficult events" include, but are not limited to, "requesting hints" and "repeated word lookup";

[0059] "Routine or meaningless operations" include, but are not limited to, "screen scrolling" and "aimless clicks";

[0060] Weighted Inter-entropy Values ​​and Probability of Occurrence And cognitive association weight All are positively correlated. The more chaotic the user's behavior pattern, that is, the higher the probability of multiple events occurring. The more evenly the distribution, and the more likely these chaotic behaviors are to have high cognitive association weights. When the event is a value, the weighted interaction entropy The calculation results will increase significantly;

[0061] Abnormal state score The output range is limited to the interval (-∞, 1), but in practical applications, the vast majority of values ​​will fall between (-1, 1), which reflects both the technical effectiveness and the stability of the data.

[0062] When the state is abnormal score The closer the output is to 1, the greater the increase in the disorder of the user's current behavior relative to its recent stable state. This implies the current weighted interaction entropy. This is significantly higher than its historical average. This occurs because user interaction behavior rapidly shifts from a low-weighted interaction entropy over a short period. The ordered state abruptly transforms into a highly weighted cross-entropy. The disordered state. State anomaly score. Approaching 1 is a clear signal from the system, indicating that the user may have encountered a cognitive bottleneck, their learning flow has been disrupted, and they are on the verge of cognitive overload.

[0063] In this embodiment, the low-weighted interaction entropy Ordered states include, but are not limited to, focusing on answering questions sequentially, where the probability of occurrence is... Concentrated on a few low cognitive association weights On the incident;

[0064] High weighted cross-entropy The disordered state includes, but is not limited to, frequently switching between different interfaces, repeatedly looking up words, and requesting hints. The probability of this occurring is... Distributed across multiple high cognitive association weights The event;

[0065] When the state is abnormal score When the output is negative or close to 0, the user's current disordered behavior remains stable or becomes more ordered relative to its recent stable state.

[0066] Abnormal status score Approaching 0 indicates the current weighted interaction entropy. The score is roughly in line with its historical average, indicating that the user's learning status is stable. Abnormal status score A negative value indicates an abnormal score for the current state. A reading below the historical average is a positive sign, indicating that the user has transitioned from a struggling state to a highly focused, single-behavioral "flow" state. At this point, the system should assess the user's condition as good and avoid unnecessary intervention. The "flow" state includes, but is not limited to, transitioning from searching for information to prolonged, continuous reading and answering questions.

[0067] The effectiveness of this embodiment depends on the reasonable setting of the time window T and the historical window length k; the specific details are determined by the expert group using the fuzzy hierarchical analysis method, which will not be elaborated here.

[0068] To demonstrate the technical capabilities of the first computing module, the following specific scenario is set up:

[0069] The user is learning from a German technical document. It defines three key types of discrete interactive events. And assign cognitive association weights to them. The weight range is [0.5, 2], with larger values ​​indicating stronger correlations.

[0070] It's sequential scrolling reading, with a weight W1=0.8, a standard operation; It uses the built-in dictionary, with a weight W2=1.5, so you might encounter unfamiliar words. It is requesting syntax hints, with a weight W3=2.0, and is experiencing difficulty in understanding;

[0071] Compare the performance of two users (User A1: learning smoothly; User B1: encountering difficulties) within two consecutive time windows T=30 seconds:

[0072] Table 1. Implementation example of the first calculation module:

[0073]

[0074] At time window t-1, the standard Shannon entropy of the two users is similar, but the weighted interaction entropy of this invention can distinguish that user A1's weighted interaction entropy (0.80) is slightly more disordered than user B1's weighted interaction entropy (0.54). At time window t, user A1 enters a more focused state, and their behavior is more concentrated on sequential reading of e1, with the weighted interaction entropy significantly decreasing to 0.54. User B1, on the other hand, encounters difficulties, and their behavior is scattered across reading, word lookup, and requesting hints. The probability of high-weighted events e2 and e3 increases, causing the weighted interaction entropy to soar to 1.71. This demonstrates that weighted interaction entropy, compared to traditional entropy, can more sensitively and accurately quantify the degree of behavioral disorder related to cognitive load.

[0075] User A1's abnormal status score A score of -0.48 is a clear negative value, indicating that the state is developing positively, and the system should avoid intervention. This is the abnormal state score for user B1. A score of 0.68 is a significant positive value, clearly indicating that the student's learning state deteriorated rapidly within a short period, reaching the conditions for triggering subsequent cognitive diagnosis. This validates the validity of the abnormal state score. As a dynamic rate of change indicator, it plays a decisive role in accurately determining the timing of intervention.

[0076] In a specific embodiment of the present invention, in order to adapt to the limited precision of the digital processing system and improve computational efficiency, the system further includes a processing step: calculating the state anomaly score. The score is compared with a preset zero-zone threshold of 0.05; if the state is abnormal... If the absolute value of the score is less than the zero-zone threshold, then in subsequent intervention decision-making steps, the score for that state will be adjusted accordingly. The value of is forcibly set to 0. This processing method is a common technique used by those skilled in the art when implementing numerical algorithms. It aims to effectively process parameters that are theoretically infinitely close to zero in engineering, thereby ignoring the minimal fluctuations that have no practical impact on the final result.

[0077] Cognitive Diagnosis Activation Module: Used to activate the cognitive diagnosis process in response to the abnormal state score exceeding a preset diagnostic trigger threshold;

[0078] Further explanation: The step of activating the cognitive diagnosis process within the cognitive diagnosis activation module further includes rules for establishing the diagnostic trigger threshold, the rules including:

[0079] Based on the time series of the user's state anomaly scores over a historical period, the baseline volatility characterizing the stability of the user's individual behavioral patterns is calculated.

[0080] Obtain the material complexity score based on the preset material fragment of the current learning material;

[0081] Based on the baseline volatility and the material complexity score, an adaptive diagnostic trigger threshold that dynamically changes with the user and the material is calculated using a preset linear combination model, and is used as the preset diagnostic trigger threshold.

[0082] In this embodiment, the "diagnostic trigger threshold" is a self-adjusting "adaptive diagnostic trigger threshold." Traditional methods use fixed thresholds, which cannot adapt to the differences between different users and learning materials, easily leading to misjudgments. This solution establishes a dynamic threshold model by introducing two dimensions: "baseline volatility" and "material complexity score." This model can set more sensitive thresholds for users with more stable behavior patterns and more lenient thresholds for users with more volatile behavior patterns; at the same time, it can automatically raise the threshold when faced with more complex learning materials, allowing users more room for "trial and error" and "exploration." This personalized and contextualized threshold establishment rule improves the accuracy and rationality of diagnostic activation and is key to ensuring the intelligence and humanization of the entire system.

[0083] The specific implementation process for establishing the "adaptive diagnostic trigger threshold" is broken down as follows:

[0084] The quantitative explanation of baseline volatility is as follows:

[0085] 2.1) Acquisition of historical data sequences: The system retrieves all historical anomaly scores generated by the current user within a relatively long historical period from the database, forming a time series; in this embodiment, the relatively long historical period includes the most recent N1 learning sessions; "historical anomaly scores" refer to a series of anomaly scores calculated and stored by the "first calculation module" in the past. These are obtained through database query operations.

[0086] 2.2) The baseline volatility is calculated as follows: the standard deviation of the time series of the "historical state anomaly score" is calculated. "Baseline volatility" is a non-negative real number, and its magnitude reflects the degree of dispersion of the user's state anomaly score throughout history. A larger value indicates higher inherent volatility in the user's learning behavior pattern. It is obtained directly by calling the standard deviation calculation function.

[0087] 2.3) The steps for obtaining the material complexity score are as follows:

[0088] 2.31) The first step is to extract complexity indicators: When the system loads learning materials, it performs text analysis on the current learning material segment and extracts multiple text complexity indicators. "Text complexity indicators" are multiple quantitative indicators used to measure the difficulty of text. In this embodiment, the "average sentence length", "low-frequency word density" and "average syntactic tree depth" are calculated by calling the existing text analysis library.

[0089] 2.32) The calculation method for the material complexity score is as follows: multiple "text complexity indicators" are weighted and summed, and then normalized to obtain a value between 0 and 1 within a specific range. The "material complexity score" is a comprehensive difficulty rating. Its weight coefficients are obtained by training a large number of texts and their difficulty labels through expert calibration or machine learning models, ensuring that the contribution of indicators such as "average sentence length" to the final result is reasonable.

[0090] The steps for determining the adaptive diagnostic trigger threshold are as follows:

[0091] 2.4) The adaptive diagnostic trigger threshold is calculated as follows: a preset "baseline threshold" is used as a benchmark, plus a first adjustment term obtained by multiplying the "baseline volatility" by a "volatility adjustment coefficient," and a second adjustment term obtained by multiplying the "material complexity score" by a "complexity adjustment coefficient." The "adaptive diagnostic trigger threshold" is the final dynamic threshold value used for comparison with the real-time "state anomaly score." The "baseline threshold" is a universally applicable threshold initially set by the system; in this embodiment, it is set to 0.5. The "volatility adjustment coefficient" and "complexity adjustment coefficient" are two preset weighted parameters greater than zero, used to control the influence of the two dynamic factors on the threshold. Their values ​​are optimized through statistical analysis of a large amount of user data. This adaptive diagnostic trigger threshold ensures that the "baseline volatility" and "material complexity score" in the input are both positively proportional to the output of the final threshold.

[0092] The "cognitive diagnosis activation module" in this embodiment solves the technical challenge of traditional fixed threshold methods in balancing sensitivity and accuracy when facing individual user differences and varying task difficulty by constructing an adaptive diagnostic trigger threshold model. This model can dynamically adjust to achieve accurate judgment of different users in different learning contexts. The technical effectiveness and logical rationale are explained in detail through the following specific parameter substitution and deduction process.

[0093] 2.5) Mark the "historical state anomaly score" as The "baseline volatility" is identified as... The "material complexity score" is marked as... The "basic threshold" is marked as The "volatility adjustment coefficient" is identified as... The "complexity adjustment factor" is marked as Mark the "Adaptive Diagnostic Trigger Threshold" as Mark the "real-time status anomaly score" as... .

[0094] The adaptive diagnostic trigger threshold The calculation, at its core, is a linear combination model, which originates from basic algebra and incorporates statistical ideas; the baseline volatility... The calculation uses the standard deviation formula, a fundamental method in statistics for measuring the dispersion of data. The adaptive diagnostic trigger threshold... The calculation uses weighted linear summation, which is the basic form of constructing multi-factor influence models in mathematics;

[0095] In this model, baseline volatility Material complexity score These are two independent input variables, providing information to the model from the dimensions of "user history" and "task status," respectively; the basic threshold... As the intercept term of the model, it provides a universally applicable basic trigger level; volatility adjustment coefficient and complexity adjustment coefficient As weighting coefficients, they are positioned in an adjustment phase and determine the baseline volatility. Material complexity score Ultimate adaptive diagnostic trigger threshold The intensity of the impact.

[0096] Adaptive diagnostic trigger threshold Compared with baseline volatility Material complexity score All show a positive proportional relationship. The more unstable the user's historical behavior pattern, i.e., the more volatile the baseline... The more complex the learning material, the higher the complexity score. Increase the calculated adaptive diagnostic trigger threshold The higher the tolerance, the better. The design logic is that the system shows a higher tolerance for users whose behavior patterns are inherently variable, avoiding frequent false triggers of diagnosis due to their inherent behavioral noise; at the same time, it is normal for users to experience temporary behavioral confusion when facing highly difficult materials, so the triggering conditions need to be relaxed.

[0097] Baseline volatility The range of values ​​is When baseline volatility The closer the baseline volatility is to 0, the more consistent and stable the user's historical behavior pattern; when the baseline volatility is close to 0, the more consistent and stable the user's historical behavior pattern is. The larger the value, the stronger the inherent volatility of the user's behavior pattern.

[0098] Through normalized design, the material complexity score The value range is [0,1]. This ensures that complexity metrics from different sources contribute to the model on a uniform scale. When the material complexity score... The closer the complexity score is to 0, the simpler the material; when the material complexity score is close to 0, the simpler the material is. The closer it gets to 1, the higher the material complexity.

[0099] Adaptive diagnostic trigger threshold The range of values ​​is Its value is directly related to the abnormal state score. A comparison was made. A lower adaptive diagnostic trigger threshold was used. A value indicating that the system is in a "high-sensitivity" state means that even slight abnormalities in user behavior will trigger an abnormality score. Even a slight increase can trigger a diagnosis. A higher adaptive diagnostic trigger threshold. A value of 0 indicates that the system is in a "high tolerance" state, requiring significant anomalies in user behavior; this is the state anomaly score. Only a significant increase will trigger a diagnosis.

[0100] 2.6) This embodiment demonstrates the technical effect of the "cognitive diagnosis activation module" by setting a specific scenario and assigning values ​​to the weight parameters: , , The system's decision-making is compared among four users with different characteristics when faced with materials of varying difficulty.

[0101] Table 2. Implementation example of the cognitive diagnosis activation module:

[0102]

[0103] Comparing users A2 and C2, both faced with the same simple learning materials, However, due to the significantly higher volatility of user C2's historical behavior compared to user A2 ( (0.80 and 0.10 respectively), the adaptive diagnostic trigger thresholds calculated by the system. This is also much higher than that of user A2. This means that the system did not trigger a diagnosis when faced with user C2's high state anomaly score (0.70), successfully filtering out its inherent behavioral noise; while for the stable user A2, a moderate state anomaly score (0.60) is enough to be identified as a signal that requires intervention.

[0104] Comparing users A2 and B2, both are stable users, and However, user B2 is faced with complex learning materials. The system automatically raised the adaptive diagnostic trigger threshold for it. The score was 0.72, higher than user A2's 0.51. Therefore, even though user B2 showed a higher abnormality score (0.65), the system judged that it was within an acceptable "struggle range" and avoided unnecessary interference when tackling the problem.

[0105] New user E2, due to limited historical data, At an average level, the system assigns a slightly above-average adaptive diagnostic trigger threshold of 0.75. When its state anomaly score... When the adaptive diagnostic trigger threshold is exceeded, the system will activate the diagnostic function in a timely manner.

[0106] In a specific embodiment of the present invention, in order to adapt to the limited precision of the digital processing system and improve computational efficiency, the system further includes a processing step: calculating the baseline fluctuation. Compare with a preset fluctuation zero-zone threshold of 0.01; if the baseline fluctuation... If the baseline volatility is less than the zero-zone threshold, then in the subsequent adaptive diagnostic trigger threshold calculation, this baseline volatility will be used. The value of is forcibly set to 0. This processing method is a common technique used by those skilled in the art when implementing numerical algorithms. It aims to effectively process parameters that are theoretically infinitely close to zero in engineering, thereby ignoring the minimal fluctuations that have no practical impact on the final result.

[0107] The second calculation module is used to calculate, in the cognitive diagnosis process, a cognitive deviation value that quantifies the degree of user comprehension deviation for the learning material segment corresponding to the time when the diagnostic trigger threshold is triggered, in combination with the eye movement trajectory data.

[0108] To further explain, in the second calculation module, calculating the cognitive bias value includes:

[0109] Natural language processing techniques are used to analyze the learning material fragments to generate semantic core maps that identify their semantic core regions;

[0110] Based on the gaze duration of each gaze point in the eye-tracking data, the corresponding gaze intensity weight is calculated. The gaze intensity weight is used to characterize the user's cognitive engagement at a single gaze point.

[0111] By combining the eye movement trajectory data and the gaze intensity weight corresponding to each gaze point, a weighted eye movement heatmap representing the spatial distribution of the user's cognitive resources is generated.

[0112] The distribution difference between the weighted eye-tracking heatmap and the semantic core map is calculated, and this distribution difference is used as the cognitive bias value.

[0113] Furthermore, the generation of the semantic core map is based on dependency parsing of the learning material fragments to identify the core syntactic components, including but not limited to the subject, predicate, and object, and assigning higher initial weight values ​​to the corresponding areas of these components on the display interface.

[0114] The specific implementation process for calculating the cognitive bias value in the second calculation module is broken down as follows:

[0115] 3.1) The generated content of the semantic core graph is as follows:

[0116] 3.11) The first step is semantic element identification and scoring: The system calls a pre-built natural language processing model to analyze the text of the current learning material segment and identify the core semantic elements; "core semantic elements" refer to words or phrases in the text that carry the main information, such as core nouns, key verbs, logical connectors, etc. These are automatically identified through algorithms such as dependency parsing and keyword extraction.

[0117] Each identified "core semantic element" is assigned a semantic importance score; the "semantic importance score" is a numerical value that quantifies the information content of the element. For example, the subject and predicate have higher scores than the attributive and adverbial modifiers. This score is directly output by the natural language processing model based on syntactic structure rules.

[0118] 3.12) The second step is to construct the semantic core map: The system divides the display interface where the learning material fragment is located into a two-dimensional grid. For each grid cell, if it covers a "core semantic element", the initial value of the cell is set to the "semantic importance score" of the "core semantic element" it covers; if it does not cover it, the initial value is zero.

[0119] The entire grid is subjected to Gaussian blurring and normalization to generate a probability distribution map, which is the semantic core map distribution. The semantic core map distribution is a two-dimensional probability distribution, where the value of each point represents the semantic importance of that location, and the sum of the values ​​of all points is 1. It is generated through the assignment in 3.11) and the image processing steps in 3.12) above.

[0120] 3.2) The cognitive bias value of the weighted eye-tracking thermogram was determined using the following method:

[0121] 3.21) The first step is to calculate the gaze intensity weight: The system extracts each independent eye gaze point and its corresponding gaze duration from the eye movement trajectory data obtained from the eye tracking device; the "eye gaze point" is the coordinate of the position on the screen where the user's gaze stays for more than a preset time. The "gaze duration" is the duration of this gaze; both parameters are directly provided by the driver of the eye tracking device; and in this embodiment, the preset duration is initially set to 100 milliseconds;

[0122] The calculation method for the gaze intensity weight is as follows: The "gaze duration" is substituted into a preset, monotonically increasing exponential function. This function uses "gaze duration" as the independent variable and "gaze intensity weight" as the dependent variable. Specifically, the "gaze duration" in milliseconds is divided by a baseline duration, then the logarithm is taken, and 1 is added. In this embodiment, the initial baseline duration is set to 200 milliseconds. In this embodiment, the "gaze intensity weight" is a dimensionless positive number, which amplifies the influence of prolonged gaze. This ensures a direct proportional relationship between "gaze duration" and "gaze intensity weight."

[0123] 3.22) The second step is to construct a weighted eye-tracking heatmap: the system also divides the display interface into a two-dimensional grid. The heatmap is generated using a kernel density estimation algorithm. In this embodiment, when performing density estimation, the contribution of each "eye-tracking fixation point" is no longer equal to 1, but rather its corresponding "fixation intensity weight".

[0124] The calculated heatmap is normalized to generate a weighted eye-tracking heatmap distribution. The "weighted eye-tracking heatmap distribution" is a two-dimensional probability distribution. The value of each point in the graph represents the intensity of cognitive resources invested by the user at that location, and the sum of the values ​​of all points is 1.

[0125] 3.23) The third step is to calculate the cognitive bias value: The cognitive bias value is calculated as follows: using the "weighted eye-tracking heatmap distribution" as the first probability distribution and the "semantic core map distribution" as the second probability distribution, the KL divergence of the first probability distribution relative to the second probability distribution is calculated. In this embodiment, the "distribution difference between the weighted eye-tracking heatmap and the semantic core map" is characterized by KL divergence; the result of this KL divergence calculation is the "cognitive bias value". The "cognitive bias value" is a non-negative real number that quantifies the degree of deviation between the user's actual cognitive resource input and the ideal semantic core. It is calculated using the KL divergence formula.

[0126] Further explanation of the implementation of the second calculation module:

[0127] 3.24) The second calculation module innovatively improves the traditional eye-tracking heatmap generation method by introducing gaze intensity weights, enabling the generated weighted eye-tracking heatmap to more accurately represent the spatial distribution of cognitive resources actually invested by the user in the learning material. By quantifying and comparing this spatial distribution of cognitive resources with the semantic core graph distribution of the material itself, this module can accurately calculate the cognitive bias value, thereby identifying deeper problems in the user's understanding, rather than just the surface visual trajectory;

[0128] The semantic importance score is identified as... The semantic core graph distribution is identified as... Mark the eye-tracking fixation point as The duration of gaze is marked as The base duration is marked as The gaze intensity weight is marked as The weighted eye-tracking thermogram distribution is identified as... The cognitive bias value is identified as... ;

[0129] 3.25) For weighted eye-tracking thermograms The construction logic and effect are as follows: the gaze intensity weight The calculation of gaze duration originates from the logarithmic function in mathematics. The logarithmic function is used to represent gaze duration. Convert to gaze intensity weights Its calculation logic is as follows: for fixation duration Divide by a preset base duration Then take its logarithm and add 1.

[0130] gaze intensity weight As a multiplier, it acts directly on each eye fixation point during the kernel density estimation process. Contribution amount. Base duration The settings determine the baseline correspondence between "fixation duration" and "fixation intensity weight". This processing makes fixation duration... The longer the duration, the greater the weight of fixation intensity. It exhibits non-linear growth, thus affecting the generation of weighted eye-tracking thermograms. At the same time, it can significantly amplify the cognitive investment represented by a user's prolonged gaze at a specific location.

[0131] gaze intensity weight With gaze duration There is a positive correlation. This refers to the duration of a user's gaze over a specific area of ​​the learning material. The longer the length, the greater the corresponding gaze intensity weight. The larger the value, the more cognitive resources are invested in that area. This makes the weighted eye-tracking heatmap... It can accurately reflect the skewed distribution of users' "cognitive resources";

[0132] 3.26) Regarding cognitive bias values The calculation logic and effect: the cognitive bias value The calculation of this value originates from the KL divergence (Kullback-Leibler-Divergence) formula in information theory.

[0133] Set cognitive bias value The range of values ​​is .

[0134] When cognitive bias value The closer the output is to 0, the higher the alignment between the user's actual cognitive resource investment and the core semantics of the content. This means that the weighted eye-tracking heatmap... Distribution of semantic core graph The higher the degree of fit, the greater the weighting of gaze intensity in areas where users gaze for extended periods. The higher the value, the more semantically important the region. This indicates that the user not only saw the key information but also engaged in in-depth cognitive processing of it, demonstrating an efficient and thorough understanding process.

[0135] When cognitive bias value As the output value increases, the deviation between the user's actual cognitive resource input and the core semantics of the content becomes greater; this indicates that the weighted eye-tracking heatmap... Distribution of semantic core graph The more significant the difference, the better. In this embodiment, the user will focus on the area for a long time, and the gaze intensity will be weighted accordingly. The higher the numerical value, the more likely the user is to focus on non-core semantic areas (such as complex illustrations or obscure words that are not part of the core meaning), or the shorter the attention duration on core semantic areas. This indicates that the user may be engaging in "pseudo-reading," having scattered attention, or having difficulty understanding key information.

[0136] 3.27) This embodiment demonstrates the technical effect of the second computing module by setting a learning scenario: a user is reading an article containing semantic core regions and non-semantic regions. Assume the semantic core graph of the article is distributed as follows: semantic core region percentage Non-semantic region proportion Set the base duration. The time interval is 100 milliseconds, and the gaze intensity weight is calculated. .

[0137] This embodiment compares cognitive bias values ​​under four typical reading modes:

[0138] Table 3. Implementation example of the second calculation module:

[0139]

[0140] The weighted eye-tracking heatmap distribution ratios for the "semantic core area" and "non-semantic core area" in the table are calculated based on the initial unweighted gaze distribution; the weighted eye-tracking heatmap distribution ratio for the "non-semantic core area" is denoted as... The weighted eye-tracking heatmap distribution ratio of the "semantic core area" is denoted as... ;

[0141] In the initial unweighted gaze distribution of this embodiment, 70% of the unweighted gaze points fall within the semantic core region, and 30% fall within the non-semantic region, combined with gaze intensity weights. The result is obtained after renormalization. For deep understanding scenarios, the weighted contribution of the semantic core region is: The weighted contribution of non-semantic regions is Total weighted contribution is The weighted eye-tracking heatmap distribution ratio of the "semantic core area" is: The weighted eye-tracking heatmap distribution ratio of the "non-semantic core area" is .

[0142] Comparing deep understanding and rapid scanning scenarios, both demonstrate attention to the semantic core region along the visual trajectory (in this embodiment, the original unweighted gaze distribution is concentrated in the semantic core region). However, due to the longer gaze duration in the core region during "deep understanding," the gaze intensity weighting is higher. The higher weighting of the "semantic core area" in the final eye-tracking heatmap (0.875) significantly exceeds the semantic core area's proportion of 0.7, resulting in a cognitive bias value of 0.124. During "rapid scanning," the fixation duration is generally shorter, leading to a higher weighting of fixation intensity. Both are close to 1, making This coincides precisely with the proportion of the semantic core area, causing the cognitive bias value to approach 0. This proves that the "gaze intensity weight" introduced in this invention can effectively distinguish whether the user "truly understands" or "merely scans," thus more accurately reflecting the cognitive state.

[0143] Accurately Identifying "Pseudo-Reading": Comparing "pseudo-reading" and "typical reading" scenarios. Both have the same original unweighted gaze distribution. In "pseudo-reading," although the user focuses their gaze on the semantic core area, their... The numerical values ​​are high, but due to the generally low average fixation duration, in-depth processing was not performed, resulting in a "cognitive bias value" of 0.025, indicating a slight bias. In "typical reading," users have longer fixation duration in the semantic core area, resulting in a "cognitive bias value" of 0.025 as well, but their behavioral patterns are healthier. This demonstrates that even with seemingly similar visual behaviors, this invention can uncover differences in cognitive quality through "fixation intensity weighting."

[0144] Effectively identify "semantic blind spots": In "semantic blind spot" scenarios, users have low fixation time in the semantic core area and high fixation time in non-semantic areas, leading to... The 0.438 is far lower than the 0.7% proportion of the semantic core region, while The value of 0.562 is much higher than the 0.3 for the non-semantic area, resulting in the highest "cognitive bias value" of 0.211. This clearly indicates that the user failed to allocate cognitive resources to the correct areas, thus effectively exposing the user's comprehension blind spots or attention deficit issues.

[0145] In a specific embodiment of the present invention, in order to adapt to the limited precision of the digital processing system and improve computational efficiency, the system further includes a processing step: comparing the calculated cognitive deviation value with a preset deviation zero-zone threshold of 0.005; if the cognitive deviation value meets the condition of being less than the deviation zero-zone threshold, then in the subsequent cognitive diagnosis process, the value of the cognitive deviation value is forcibly set to 0. This processing method is a conventional technique used by those skilled in the art when implementing numerical algorithms, aiming to effectively process parameters that theoretically approach zero in engineering, thereby ignoring minimal fluctuations that have no practical impact on the final result.

[0146] Intervention module: Based on the state abnormality score and the cognitive bias value, it selects and executes the corresponding adaptive intervention operation from a preset intervention strategy library.

[0147] To further explain, the intervention module, which selects and executes intervention operations based on the abnormal state score and the cognitive bias value, includes:

[0148] Construct a two-dimensional decision space with the state abnormality score as the first dimension and the cognitive bias value as the second dimension;

[0149] The two-dimensional decision space is divided into multiple preset intervention regions, each corresponding to a specific adaptive intervention operation;

[0150] Map the current user's state to a user state coordinate point within this space;

[0151] Based on the location of the user's state coordinates in multiple pre-divided intervention areas, a basic intervention type corresponding to that area is determined;

[0152] Based on the relative positional relationship between the user's state coordinates and the preset boundary or center point of the intervention area, an intervention intensity score is calculated to quantify the strength of the intervention.

[0153] By combining the basic intervention type and the intervention intensity score, a final intervention operation with a specific intensity of effect is generated and executed.

[0154] This solution introduces the key technical feature of "intervention intensity score," making intervention no longer a simple action selection, but a dynamic combination of "action type" and "action intensity." The system not only selects the basic intervention type but also calculates the intervention intensity score. For example, with the same highlighting prompt, users newly entering the problem area will receive a soft, low-intensity glow, while users deeply immersed in the problem area will receive a bright, high-intensity highlight. This upgrade from "qualitative" to "quantitative" intervention improves the smoothness of the user experience and the accuracy of intervention measures, and is one of the key innovations of this solution.

[0155] Furthermore, the "user status coordinate point" is compared with the regional boundary thresholds of multiple preset intervention areas. Based on the comparison results, the intervention area into which the "user status coordinate point" falls is determined, and the basic intervention type bound to that area is queried and determined from the intervention strategy library.

[0156] The threshold of the intervention area boundary is dynamically adjusted based on the user's historical learning proficiency level; for users with higher proficiency levels, the boundary threshold is lowered accordingly to improve the sensitivity of the intervention.

[0157] The specific implementation process of the intervention module generating and executing the final intervention operation is broken down as follows:

[0158] 4.1) Generation of user status coordinates and region positioning:

[0159] 4.11) The first step is to generate coordinate points: the system obtains the calculated state anomaly score and cognitive bias value.

[0160] By using a linear mapping or a nonlinear compression function, the two original scores are normalized to obtain the normalized state abnormality score and the normalized cognitive bias value, both of which have a value range of [0,1]. In this embodiment, the Sigmoid function is selected for linear mapping.

[0161] These two normalized scores are used as the horizontal and vertical coordinates in a two-dimensional Cartesian coordinate system to form the user status coordinate points.

[0162] The "user state coordinate point" is a two-dimensional vector that accurately defines the user's current macroscopic state and microscopic issues within a unified decision space. It is generated through the aforementioned normalization and coordinate construction steps; in this embodiment, the macroscopic state represents the degree of anomaly; and the microscopic issues represent the type of deviation.

[0163] 4.12) The second step is to perform area positioning and type selection: The system compares the "user status coordinate point" with the area boundary thresholds of multiple preset intervention areas. The "area boundary threshold" is a constant value set by the system designer for different intervention areas to divide the decision space after statistical analysis of a large amount of user experimental data. Based on the comparison results, the system determines the intervention area in which the "user status coordinate point" falls, and queries and determines the basic intervention type bound to the area from the intervention strategy library. In this embodiment, the basic intervention types include, but are not limited to, "visual highlighting", "keyword interpretation" or "relaxation video push".

[0164] In this embodiment, the specific implementation details of "the regional boundary threshold of the intervention area is dynamically adjusted based on the user's historical learning proficiency level" are as follows:

[0165] The "dynamic adjustment of regional boundary thresholds" is based on a comprehensive evaluation of users' historical learning behavior, generating a quantified "user proficiency score." This score is then used to dynamically calculate the boundary of each intervention region through a pre-defined adjustment model. The core logic is: for users with higher proficiency, the system intervention "trigger" should be more sensitive, meaning the regional boundary threshold should be lowered accordingly to provide subtle guidance at the initial stage of a problem; for users with lower proficiency, a more lenient "basic threshold" should be used to avoid dampening their learning enthusiasm due to frequent interventions.

[0166] The specific rules for determining the region boundary threshold are broken down as follows:

[0167] The system first needs to calculate a user proficiency score that comprehensively reflects the user's historical learning level. This score is calculated through a multi-parameter weighted average process, ensuring the comprehensiveness and accuracy of the assessment.

[0168] The system extracts at least the following three types of basic indicators from the historical database and normalizes them to the [0,1] interval:

[0169] 1. Historical Average Performance Score: The average score a user has achieved across all past learning units' tests or exercises. This is normalized to obtain the normalized performance score.

[0170] 2. Historical Learning Efficiency Score: The average time a user takes to complete historical learning tasks, normalized by comparing it with the standard time. The shorter the time, the higher the efficiency, and the closer the score is to 1. The normalized efficiency score is obtained after normalization.

[0171] 3. Historical Intervention Frequency Score: The frequency with which users trigger the intervention module during their historical learning process. The lower the frequency, the stronger their ability to solve problems independently, and the closer the score is to 1. This score is obtained by taking the reciprocal of the intervention frequency and then normalizing it. The normalized disturbance rejection score is obtained after normalization.

[0172] The user proficiency score is calculated by weighting and summing the three normalized scores mentioned above.

[0173] User proficiency score = (normalized performance score × performance weight) + (normalized efficiency score × efficiency weight) + (normalized immunity score × immunity weight).

[0174] The three weights (performance weight, efficiency weight, and anti-interference weight) are preset constants that sum to 1. In this embodiment, they are set to 0.5, 0.3, and 0.2 respectively to reflect the emphasis on academic performance. The final "user proficiency score" is a value in the range [0,1]. The higher the score, the higher the user proficiency level.

[0175] After obtaining the "user proficiency score", the system will use a linear adjustment model to calculate the final region boundary threshold.

[0176] Step 1: For each region boundary threshold that needs adjustment, the system needs to obtain two preset basic parameters:

[0177] 1. Base threshold: A standard threshold set to characterize novice or average users, determined by the system designer based on statistical analysis of a large amount of user experimental data.

[0178] 2. Maximum adjustment range: This is a constant that defines the maximum amount by which the threshold can be lowered. It ensures that even for the most skilled users, the threshold will not be reduced to an unreasonable level.

[0179] Step 2: The calculation method for the regional boundary threshold is: the base threshold minus the product of "user proficiency score" and "maximum adjustment range".

[0180] Assume that the starting threshold of the horizontal axis of the "low-light guidance zone", that is, the trigger threshold of the abnormal state score, is set to 0.40, and its maximum adjustment range is set to 0.20.

[0181] Scenario 1, New User: A new user's "User Proficiency Score" calculated from historical data is 0.1.

[0182] Its region boundary threshold = 0.40 - (0.1 × 0.20) = 0.40 - 0.02 = 0.38.

[0183] The boundary threshold of this region is very close to the basic threshold of 0.40. The system's intervention in its learning process is kept at a relatively lenient standard, tolerating a relatively large number of minor abnormal states.

[0184] Scenario 2, Skilled User: The "User Skill Score" calculated from the historical data of a skilled user is 0.9.

[0185] Its region boundary threshold = 0.40 - (0.9 × 0.20) = 0.40 - 0.18 = 0.22.

[0186] The boundary threshold for this area has been significantly reduced. This means that the system will intervene as soon as a user's abnormal status score exceeds 0.22. This highly sensitive setting can provide a subtle reminder as soon as a skilled user shows signs of deviating from their optimal learning state, helping them quickly return to an efficient learning state, which meets the needs of expert users for "minor but timely" assistance.

[0187] 4.2) Calculate the intervention intensity score:

[0188] 4.21) Calculate the offset distance: For the located intervention area, the system obtains its preset area starting boundary point; the "area starting boundary point" is the point where the user's state coordinates intersect the area boundary when the user's trajectory line enters the area, representing the "entrance" of the area. It is obtained through geometric calculation or table lookup;

[0189] The Euclidean distance between the current "user state coordinates" and the "region's starting boundary point" is calculated to obtain the state point offset distance. The "state point offset distance" is a non-negative real number that quantifies the "depth" of the user state within the problem area; the larger the distance, the further the user state deviates from the normal area. It is calculated using the spatial distance formula.

[0190] 4.22) The intervention intensity score is calculated as follows: the "state point offset distance" is divided by a preset maximum offset distance for the region to obtain a ratio; if the ratio is greater than 1, it is set to 1; the "maximum offset distance for the region" is the distance from the "region starting boundary point" to the farthest point such as the diagonal or center of the region, representing the maximum possible offset within the region, and is a preset constant. The "intervention intensity score" is a dimensionless value in the interval [0,1], which converts the absolute offset distance into a relative intensity level; it is obtained through the above normalized division operation, ensuring a direct proportional relationship between the "state point offset distance" and the "intervention intensity score".

[0191] 4.31) Dynamically set the names of intervention parameters: Based on the determined "basic intervention type", the system obtains the corresponding adjustable intervention parameter names from the intervention strategy library, including but not limited to "highlight transparency", "explanation card pop-up speed" or "video volume".

[0192] A pre-defined linear transformation model maps the "intervention intensity score" to the specific value of the "intervention parameter name".

[0193] The calculation method for "highlight transparency" is as follows: add the product of the "intervention intensity score" and the intensity adjustment range to the base transparency; both "base transparency" and "intensity adjustment range" are preset parameters used to define the lower limit and dynamic range of the intervention effect. The final parameter value is obtained through the above linear calculation; in this embodiment, the initial base transparency is set to 0.2, and the upper limit represented by the intensity adjustment range is 0.6.

[0194] As the "intervention intensity score" changes from 0 to 1, the "highlight transparency" smoothly increases from "slight glow" represented by 0.2 to "striking brightness" represented by 0.8.

[0195] The calculation method for "explanation card pop-up speed" is as follows:

[0196] In this embodiment, the "speed" in "explanation card pop-up speed" is technically achieved by controlling the "animation duration," meaning the shorter the duration, the faster the speed. The final pop-up animation duration is calculated as: "maximum pop-up duration" minus the product of the "intervention intensity score" and a "duration adjustment range."

[0197] The "maximum pop-up duration" is defined as the gentlest and least disruptive pop-up effect, while the "duration adjustment range" defines the range of variation from the slowest to the fastest. In this embodiment, the initial "maximum pop-up duration" is set to 500 milliseconds, and the "duration adjustment range" is set to 400 milliseconds. This design ensures that when the "intervention intensity score" is 0, the card pops up with a gentle animation of 500 milliseconds; when the "intervention intensity score" is 1, the animation duration is shortened to 100 milliseconds, achieving a fast and direct presentation to address scenarios where users urgently need help.

[0198] The calculation method for "video volume" is: "base volume" plus the product of "intervention intensity score" and "volume adjustment range".

[0199] "Base volume" defines the minimum playback volume of intervention videos (such as short videos used for relaxation or emotional buffering) to avoid interference when not needed. "Volume adjustment range" defines the maximum range of volume variation. In this embodiment, the initial "Base volume" is set to 0.3 (i.e., 30% of the maximum volume), and the "Volume adjustment range" is set to 0.5. This design ensures that as the "Intervention Intensity Score" changes from 0 to 1, the "Video Volume" linearly increases from 0.3 to 0.8, accommodating different user states from mild distraction to the need for strong emotional guidance.

[0200] In this embodiment, "video volume" of 0.3 represents background volume; "video volume" of 0.8 represents a volume that is clearly audible but not too loud; these two values ​​are adjusted based on the user's actual usage and will not be elaborated further.

[0201] 4.32) The system calls the user interface control interface, applies the specific parameter values ​​calculated in the previous step to the corresponding intervention operation, and presents them to the user.

[0202] Further explanation of the intervention module:

[0203] The normalized state anomaly score is identified as... The normalized cognitive bias value is identified as... Mark the user's status coordinates as Mark the starting boundary point of the region as The offset distance of the state point is marked as The maximum offset distance in the region is marked as... The intervention intensity score is marked as Set the basic transparency as The intensity adjustment range is marked as Highlight transparency is marked as .

[0204] The state point offset distance The calculation uses the Euclidean distance formula, a fundamental method in analytic geometry for calculating the straight-line distance between two points; in the computational logic chain, the initial boundary point of the region... It is dynamic input, representing the user's real-time status. Region starting boundary point. and the maximum offset distance of the region These are static parameters bound to a specific intervention area, which together constitute the magnitude of the calculated intensity. State point offset distance. This is an intermediate calculation result, directly reflecting the degree to which the user's state deviates from the normal boundary. Intervention intensity score. It is the final normalized output, which will offset the absolute state point by distance. This is converted into relative intensity levels. This structured arrangement ensures that the calculation of intervention intensity can respond to dynamic changes in user status while maintaining a uniform and comparable metric within each intervention area;

[0205] 4.33) Calculation Logic and Trend Analysis: Intervention Intensity Score Distance from state point They exhibit a strictly proportional relationship. When the user's state coordinates... Distance from the starting boundary point of the region into which it enters The farther away, the greater the offset distance of the state point. The higher the score, the higher the calculated intervention intensity score. The larger the value, the further the user's state deviates from the normal trajectory, indicating a deeper problem and thus requiring stronger intervention.

[0206] State point offset distance The range of values ​​is Its physical meaning is the absolute distance the user's state coordinates "drift" within the problem area;

[0207] Intervention intensity score The value range is [0,1]; when the intervention intensity score The closer the output is to 0, the weaker the intervention intensity for the user, and the less significant the intervention effect.

[0208] When the intervention intensity score Approaching 0, representing the offset distance of the state point. Approaching 0. This indicates that the user's state coordinates have just crossed the region boundary and are close to the region's starting boundary point. The overlap is high; at this point, the system determines that the user is only on the "edge" of the problem, so the parameters of the generated intervention operation will be close to the base value, providing a very slight and imperceptible guidance to avoid sudden interruption of the user's learning flow; in this embodiment, the intervention operation is to adjust the highlight transparency; and an intervention intensity score is set. The interval approaching 0 is within the range of (0, 0.15);

[0209] When the intervention intensity score The closer the output is to 1, the stronger the intervention intensity and the more significant the intervention effect on the user;

[0210] Intervention intensity score When it approaches 1, it represents the offset distance of the state point. The maximum offset distance of the region approaching this area This indicates that the user's state coordinates have penetrated deep into the core or farthest reaches of the problem area. Based on this, the system determines that the user is in significant distress and requires assistance. Therefore, the parameters of the generated intervention will be adjusted to maximize its effectiveness; for example, highlights will become extremely prominent, or prompt cards will pop up in a more attractive manner to ensure the user receives clear assistance signals.

[0211] This embodiment sets an intervention intensity score. The intervals approaching 1 are within the range of (0.9, 1);

[0212] 4.34) This embodiment demonstrates the technical effect of the "intervention module," specifically setting a "micro-light guiding area," with the basic intervention type being "visual highlighting," and the adjustable parameter being highlighting transparency. The region is defined in the normalized decision space as follows: and Its region's starting boundary point for Its maximum offset distance in the region The calculated value is 0.424. This embodiment represents the maximum offset distance in the region. It is a calculation from Distance to the diagonal point (0.7, 0.6);

[0213] The method for calculating highlight transparency is as follows: Setting the base transparency Intensity adjustment range .

[0214] Table 4. Implementation examples of the intervention module:

[0215]

[0216] The data in the table above illustrates the gradual nature of the intervention intensity: from state one to state five, as the user's state coordinates move deeper into the problem area, the offset distance of the state point increases. The intervention intensity score increased linearly from 0.022 to 0.424. It also smoothly increased from 0.052 to 1.000. This directly resulted in increased highlight transparency. The transition from 0.231 to a striking 0.800 demonstrates the gradual intervention effect of this invention, avoiding the abrupt jump from "no intervention" to "fixed intensity intervention" in traditional methods.

[0217] The table clearly illustrates the deterministic mapping relationship between the quantitative input of user status and the quantitative output of intervention effect. For example, when the user is near the midpoint of the problem area in state three, the system will provide a "moderate intensity" intervention with an intensity of 0.667 and an opacity of 0.600. This precise quantitative control is the foundation for achieving personalized and adaptive intervention.

[0218] State 6 demonstrates that even if a user exhibits significant issues only in the state anomaly score dimension, while having minor issues in the cognitive bias dimension, the system can still comprehensively assess the overall degree of deviation through Euclidean distance calculation and assign a reasonable intervention intensity score of 0.601. This proves that the proposed method can effectively integrate multi-source evidence to make more comprehensive and robust decisions than single-dimensional threshold judgments.

[0219] In a specific embodiment of the present invention, in order to adapt to the limited precision of the digital processing system and improve computational efficiency, the system further includes a processing step: calculating the intervention intensity score. The intervention intensity score is compared with a preset intensity zero-zone threshold of 0.01; if the intervention intensity score is... If the score is less than the zero-intensity threshold, then the intervention intensity score will be adjusted during the subsequent final intervention operation. The value of is forcibly set to 0. This processing method is a common technique used by those skilled in the art when implementing numerical algorithms. It aims to effectively process parameters that are theoretically infinitely close to zero in engineering, thereby ignoring the extremely small fluctuations that have no practical impact on the final result and avoiding unnecessary minor interventions in the system.

[0220] It should be noted that all calculation formulas in this application employ regression analysis, including but not limited to machine learning algorithms, to deeply analyze the collected parameters and identify their natural trends and interrelationships. Specialized software, such as Python's Scikit-learn library or the R language, is used to automatically generate mathematical models that match the data. Then, cross-validation and other methods are used to objectively evaluate the model performance, and continuous feedback and optimization are combined to ensure that the created formulas truly reflect the inherent laws of the data, thereby guaranteeing their effectiveness and accuracy. In all calculation formulas in this application, the parameters in each formula undergo dimensionless processing within a consistent range to ensure that different physical quantities are compared on the same scale; dimensionless processing techniques include, but are not limited to, min-max-normalization and Z-score standardization.

[0221] The technical solution of this invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random-access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments of this invention.

[0222] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0223] 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 it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A language learning dynamic resource allocation and interaction system based on an internet platform, executed by a computing device, characterized in that: Specifically, it includes: Data acquisition module: used to continuously acquire interactive behavior data that characterizes the user's macro-level learning status, as well as eye movement trajectory data collected by eye-tracking devices when the user reads learning materials; First calculation module: used to calculate, in real time, a state anomaly score representing the stability of user behavior patterns based on the interactive behavior data; The interaction behavior data is processed into a time series; based on the time series, a weighted interaction entropy representing the degree of disorder of the user behavior pattern is calculated; the time change rate of the weighted interaction entropy is used as the state anomaly score. Cognitive Diagnosis Activation Module: Used to activate the cognitive diagnosis process in response to the abnormal state score exceeding a preset diagnostic trigger threshold; The second calculation module is used to calculate, in the cognitive diagnosis process, a cognitive deviation value that quantifies the degree of user comprehension deviation for the learning material segment corresponding to the time when the diagnostic trigger threshold is triggered, in combination with the eye movement trajectory data. The learning material fragments are analyzed to generate semantic core maps that identify their semantic core regions; Based on the gaze duration of each gaze point in the eye-tracking data, the corresponding gaze intensity weight is calculated. The gaze intensity weight is used to characterize the user's cognitive engagement at a single gaze point. By combining the eye movement trajectory data and the gaze intensity weight corresponding to each gaze point, a weighted eye movement heatmap representing the spatial distribution of the user's cognitive resources is generated. Calculate the distribution difference between the weighted eye-tracking heatmap and the semantic core map, and use the distribution difference as the cognitive bias value; Intervention module: Based on the state abnormality score and the cognitive bias value, it selects and executes the corresponding adaptive intervention operation from a preset intervention strategy library.

2. The language learning dynamic resource allocation and interaction system based on an internet platform according to claim 1, characterized in that: The calculation steps for the abnormal status score specifically include: Within a preset time window, the interactive behavior data is identified into multiple discrete interactive event types, and the probability of each discrete interactive event type occurring within the time window is determined. Use "discrete interactive event type" Characterization; the probability of the i-th event occurring is denoted as... ; probability of occurrence It is the ratio of the number of occurrences of the i-th event to the total number of occurrences of all events within this time window; Obtain a preset cognitive association weight that corresponds one-to-one with each discrete interaction event type. The cognitive association weight is used to characterize the degree of correlation between the discrete interaction event type and the user's cognitive load. Based on the probability of occurrence of each discrete interaction event type and its corresponding cognitive association weight, the weighted interaction entropy is calculated as a quantitative representation of the degree of disorder in the user's current behavior pattern. The weighted interaction entropy is calculated by assigning the probability of occurrence of each "discrete interaction event type" to the following values: The corresponding cognitive association weights are multiplied to obtain a weighted probability term; this weighted probability term is then multiplied by the probability of the event occurring. Multiply the logarithmic values ​​again; sum the results calculated for all event types and take their negatives to obtain the weighted cross-entropy; The state anomaly score is determined based on the change of the weighted interaction entropy within a continuous time window; the higher the state anomaly score, the greater the degree to which the user's state deviates from its recent stable state. The discrete interactive event types include: mouse click events, keyboard input events, interface switching events, content scrolling events, and text highlighting events.

3. The language learning dynamic resource allocation and interaction system based on an internet platform according to claim 2, characterized in that: The step of activating the cognitive diagnosis process within the cognitive diagnosis activation module further includes rules for establishing the diagnostic trigger threshold, the rules including: Based on the time series of the user's state anomaly scores over a historical period, the baseline volatility characterizing the stability of the user's individual behavioral patterns is calculated. Obtain the material complexity score based on the preset material fragment of the current learning material; Based on the baseline volatility and the material complexity score, an adaptive diagnostic trigger threshold that dynamically changes with the user and the material is calculated using a preset linear combination model, and is used as the preset diagnostic trigger threshold.

4. The language learning dynamic resource allocation and interaction system based on an Internet platform according to claim 3, characterized in that: The generation of the semantic core map is based on dependency parsing of learning material fragments to identify core syntactic components, including subjects, predicates, and objects, and assigning high initial weight values ​​to the corresponding areas of these components on the display interface.

5. The language learning dynamic resource allocation and interaction system based on an Internet platform according to claim 4, characterized in that: The intervention module, based on the abnormal state score and the cognitive bias value, selects and executes intervention operations, including: Construct a two-dimensional decision space with the state abnormality score as the first dimension and the cognitive bias value as the second dimension; The two-dimensional decision space is divided into multiple preset intervention regions, each corresponding to a specific adaptive intervention operation, including the following: Map the current user's state to a user state coordinate point within this space; Based on the location of the user's state coordinates within a pre-divided set of intervention areas, a basic intervention type corresponding to the intervention area of ​​that location is determined. Based on the relative positional relationship between the user's state coordinates and the preset boundary of the intervention area, an intervention intensity score is calculated to quantify the strength of the intervention. By combining the basic intervention type with the intervention intensity score, the final intervention operation is generated and executed. The "user status coordinate point" is compared with the regional boundary thresholds of multiple preset intervention areas. Based on the comparison results, the intervention area into which the "user status coordinate point" falls is determined and designated as the location intervention area. The basic intervention type bound to the location intervention area is then queried from the intervention strategy library. The threshold of the intervention area's boundary is dynamically adjusted based on the user's historical learning proficiency level.

Images