Online learning energy concentration degree evaluation method combined with formability evaluation

Abstract

The invention provides an online learning energy concentration degree evaluation method combined with formability evaluation, and belongs to the technical field of online education, and the method comprises the steps: capturing eye movement data of a learner in an online learning process, carrying out feature extraction and time sequence analysis, and combining formability evaluation data to construct a deep neural network energy concentration degree evaluation model. And performing real-time dynamic evaluation on the energy concentration state of the learner by using multi-dimensional indexes such as attention continuity, visual concentration degree and cognitive input degree. Through a sliding time window technology and a specially-designed evaluation equation, the score of the energy concentration degree of each learning stage can be accurately calculated, and finally the overall energy concentration level of the learner is comprehensively output, so that data support is provided for personalized learning. According to the method, learning energy concentration degree monitoring and analysis support is provided, and the problem that it is difficult to continuously, objectively and comprehensively monitor the attention state of a learner on the premise that learning activities are not affected in the prior art is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of online education, and in particular, relates to an online learning concentration evaluation method combined with formative evaluation. Background Art

[0002] In the contemporary networked and intelligent educational environment, online learning has become one of the important channels for learners to acquire knowledge. Compared with traditional classroom teaching, online learning has the flexibility of time and space, and can provide learners with personalized learning experience. But at the same time, online learning also faces problems such as distraction and low learning participation, which seriously affect the learning effect. Therefore, how to effectively monitor and evaluate learners' learning concentration has become one of the key issues that need to be solved in the current field of educational technology.

[0003] The existing methods for assessing learning concentration mainly include the following two methods: one is the assessment method that relies on learners' subjective reports, such as questionnaires and interviews. This method has the advantages of simple operation and low implementation cost, but it is difficult to objectively and accurately reflect the actual state of attention due to the influence of learners' own cognitive level and subjective emotions. The other is an assessment method based on physiological signal detection, such as using physiological indicators such as electroencephalogram, heart rate variability, and pupil reaction to infer the learner's cognitive engagement. This method can more objectively assess the level of attention, but it requires professional testing equipment, which causes certain interference to learners' learning activities, and it is also difficult to continuously monitor changes in learners' attention state.

[0004] In summary, the existing methods for assessing learning concentration have the following key problems: 1) subjective assessment methods are difficult to objectively reflect the learner's true attention state; 2) assessment methods based on physiological signals require additional detection equipment and are difficult to monitor continuously; 3) both methods are difficult to fully characterize the temporal dynamic characteristics of the learner's attention state. Therefore, it is urgent to develop a new concentration assessment technology that can continuously, objectively and comprehensively monitor the learner's attention state without affecting learning activities. Summary of the invention

[0005] In view of this, the present invention provides an online learning concentration assessment method combined with formative evaluation, which can solve the problem that the existing technology is difficult to continuously, objectively and comprehensively monitor the learner's attention state without affecting the learning activities.

[0006] The present invention is achieved in that:

[0007] The present invention provides an online learning concentration evaluation method combined with formative evaluation, comprising the following steps:

[0008] S10, obtaining learners’ eye movement data during online learning;

[0009] S20, extracting features from the eye movement data to obtain eye movement feature parameters;

[0010] S30, performing time series analysis on the eye movement feature parameters using a multi-scale sliding time window method to obtain eye movement feature time series data;

[0011] S40, obtaining formative assessment data of the learner and normalizing the formative assessment data, and generating a formative assessment adjustment factor using an adjustment factor empirical equation;

[0012] S50, using the concentration evaluation equation group to calculate the attention duration score, visual concentration score and cognitive engagement score;

[0013] S60, constructing a concentration evaluation model using a deep neural network based on the attention duration score, the visual focus score, and the cognitive engagement score;

[0014] S70, applying the concentration evaluation model to the real-time eye movement feature time series data to obtain a concentration score within each sliding time window;

[0015] S80: Calculate the overall concentration level using a concentration summary equation based on the concentration level scores in each sliding time window, and output the result.

[0016] On the basis of the above technical solution, the online learning concentration evaluation method combined with formative evaluation of the present invention can also be improved as follows:

[0017] Among them, the eye movement data includes pupil size data, blink frequency data, gaze point position data, eye movement speed data, pupil light reaction data, gaze point jump data, binocular coordination data, and eye micro-movement frequency data.

[0018] Furthermore, the eye movement characteristic parameters include pupil size change rate, blink interval duration, gaze point spatial distribution density, eye movement speed change rate, pupil light reaction rate, gaze point jump pattern characteristic value, binocular coordination index, and eye micro-movement frequency change rate.

[0019] Furthermore, the multi-scale sliding time window includes a short-time window, a medium-time window and a long-time window, corresponding to attention features of different time scales.

[0020] Furthermore, the formative evaluation data includes a learning task completion quality score, a learning process participation score, a knowledge mastery degree score, a learning content difficulty coefficient, a knowledge point relevance score, a learning task time management score, a learning strategy use frequency score, and an error correction ability score.

[0021] Furthermore, the deep neural network includes a time series attention mechanism module, an adaptive weight adjustment module and a multimodal feature fusion layer.

[0022] Furthermore, the concentration evaluation equation group includes an attention duration calculation equation, a visual focus calculation equation, and a cognitive engagement calculation equation.

[0023] Furthermore, the attention duration calculation equation inputs the pupil size change rate, blink interval duration, pupil light reaction rate, eye movement frequency change rate and binocular coordination index, and outputs an attention duration score.

[0024] Furthermore, the visual concentration calculation equation inputs the spatial distribution density of the gaze point, the rate of change of eye movement speed, the characteristic value of the gaze point jump pattern, the complexity of the gaze trajectory and the gaze stability index, and outputs a visual concentration score.

[0025] Furthermore, the cognitive engagement calculation equation inputs the attention duration score, visual concentration score, formative assessment adjustment factor, content difficulty coefficient, knowledge relevance and learning strategy score, and outputs a cognitive engagement score.

[0026] The pupil size data is obtained by capturing pupil images with an infrared camera of an eye tracking device, calculating pupil area or diameter using an image processing algorithm, with a sampling frequency of no less than 60 Hz, and performing noise filtering and smoothing on the raw data;

[0027] The blink frequency data is obtained by detecting the continuity interruption of the pupil image, identifying a blink when the pupil is blocked by the eyelid, recording the number of blinks per unit time, and setting the minimum effective blink duration threshold to 50ms to filter out pseudo-blink events;

[0028] The acquisition of the gaze point position data is specifically to calculate the relative position relationship between the pupil center and the corneal reflection point through an eye tracking device, and convert it into a two-dimensional coordinate point on the display screen through a geometric mapping algorithm, with a sampling frequency of not less than 120 Hz, and simultaneously record the timestamp information of the gaze point;

[0029] The acquisition of the eye movement speed data is specifically performed by calculating the displacement difference of the gaze point position at adjacent sampling moments and dividing the difference by the sampling time interval to obtain the instantaneous speed of the eye movement, and performing median filtering on the speed data to eliminate mutation noise;

[0030] The pupil light reaction data is obtained by recording the dynamic change process of pupil size when the light intensity changes, including parameters such as contraction latency, contraction speed, contraction amplitude and recovery time, with a sampling frequency of not less than 100 Hz;

[0031] The acquisition of the gaze point jump data is specifically through identifying the rapid eye movement (saccade) event, when the eye movement speed exceeds 30° / s, it is determined that the jump starts, when the speed is less than 15° / s, it is determined that the jump ends, and the starting point, end point, duration and trajectory characteristics of the jump are recorded;

[0032] The binocular coordination data is obtained by simultaneously tracking the movement state of both eyes, calculating the parallax angle of the left and right eye gaze points, fusion stability and the phase difference of the coordinated movement, with a sampling frequency of not less than 200 Hz, and performing real-time synchronization of the binocular data;

[0033] The acquisition of the eye micro-movement frequency data is specifically to capture the tiny shaking movement of the eye through a high-precision eye tracking device, with a frequency range of 30-100 Hz and an amplitude of less than 1° visual angle, and to extract the micro-movement spectrum characteristics using fast Fourier transform and compensate for environmental vibration noise.

[0034] The steps of dividing the short-term window, the medium-term window and the long-term window specifically include:

[0035] Step 1: The short-term window, medium-term window and long-term window are hierarchically designed based on the sampling frequency of eye movement data and human cognitive characteristics. The short-term window is set to 3 seconds in size, 1 second in step length, and 5 consecutive windows are maintained to capture instantaneous attention fluctuations. The medium-term window is set to 30 seconds in size, 10 seconds in step length, and 6 consecutive windows are maintained to reflect the level of sustained attention. The long-term window is set to 300 seconds in size, 60 seconds in step length, and 4 consecutive windows are maintained to analyze long-term attention trends. The windows of the three time scales need to be aligned and synchronized in time sequence.

[0036] Step 2: Use the real-time data stream processing mechanism to realize the dynamic sliding and updating of the window, and smoothly move the windows of each time scale according to the preset step interval. When the amount of historical data of a certain scale window exceeds the preset retention quantity, the first-in-first-out principle is adopted to remove the earliest window data, and always keep a fixed number of the latest historical windows for subsequent feature extraction and analysis, while ensuring the continuity and integrity of the data during the window sliding process;

[0037] Step 3: Extract features and conduct statistical analysis on the eye movement data in each time window, including calculating characteristic parameters such as pupil size variability, blink frequency, gaze point dispersion, saccade speed, pupil light response, gaze duration, binocular coordination, and micro-saccade frequency. Standardization is used to eliminate the influence of dimensions, and filtering and other methods are used to suppress noise interference to ensure the stability and reliability of feature extraction.

[0038] Step 4: Realize weighted fusion of multi-scale features and set a reasonable weight distribution scheme based on cognitive science research results. The weight of short-term window features is 0.4, which is used to reflect instantaneous attention changes. The weight of medium-term window features is 0.35, which is used to characterize the level of sustained attention. The weight of long-term window features is 0.25, which is used to characterize the overall attention trend. A comprehensive attention feature index is obtained through weighted combination.

[0039] Step 5: Establish a strict data synchronization and alignment mechanism, uniformly calibrate the timestamps of window features at different time scales, handle data asynchrony caused by sampling frequency differences, and use interpolation or extrapolation methods to compensate for possible data delays and losses to ensure the temporal consistency of multi-scale feature fusion;

[0040] Step 6: Optimize the real-time calculation and update strategy, adopt efficient data structures and parallel computing methods to improve processing efficiency, realize real-time feature extraction of continuous data streams, dynamically update the feature parameters of each time window, ensure the accuracy of the calculation results while ensuring real-time requirements, and provide reliable feature input for subsequent concentration evaluation.

[0041] Furthermore, the relevant calculations involved in the present invention are described in detail below.

[0042] 1. Pupil size change rate calculation equation:

[0043]

[0044] In the formula, R p (t) is the pupil size change rate at time t; D p (t) is the pupil diameter at time t; Δt is the sampling time interval, ranging from [0.01, 0.02] s; α 1 is the rate of change adjustment coefficient, with a value range of [0.1, 0.3]; β 1 is the noise suppression coefficient, ranging from [0.05, 0.15]; ∈ 1 is a random error term, which follows a normal distribution N(0,0.01); n is the number of sampling points; is the partial derivative of pupil diameter with respect to time.

[0045] 2. Calculation equation for the spatial distribution density of gaze points:

[0046]

[0047] Where ρ(x,y) is the gaze density at the spatial point (x,y); μ x ,μ y is the mean of the gaze point distribution; σ x ,σ y is the standard deviation in the x and y directions; 1 is the spatial distribution weight coefficient, with a value range of [0.2, 0.4]; F(x, y) is the gaze point distribution correction function; ∈ 2 is a random error term, which follows the normal distribution N(0,0.02).

[0048] 3. Eye movement velocity change rate calculation equation:

[0049]

[0050] Where V(t) is the eye movement velocity at time t; is the velocity component in the x and y directions; 1 ,λ 2 is the acceleration weight coefficient, the value range is [0.1, 0.2]; is the acceleration component in the x and y directions; ∈ 3 is a random error term, which follows the normal distribution N(0,0.015).

[0051] 4. Calculation equation of the characteristic value of the gaze point jump mode:

[0052]

[0053] Where S(t) is the jump characteristic value at time t; Δθ(t) is the jump angle change; Δd(t) is the jump distance; T s (t) is the jump duration; ω 1 ,ω 2 ,ω 3 is the feature weight coefficient, with a value range of [0.2, 0.4]; δ 1 is the angular velocity weight coefficient, with a value range of [0.1, 0.3]; is the angular velocity; ∈ 4 is a random error term, which follows the normal distribution N(0,0.02).

[0054] 5. Binocular coordination index calculation equation:

[0055]

[0056] Where C(t) is the binocular coordination index at time t; ΔP(t) is the parallax angle; φ(t) is the phase difference; V f (t) is the fusion stability; η 1 ,η 2 ,η 3 is the feature weight coefficient, with a value range of [0.2, 0.4]; κ 1 is the acceleration weight coefficient, the value range is [0.1, 0.2]; is the parallax angular acceleration; ∈ 5 is a random error term, which follows the normal distribution H(0,0.01).

[0057] 6. Attention duration calculation equation:

[0058] A(t)=ξ 1 R p (t)+ξ 2 T b (t)+ξ 3 R l (t)+ξ 4 F m (t)+ξ 5 C(t)+∈ 6 ;

[0059] Where A(t) is the attention duration score at time t; R p (t) is the pupil size change rate; T b (t) is the blink interval; R l (t) is the pupil reaction rate to light; F m (t) is the rate of change of eye micro-movement frequency; C(t) is the binocular coordination index; ξ 1 ,ξ 2 ,ξ 3 ,ξ 4 ,ξ 5 is the feature weight coefficient, ranging from [0.15, 0.25]; ∈ 6 is a random error term, which follows the normal distribution N(0,0.02).

[0060] 7. Visual concentration calculation equation:

[0061] V a (t) = μ 1 ρ(x,y)+μ 2 V(t)+μ 3 S(t)+μ 4 C t (t)+μ 5 S t (t)+∈ 7 ;

[0062] Where V a (t) is the visual concentration score at time t; ρ(x, y) is the spatial distribution density of fixation points; V(t) is the rate of change of eye movement speed; S(t) is the characteristic value of fixation point jump pattern; C t (t) is the complexity of gaze trajectory; S t (t) is the gaze stability index; μ 1 ,μ 2 ,μ 3 ,μ 4 ,μ 5 is the feature weight coefficient, ranging from [0.15, 0.25]; ∈ 7 is a random error term, which follows the normal distribution N(0,0.02).

[0063] 8. Cognitive engagement calculation equation:

[0064] E(t)=v 1 A(t)+v 2 V a (t)+v 3 F r (t)+v 4 D c (t)+v 5 R k (t)+v 6 S l (t)+∈ 8 ;

[0065] Where, E(t) is the cognitive engagement score at time t; A(t) is the attention duration score; V a (t) is the visual concentration score; F r (t) is the formative assessment adjustment factor; D c (t) is the content difficulty coefficient; R k (t) is the knowledge relevance; S l (t) rate the learning strategy; v 1 ,v 2 ,v 3 ,v 4 ,v 5 ,v 6 is the feature weight coefficient, ranging from [0.1, 0.2]; ∈ 8 is a random error term, which follows the normal distribution N(0,0.02).

[0066] 9. Multi-scale feature fusion matrix calculation:

[0067]

[0068] Where M fis the fusion feature matrix; F s ,F m ,F l are short, medium and long time window feature vectors respectively; w ij (i,j=1,2,3) is the weight coefficient matrix, and the element value range is [0.1,0.4]; ∈ 9 ,∈ 10 ,∈ 11 is a random error vector, which follows the normal distribution N(0,0.01).

[0069] 10. Summary equation of concentration:

[0070]

[0071] In the formula, C f (t) is the overall concentration score at time t; F s (t i ),F m (t j ),F l (t k ) are the feature scores of short, medium and long time windows respectively; w s ,w m ,w l is the weight coefficient of each time scale; n s ,n m ,n l are the number of windows at each time scale; τ 1 ,τ 2 ,τ 3 is the time scale weight coefficient, ranging from [0.25, 0.4]; ∈ 12 is a random error term, which follows the normal distribution N(0,0.02).

[0072] The construction principles and meanings of these equations are as follows:

[0073] 1. The pupil size change rate equation takes into account the first-order and higher-order change characteristics of the pupil diameter, and introduces partial derivatives to capture fast-changing characteristics, which is suitable for reflecting instantaneous attention changes;

[0074] 2. The spatial distribution density equation of gaze points is based on a two-dimensional Gaussian distribution model, and a correction function is added to deal with the boundary effect, which can effectively describe the spatial characteristics of gaze behavior;

[0075] 3. The eye movement velocity change rate equation comprehensively considers the velocity and acceleration characteristics, uses vector decomposition to improve calculation accuracy, and reliably reflects the dynamic characteristics of eye movement;

[0076] 4. The characteristic value equation of the gaze point jump pattern combines the angle, distance and time characteristics, introduces the angular velocity term to describe the motion characteristics, and comprehensively characterizes the scanning behavior pattern;

[0077] 5. The binocular coordination index equation integrates disparity, phase and stability characteristics, takes into account acceleration changes, and accurately evaluates the degree of binocular coordination;

[0078] 6. The attention duration equation adopts a multi-feature weighted fusion strategy to balance the contribution of various physiological indicators and provide reliable attention assessment results;

[0079] 7. The visual attention equation combines spatial and temporal characteristics, introduces trajectory complexity and stability indicators, and comprehensively evaluates the visual attention level;

[0080] 8. The cognitive engagement equation integrates attention, concentration and learning characteristics, considers the impact of formative assessment, and accurately reflects the degree of cognitive engagement;

[0081] 9. The multi-scale feature fusion matrix realizes the effective combination of features at different time scales, and the weight matrix ensures the rationality of feature extraction;

[0082] 10. The concentration summary equation adopts a hierarchical weighting strategy to balance the characteristics of each time scale and provide a stable and reliable overall evaluation result.

[0083] Furthermore, the overall architecture of the concentration evaluation model adopts a deep neural network design, with multimodal feature fusion as the core, including three key components: time series attention mechanism module, adaptive weight adjustment module and multimodal feature fusion layer. Through the hierarchical feature extraction and fusion process, the accurate evaluation of learners' concentration is realized. The time series attention mechanism module is responsible for processing the time series data of eye movement features collected by multi-scale sliding time windows. The module is composed of a multi-layer bidirectional long short-term memory network (Bi-LSTM). The hidden state dimension of each LSTM unit is 128. The self-attention mechanism is used to perform weighted aggregation of features of different time scales, thereby capturing short-term, medium-term and long-term attention change patterns; at the same time, the module also integrates a residual connection structure to alleviate the gradient vanishing problem in deep network training, and introduces a batch normalization layer and a dropout layer (dropout rate is 0.3) after each LSTM layer to enhance the generalization ability and robustness of the model.

[0084] As the core control center of the model, the adaptive weight adjustment module receives multi-dimensional feedback signals from formative evaluation, including scoring indicators such as the quality of learning task completion, process participation, and knowledge mastery, and dynamically generates feature importance weights through a multi-layer perceptron network. The structure of this module includes three fully connected layers with 256, 128, and 64 neurons respectively. The activation function uses LeakyReLU, and the output weight distribution is normalized using Softmax in the last layer. At the same time, an attention gating mechanism is introduced to adaptively adjust the contribution of each feature according to the characteristics and difficulty of different learning tasks. In order to improve the model's adaptability to individual differences among learners, this module also integrates a parameter update strategy based on meta-learning to achieve rapid adaptation and personalized tuning through a small number of samples.

[0085] The multimodal feature fusion layer is located at the last stage of the network. It is responsible for integrating time series features and formative assessment features to achieve a comprehensive evaluation of concentration. This layer adopts a feature fusion mechanism based on Graph Attention Network to construct feature representations of different modalities into a feature graph, and learns the relationship between feature nodes through graph attention operations. The output of the fusion layer is reduced in dimension through two fully connected layers (with dimensions of 128 and 64, respectively), and finally a fully connected layer with an output dimension of 1 is used to generate a concentration score. The mean square error loss function is used in the training process, and L1 and L2 regularization constraints (regularization coefficients are 0.001 and 0.0005, respectively) are combined to prevent overfitting. In addition, in order to improve the interpretability of the model, a feature importance visualization module is added to the fusion layer, which can output a distribution diagram of the contribution of different features to the final score, providing educators with an intuitive analysis basis.

[0086] Furthermore, the steps of obtaining the training data set for the concentration evaluation model are described in detail below:

[0087] During the data collection phase, we first need to build a standardized online learning experimental environment, including high-precision eye tracking equipment (sampling rate not less than 200Hz), standard lighting conditions (illuminance maintained at 300-500 lux) and an interference-free learning space. At the same time, we need to develop a supporting learning content management system to present learning materials and record learning process data. During the experiment, we need to recruit volunteers with different learning backgrounds, age groups and cognitive levels to participate in data collection. Each participant needs to complete multiple learning tasks with increasing difficulty. Task types include text reading comprehension, video course learning, interactive problem solving, etc. The duration of each task ranges from 15 to 45 minutes. At the same time, participants are required to report their attention status through the think-aloud method during the task, and experienced educational researchers will observe and record on-site and give real-time scores.

[0088] In the data preprocessing and labeling stage, the collected raw eye movement data are first cleaned and preprocessed, including outlier detection and elimination, data smoothing, time synchronization calibration, etc., to ensure that the data quality meets the modeling requirements; then multi-dimensional data labeling is carried out, and a trained professional labeling team conducts fine-grained labeling of the learner status in each time window. The labeling content includes attention level (1-5 points), cognitive involvement (1-5 points), visual focus (1-5 points), etc. At the same time, the final comprehensive score is determined through the hierarchical analysis method based on the learner's self-report, the on-site scoring of educational experts and the objective learning effect test results; to ensure the quality of labeling, a two-person cross-labeling method is adopted, and third-party experts conduct random inspections and arbitration. The consistency coefficient of the labeling results must reach 0.85 or above before they can be included in the training set.

[0089] During the dataset construction and division stage, the processed data was randomly divided into training set, validation set and test set in a ratio of 8:1:1, while ensuring that the data distribution of learning tasks of different difficulty levels and different learner groups remained consistent in each dataset; in order to enhance the generalization ability of the model, data enhancement technology was used to expand the training samples, including time window sliding sampling, adding random noise, feature combination resampling and other methods; at the same time, different types of attention state samples were balanced, and a stratified sampling strategy was used to ensure that the number of samples in various states was relatively balanced. For special states with a small number of samples (such as significant fluctuations in instantaneous attention), oversampling technology was used to supplement them; the training dataset finally constructed contained learning process data of no less than 1,000 people, with a cumulative learning time of more than 500 hours, covering real cases of various learning scenarios and attention states.

[0090] Compared with the prior art, the beneficial effect of the online learning concentration assessment method combined with formative evaluation provided by the present invention is: using eye tracking technology, comprehensively collecting multi-dimensional physiological data such as pupil size, blinking frequency, fixation point position, eye movement speed, pupil light reaction, fixation point jump, binocular coordination, and eye micro-movement frequency of learners during online learning. Through in-depth analysis and feature extraction of these eye movement data, the learner's attention state can be quantitatively described, including instantaneous attention changes, sustained attention levels, and long-term attention trends.

[0091] In order to further improve the accuracy and reliability of the evaluation, the present invention also incorporates formative evaluation information of learners, such as the quality of learning task completion, participation in the learning process, knowledge mastery, etc. Based on these subjective evaluation indicators, a formative evaluation adjustment factor is constructed in the invention to adjust the contribution weight of eye movement characteristics to concentration, making up for the limitations of single physiological data evaluation.

[0092] In addition, the present invention adopts a multi-scale sliding time window analysis method, which can more comprehensively capture the dynamic change characteristics of learners' attention state. At the same time, the invention also constructs a concentration evaluation model based on a deep neural network, which can adaptively adjust feature weights and improve the accuracy and generalization ability of the evaluation.

[0093] Compared with the prior art, the present invention has the following significant advantages: 1) It adopts non-invasive eye tracking technology, which will not interfere with learning activities and can continuously monitor the learner's attention state; 2) It integrates subjective formative evaluation information to make up for the limitations of single physiological data evaluation and improve the accuracy and credibility of the evaluation; 3) It adopts a multi-scale time series analysis method to comprehensively characterize the dynamic change characteristics of the attention state; 4) The adaptive evaluation model based on deep learning enhances the generalization performance of the method. In general, the present invention can provide reliable learning concentration monitoring and analysis support for intelligent online education systems, solving the problem that the prior art is difficult to continuously, objectively and comprehensively monitor the learner's attention state without affecting learning activities. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] Figure 1 A flow chart of the method provided by the present invention. DETAILED DESCRIPTION

[0095] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0096] like Figure 1 As shown, it is a flow chart of an online learning concentration assessment method combined with formative evaluation provided by the present invention. The specific implementation method of each step is described in detail below:

[0097] The specific implementation of step S10 is as follows:

[0098] First, the pupil image of the learner during online learning is captured by the infrared camera of the eye tracking device. Then the pupil area or diameter D is calculated using the image processing algorithm. p (t), the sampling frequency is not less than 60 Hz. On this basis, the original data is subjected to noise filtering and smoothing.

[0099] Next, blink events are identified by detecting the continuity interruption of the pupil image. When the pupil is blocked by the eyelid, it is considered a blink. The system records the number of blinks per unit time N b , and set the minimum effective blink duration threshold The timeout is 50 milliseconds, which is used to filter out possible false blink events.

[0100] For the acquisition of gaze point position data, the system calculates the relative position relationship between the pupil center and the corneal reflection point, and converts it into a two-dimensional coordinate point (x, y) on the display screen through a geometric mapping algorithm. The sampling frequency is not less than 120 Hz, and the timestamp information t of each gaze point is recorded.

[0101] The eye movement velocity data is obtained by calculating the displacement difference Δx, Δy of the gaze point position at adjacent sampling moments and dividing it by the sampling time interval Δt to obtain the instantaneous eye movement velocity V. x (t),V y (t). At the same time, median filtering is performed on the velocity data to eliminate possible mutation noise.

[0102] In order to obtain pupil light response data, the system records the pupil size D when the light intensity changes. p (t) dynamic change process. Including the contraction latency T l , shrinkage rate R c , contraction amplitude ΔD p and recovery time T r The sampling frequency is not less than 100 Hz.

[0103] The data of gaze point jump is obtained by identifying rapid eye movement events. When the eye movement speed V(t) exceeds 30 degrees per second, it is judged as the start of the jump, and when the speed is less than 15 degrees per second, it is judged as the end of the jump. The system will record the starting point of the jump end Duration T s and trajectory characteristics

[0104] To obtain binocular coordination data, the system needs to track the movement of the left and right eyes at the same time. Calculate the parallax angle ΔP(t) of the left and right eye gaze points and the fusion stability V f The phase difference φ(t) of the coordinated movement is obtained. The sampling frequency is not less than 200 Hz, and the left and right eye data are synchronized in real time.

[0105] Finally, high-precision eye tracking equipment can capture the tiny jitter movements of the eyeball. The micro-motion frequency characteristics between 30 and 100 Hz are extracted by fast Fourier transform. m (t). At the same time, it is necessary to compensate for environmental vibration noise to ensure the reliability of eye movement frequency data.

[0106] In general, the purpose of step S10 is to comprehensively obtain various eye movement data of learners during online learning, laying a foundation for subsequent feature extraction and concentration evaluation. These data include pupil size D p (t), blink frequency N b, gaze point position (x, y), eye movement speed V x (t),V y (t), pupil light response parameter T l ,R c ,ΔD p ,T r , fixation jump features Binocular coordination index ΔP(t),V f ,φ(t) and eye movement frequency F m (t) etc.

[0107] The specific implementation of step S20 is as follows:

[0108] First, feature extraction is performed on various eye movement data obtained in step S10. The extracted eye movement feature parameters include: pupil size change rate R p (t), blink interval length T b (t), the spatial distribution density of the gaze point ρ(x,y), the rate of change of eye movement speed V(t), and the pupil's reaction rate to light R l (t), the characteristic value of the gaze point jump pattern S(t), the binocular coordination index C(t) and the eye movement frequency change rate F m (t).

[0109] The pupil size change rate R p The calculation formula of (t) is:

[0110]

[0111] Where, Δt is the sampling time interval, ranging from [0.01, 0.02] seconds; α 1 is the rate of change adjustment coefficient, with a value range of [0.1, 0.3]; β 1 is the noise suppression coefficient, ranging from [0.05, 0.15]; ∈ 1 is a random error term, which follows a normal distribution N(0,0.01); n is the number of sampling points; is the partial derivative of pupil diameter with respect to time. This formula takes into account the first-order and higher-order changes in pupil diameter, introduces partial derivatives to capture fast-changing features, and can reflect changes in instantaneous attention.

[0112] The calculation formula of the spatial distribution density of the gaze point ρ(x,y) is:

[0113]

[0114] In the formula, μ x ,μ y is the mean of the gaze point distribution; σ x ,σ yis the standard deviation in the x and y directions; 1 is the spatial distribution weight coefficient, with a value range of [0.2, 0.4]; F(x, y) is the gaze point distribution correction function; ∈ 2 is a random error term, which follows a normal distribution N(0,0.02). This formula is based on a two-dimensional Gaussian distribution model, and a correction function is added to deal with the boundary effect, which can effectively describe the spatial characteristics of gaze behavior.

[0115] The calculation formula of eye movement velocity change rate V(t) is:

[0116]

[0117] In the formula, is the velocity component in the x and y directions; 1 ,λ 2 is the acceleration weight coefficient, the value range is [0.1, 0.2]; is the acceleration component in the x and y directions; ∈ 3 is a random error term, which follows a normal distribution

[0118] N(0,0.015). This formula takes into account the characteristics of velocity and acceleration, uses vector decomposition to improve calculation accuracy, and reliably reflects the dynamic characteristics of eye movement.

[0119] The calculation formula of the gaze point jump mode characteristic value S(t) is:

[0120]

[0121] Where Δθ(t) is the jump angle change; Δd(t) is the jump distance; T s (t) is the jump duration; ω 1 ,ω 2 ,ω 3 is the feature weight coefficient, with a value range of [0.2, 0.4]; δ 1 is the angular velocity weight coefficient, with a value range of [0.1, 0.3]; is the angular velocity; ∈ 4 is a random error term, which follows a normal distribution N(0,0.02). The formula combines the angle, distance and time characteristics, introduces the angular velocity term to describe the motion characteristics, and comprehensively characterizes the scanning behavior pattern.

[0122] The calculation formula of binocular coordination index C(t) is:

[0123]

[0124] Where ΔP(t) is the parallax angle; φ(t) is the phase difference; V f (t) is the fusion stability; η 1,η 2 ,η 3 is the feature weight coefficient, with a value range of [0.2, 0.4]; κ 1 is the acceleration weight coefficient, the value range is [0.1, 0.2]; is the parallax angular acceleration; ∈ 5 is a random error term, which follows a normal distribution N(0,0.01). The formula integrates the characteristics of disparity, phase and stability, takes acceleration changes into consideration, and accurately evaluates the degree of binocular coordination.

[0125] In general, the purpose of step S20 is to extract and analyze the original eye movement data obtained in step S10 to obtain a series of characteristic parameters that can quantify the learner's attention state. These characteristic parameters provide reliable data support for the subsequent concentration evaluation.

[0126] The specific implementation of step S30 is as follows:

[0127] The multi-scale sliding time window method is used to perform time series analysis on the eye movement feature parameters extracted in step S20. Among them, the short-term window is set to 3 seconds in size and 1 second in step length, and 5 consecutive windows are maintained to capture the fluctuation characteristics of instantaneous attention. The medium-term window is set to 30 seconds in size and 10 seconds in step length, and 6 consecutive windows are maintained to reflect the level of sustained attention. The long-term window is set to 300 seconds in size and 60 seconds in step length, and 4 consecutive windows are maintained to analyze the trend changes of long-term attention.

[0128] The sliding windows of these three different time scales need to be aligned and synchronized in time series. The system uses a real-time data stream processing mechanism to smoothly move the windows of each time scale according to the preset step interval. When the amount of historical data in a certain scale window exceeds the preset retention number n s ,n m ,n l When performing feature analysis, the first-in-first-out principle is used to remove the earliest window data, and a fixed number of the latest historical windows are always kept for subsequent feature analysis.

[0129] For the eye movement data in each time window, the system will perform feature extraction and statistical analysis, including calculating the pupil size variability R p (t), blink frequency N b , gaze point dispersion ρ(x,y), scanning speed V(t), pupil light reaction parameter R l (t), fixation duration T s , binocular coordination index C(t) and micro-scanning video frequency F m On this basis, standardization is used to eliminate the dimension effect, and filtering and other methods are used to suppress noise interference to ensure the stability and reliability of feature extraction.

[0130] Finally, the system realizes the weighted fusion of multi-scale features. According to the research results of cognitive science, a reasonable weight distribution scheme is set. Among them, the short-term window feature weight w s is 0.4, which is used to reflect the instantaneous attention change; the mid-time window feature weight w m is 0.35, which is used to characterize the level of sustained attention; the long-term window feature weight w l is 0.25, which is used to characterize the overall attention trend. The comprehensive attention feature index M is obtained through weighted combination f :

[0131]

[0132] At the same time, the system has established a strict data synchronization and alignment mechanism. The timestamps of window features at different time scales are uniformly calibrated to handle the data asynchrony problem caused by sampling frequency differences. Interpolation or extrapolation methods are used to compensate for possible data delays and losses to ensure the temporal consistency of multi-scale feature fusion.

[0133] In general, the purpose of step S30 is to use a multi-scale sliding window method to conduct an in-depth time series analysis of the eye movement feature parameters extracted in step S20. Through the comprehensive evaluation of instantaneous, continuous and long-term attention features, the changes in the learner's attention state can be more comprehensively described. At the same time, the system has also established a strict data synchronization and alignment mechanism to ensure the reliability of multi-scale feature fusion and provide stable and reliable data input for subsequent concentration evaluation.

[0134] The specific implementation of step S40 is as follows:

[0135] First, the system obtains the formative evaluation data of learners. These data include: learning task completion quality score Q, learning process participation score P, knowledge mastery score K, learning content difficulty coefficient D c , knowledge point relevance score R k , learning task time management score T, learning strategy use frequency score S l and the error correction ability score E.

[0136] Then, the system normalizes the above formative assessment data to eliminate the dimension effect. Next, the formative assessment adjustment factor F is generated using the adjustment factor empirical equation. r :

[0137] F r (t) = γ 2 Q+γ 3 P+γ 4 K+γ 5 Dc +γ 6 R k +γ 7 T+γ 8 S l +γ 9 E+∈ 13 ;

[0138] In the formula, γ 2 ,γ 3 ,…,γ 9 is the weight coefficient of each evaluation index, with a value range of [0.1, 0.2]; ∈ 13 is a random error term, which follows a normal distribution N(0,0.02).This equation comprehensively considers the influence of various formative assessment indicators on learners' concentration, and obtains a reasonable weight distribution through experimental research and statistical analysis.

[0139] The introduction of formative assessment adjustment factors is to better integrate the subjective evaluation information in the learning process and make up for the limitations of relying solely on eye movement data. It can reflect the impact of factors such as the difficulty of learning tasks, knowledge relevance, and learning strategies on the state of attention.

[0140] In general, the purpose of step S40 is to obtain the learner's formative assessment data and generate a formative assessment adjustment factor F according to its impact on the attention state. r This provides important auxiliary information for the subsequent concentration assessment and helps to more accurately characterize the learner's cognitive engagement state.

[0141] The specific implementation of step S50 is as follows:

[0142] The concentration evaluation equations are used to calculate the attention duration score A(t) and the visual concentration score V a (t) and cognitive engagement score E(t).

[0143] The calculation formula for attention duration is:

[0144] A(t)=ξ 1 R p (t)+ξ 2 T b (t)+ξ 3 R l (t)+ξ 4 F m (t)+ξ 5 C(t)+∈ 6 ;

[0145] In the formula, ξ 1 ,ξ 2 ,ξ 3 ,ξ4 ,ξ 5 is the feature weight coefficient, ranging from [0.15, 0.25]; ∈ 6 is a random error term, which follows a normal distribution N(0,0.02). This formula comprehensively considers the pupil size change rate R p (t), blink interval length T b (t), pupil light reaction rate R l (t), eye movement frequency change rate F m The features of the learner’s sustained attention are obtained by weighted fusion, namely, the sustained attention score, which can reflect the learner’s sustained attention level.

[0146] The formula for calculating visual concentration is:

[0147] V a (t) = μ 1 ρ(x,y)+μ 2 V(t)+μ 3 S(t)+μ 4 C t (t)+μ 5 S t (t)+∈ 7 ;

[0148] In the formula, μ 1 ,μ 2 ,μ 3 ,μ 4 ,μ 5 is the feature weight coefficient, ranging from [0.15, 0.25]; ∈ 7 is a random error term, which follows a normal distribution N(0,0.02). This formula combines the spatial distribution density of the fixation point ρ(x,y), the rate of change of eye movement speed V(t), the characteristic value of the fixation point jump pattern S(t), and the complexity of the fixation trajectory C t (t) and the gaze stability index S t (t) and other features to comprehensively evaluate the learners' visual attention state.

[0149] The formula for calculating cognitive engagement is:

[0150] E(t)=v 1 A(t)+v 2 V a (t)+v 3 F r (t)+v 4 D c (t)+ν 5 R k (t)+v 6 S l (t)+∈8 ;

[0151] Where, ν 1 ,ν 2 ,ν 3 ,v 4 ,v 5 ,ν 6 is the feature weight coefficient, ranging from [0.1, 0.2]; ∈ 8 is a random error term, which follows a normal distribution N(0,0.02). This formula is effective in the attention duration score A(t) and the visual concentration score V a (t), further integrates the formative assessment adjustment factor F r (t), content difficulty coefficient D c (t), knowledge relevance R k (t) and learning strategy score S l (t) and other information can accurately reflect the learners' cognitive involvement.

[0152] The coefficients involved in these three calculation formulas, such as the weight coefficient ξ i ,μ j ,v k Etc., all require a large amount of experimental data verification and statistical analysis to obtain a reasonable value range to ensure the reliability and applicability of the algorithm.

[0153] In general, the purpose of step S50 is to use the concentration evaluation equation group to calculate the attention duration score A(t) and the visual concentration score V according to the eye movement feature parameters extracted in step S20 and the formative evaluation data obtained in step S40. a (t) and cognitive engagement score E(t). These three indicators comprehensively describe the learner's attention state and provide reliable basic data for the subsequent comprehensive evaluation of concentration.

[0154] The specific implementation of step S60 is as follows:

[0155] First, the system uses a deep neural network to build a concentration evaluation model. The model consists of three key modules: time series attention mechanism module, adaptive weight adjustment module and multimodal feature fusion layer.

[0156] The function of the time series attention mechanism module is to capture the multi-scale eye movement feature time series data M extracted in step S30 using the recurrent neural network and the attention mechanism. f This helps to accurately characterize the instantaneous changes, continuous trends, and long-term evolution of attention states.

[0157] The adaptive weight adjustment module automatically adjusts the contribution weight w of each feature according to the feature importance in different learning situations. ij For example, in simple tasks, attention span may be more critical, while in complex tasks, visual focus may play a more important role. This module can dynamically adapt to changes in learning tasks and improve the generalization performance of the evaluation model.

[0158] The multimodal feature fusion layer integrates the eye movement features extracted in step S20 and the formative assessment data F obtained in step S40. r ,D c ,R k ,S l ,The association rules between features are automatically learned through deep learning.,This can give full play to the complementary advantages of multi-source information and obtain more ,comprehensive and accurate results of concentration evaluation.

[0159] The training of the entire deep neural network model requires a large amount of labeled sample data, including eye movement data M f , formative assessment data and corresponding concentration standards C f Through supervised learning, the model can automatically optimize the parameters of each layer and improve the accuracy and generalization ability of the evaluation.

[0160] In general, the purpose of step S60 is to construct a deep learning-based concentration evaluation model. The model integrates time series analysis, adaptive weight adjustment, and multi-source feature fusion techniques, and can make full use of eye movement data and formative evaluation information to give reliable concentration evaluation results.

[0161] The specific implementation of step S70 is as follows

[0162] The concentration evaluation model constructed in step S60 is applied to the real-time eye movement feature time series data M obtained in step S30. f For each feature data in the sliding time window, the model will output the corresponding concentration score C f (t i ),C f (t j ),C f (t k ).

[0163] This process requires a strict data synchronization and alignment mechanism to ensure that the eye movement feature data is completely consistent with the evaluation model input to avoid erroneous evaluations due to timestamp offsets. At the same time, the system also needs to adopt efficient parallel computing methods to ensure real-time processing of continuous data streams to meet the timeliness requirements of application scenarios.

[0164] By scoring the concentration level in each time window, we can deeply analyze the dynamic changes of learners' attention state. For example, we can find that learners have higher concentration in certain periods, while their attention is easily distracted in other periods. These detailed information can help teachers or systems better understand learners' cognitive input and adopt targeted teaching strategies.

[0165] In general, the purpose of step S70 is to apply the concentration evaluation model constructed in step S60 to the real-time eye movement data and output the concentration score C in each sliding time window. f (t i ),C f (t j ),C f (t k ). These scoring results provide reliable basic data for the subsequent summary of overall concentration.

[0166] The specific implementation of step S80 is as follows:

[0167] According to the concentration score C in each sliding time window obtained in step S70 f (t i ),C f (t j ),C f (t k ), using the concentration summary equation to calculate the overall concentration score C f (t).

[0168] The summary equation adopts a hierarchical weighted strategy, comprehensively considering the contributions of short-term window, medium-term window and long-term window features:

[0169]

[0170] In the formula, τ 1 ,τ 2 ,τ 3 is the weight coefficient of each time scale, ranging from [0.25, 0.4]; n s ,n m ,n l are the number of short, medium and long time windows respectively; w s ,w m ,w l is the weight coefficient corresponding to the time scale; ∈ 12 is a random error term, which follows the normal distribution N(0,0.02).

[0171] By weighted combination of the characteristic scores of each time scale, a stable and reliable overall concentration evaluation result C can be obtained. f(t). This hierarchical weighting strategy can balance the contributions of instantaneous, sustained, and long-term attention features and provide a more comprehensive evaluation.

[0172] Finally, the system will calculate the overall concentration score C f (t) Output: Provides teachers or systems with a comprehensive assessment of learners’ attention status. This information helps to timely identify learners’ inattention problems and take targeted intervention measures to improve learning outcomes.

[0173] In general, the purpose of step S80 is to calculate the overall concentration score C using the concentration summary equation based on the concentration scores of each time window obtained in step S70. f (t) and output it. This result provides the final quantitative result for the complete online learning concentration evaluation.

[0174] In general, this invention uses advanced eye movement signal processing and deep learning technology to realize an online learning concentration assessment method combined with formative evaluation. It constructs a concentration assessment model by comprehensively acquiring learners' eye movement data during online learning, extracting characteristic parameters that can quantify the state of attention, and integrating formative evaluation information. The model can output the concentration scores of each time scale in real time, and finally comprehensively calculate the learner's overall concentration. These evaluation results provide teachers or systems with an objective analysis of the state of attention, which helps to adopt targeted teaching strategies and improve learning outcomes.

[0175] Specifically, the principle of the present invention is to use eye tracking technology to comprehensively obtain learners' physiological data during online learning, and integrate formative evaluation information to build a concentration evaluation model. This method can effectively solve the problems existing in the existing technology, mainly in the following aspects:

[0176] First, eye movement data can objectively reflect the learner's attention state. A large number of cognitive neuroscience studies have shown that human visual attention is closely related to eye movement. Eye movement indicators such as pupil size, blinking frequency, fixation point position, and eye movement speed can reflect the learner's cognitive processing and attention level. Compared with physiological signals such as electroencephalogram and heart rate variability, eye movement data is easy to collect and does not interfere with learning activities. It is very suitable for the assessment of concentration in online learning environments.

[0177] Secondly, the present invention integrates formative assessment information, which can make up for the limitations of assessment based solely on eye movement data. Formative assessment includes multiple dimensions such as the quality of learning tasks, the degree of knowledge mastery, and the use of learning strategies, which can more comprehensively reflect the learning status of learners. By constructing formative assessment adjustment factors and combining these subjective evaluation indicators with eye movement characteristics, the learners' cognitive engagement can be more accurately portrayed, and the reliability of the assessment can be improved.

[0178] In addition, the present invention adopts a multi-scale sliding time window analysis method, which can more comprehensively capture the dynamic characteristics of learners' attention state. Through the comprehensive evaluation of instantaneous, continuous and long-term attention characteristics, the temporal change law of learners' attention can be deeply analyzed, providing teachers or intelligent systems with richer feedback information.

[0179] Finally, the adaptive assessment model based on deep learning further enhances the performance of the proposed method. The model can autonomously adjust feature weights according to different learning situations, which improves the accuracy of the assessment while also enhancing the generalization ability of the method. Compared with traditional linear weighted models, deep neural networks have stronger nonlinear fitting capabilities and can better capture the complex relationship between eye movement data and formative assessment.

[0180] In summary, the concentration evaluation method proposed in this invention makes full use of the advantages of eye tracking technology, integrates formative evaluation information, and adopts multi-scale time series analysis and deep learning technology, which can effectively solve the shortcomings of existing technologies in terms of objectivity, sustainability and comprehensiveness. This method provides a reliable learning energy monitoring method for intelligent online education systems, and has important application value for improving learning effects.

[0181] The following is an example of a specific application scenario of the present invention: In order to better monitor students' learning concentration and improve the teaching quality of the course, teachers of a certain institution decide to adopt the online learning concentration assessment method combined with formative evaluation proposed by the present invention.

[0182] First, the course team equipped students with eye tracking equipment to record their eye movement data while they were learning online. The eye movement data collection process is as follows:

[0183] The pupil images of students during the learning process are captured by the infrared camera of the eye tracking device at a sampling frequency of 60 Hz. The pupil diameter D is calculated using the image processing algorithm. p (t), and performs noise filtering and smoothing on the raw data. At the same time, the system also monitors the students’ blinking events and records the number of blinks per unit time N. b , and set the minimum effective blink duration threshold to 50 milliseconds to filter out possible pseudo-blinks.

[0184] For the acquisition of gaze point position data, the system calculates the relative position relationship between the pupil center and the corneal reflection point, and converts it into two-dimensional coordinates (x, y) on the display screen through geometric mapping. The sampling frequency reaches 120 Hz, and the timestamp information t of each gaze point is recorded.

[0185] The eye movement velocity data is obtained by calculating the displacement difference of the gaze point position at adjacent sampling moments and dividing it by the sampling time interval to obtain the instantaneous eye movement velocity V. x (t),V y (t). Then the velocity data is median filtered to eliminate mutation noise.

[0186] In addition, the system also records the dynamic changes of students' pupil response to light, including the contraction latency T l , shrinkage rate R c , contraction amplitude ΔD p and recovery time T r The sampling frequency is 100 Hz.

[0187] For gaze point jump data, the system identifies rapid saccadic behavior by detecting events where the eye movement speed V(t) exceeds 30 degrees per second and records the starting point of the jump. end Duration T s and trajectory characteristics When the speed is lower than 15 degrees per second, the jump is considered complete.

[0188] The acquisition of students' binocular coordination data includes disparity angle ΔP(t), fusion stability V f The system sampling frequency reaches 200 Hz, and the left and right eye data are synchronized in real time.

[0189] Finally, the high-precision eye tracking device can also capture the tiny shaking movements of the students' eyes, and extract the micro-movement frequency characteristics between 30 and 100 Hz through fast Fourier transform. m (t), and compensate for environmental noise.

[0190] Through the above methods, the system comprehensively obtains various eye movement data of students during online learning, including pupil size, blinking frequency, fixation point position, eye movement speed, pupil light reaction, fixation point jump, binocular coordination, eye micro-movement frequency, etc. These data provide a reliable basis for the subsequent concentration assessment.

[0191] At the same time, the course team also collected formative evaluation information from students, including:

[0192] The quality of learning task completion is scored on a 5-point scale, with 1 indicating poor completion quality and 5 indicating excellent completion quality.

[0193] Learning process participation score P: A 5-point score is used, with 1 indicating very low participation and 5 indicating very high participation.

[0194] Knowledge mastery level score K: A 5-point score is used, 1 means very poor knowledge mastery, and 5 means very good knowledge mastery.

[0195] Difficulty level of learning content D c : According to the difficulty level of the course setting, an integer rating of 1-5 is used, with 1 representing the easiest and 5 representing the most difficult.

[0196] Knowledge point relevance score R k : A 5-level rating is used, where 1 means the knowledge point has a very low relevance and 5 means the relevance is very high.

[0197] Learning task time management score T: A 5-point score is used, 1 means very poor time management and 5 means very good time management.

[0198] Learning strategy use frequency score S l : Use 5-level rating, 1 means rarely used and 5 means frequently used.

[0199] Error correction ability score E: A 5-point score is used, where 1 means very poor error correction ability and 5 means very strong error correction ability.

[0200] The above 8 formative evaluation indicators are evaluated by the teachers according to the students' actual performance. Through these subjective evaluation information, the students' learning status and cognitive involvement can be more comprehensively reflected.

[0201] With abundant eye movement data and formative evaluation information, the course group then began to implement the concentration evaluation method of the present invention. First, the various eye movement feature parameters extracted in step S20 were analyzed and calculated:

[0202] Pupil size change rate R p (t): calculated with a sampling time interval of Δt = 0.02 seconds, α 1 =0.2,β 1 =0.1. In a certain learning session, at the 10th second, D p (10) = 4.2 mm, D p (9.98) = 4.1 mm, then

[0203] The spatial distribution density of fixation points ρ(x,y): According to the distribution statistics of fixation points at the 20th second, μ is calculated. x =512,μy =384,σ x =32,σ y =24,γ 1 =0.3,F(x,y)=0.1(x-μ x ) 2 +0.1(y-μ y ) 2 ,∈ 2 =0.015. Then the gaze density at the coordinate (480,360) is

[0204] Eye movement velocity change rate V(t): At the 30th second, the gaze point displacement Δx = 8 pixels, Δy = 12 pixels, and the sampling time interval Δt = 0.0083 seconds. Pixels / second.

[0205] The characteristic value of the gaze point jump pattern S(t): At the 40th second, the jump angle change Δθ(40) = 20 degrees, the jump distance Δd(40) = 128 pixels, and the jump duration t s (40) = 0.15 seconds, angular velocity degrees / second. Then S(40)=0.3×20+0.3×128+0.3×0.15+0.2×133.33+0.02=89.667.

[0206] Binocular coordination index C(t): At 50 seconds, parallax angle ΔP(50) = 2 degrees, phase difference φ(50) = 0.05 radians, fusion stability V f (50) = 0.98, parallax angular acceleration Degree / second 2 Then C(50)=0.3×2+0.3×0.05+0.3×0.98+0.15×0.2+0.01=1.245.

[0207] Next, the course group used a multi-scale sliding time window method to perform a time series analysis on the above eye movement feature parameters. The details are as follows:

[0208] Short-term window: 3 seconds window size, 1 second step length, 5 consecutive windows are maintained. It is used to reflect instantaneous attention fluctuations.

[0209] Medium time window: window size is 30 seconds, step length is 10 seconds, and 6 consecutive windows are maintained. It is used to characterize the level of sustained attention.

[0210] Long-term window: Window size 300 seconds, step length 60 seconds, keep 4 consecutive windows. Used to analyze long-term attention trends.

[0211] For the eye movement data in each time window, the course group extracted characteristic parameters such as pupil size variability, blink frequency, gaze point dispersion, saccade speed, pupil light response, gaze duration, binocular coordination, and micro-saccade frequency. After standardization and filtering, the characteristic values ​​shown in Table 1 were obtained:

[0212] Table 1 Eye movement feature values ​​in different time windows

[0213]

[0214]

[0215] Through weighted fusion, we get the comprehensive attention feature index M f :

[0216]

[0217] Among them, F S ,F m ,F l are the feature vectors of short, medium and long time windows respectively.

[0218] At the same time, the course group also generated the formative assessment adjustment factor F based on the students' formative assessment data through the adjustment factor empirical equation. r :

[0219] F r =0.15Q+0.12P+0.13K+0.11D c +0.13R k +0.12T+0.11S l +0.13E+0.02;

[0220] Next, the course group used the concentration evaluation equation group to comprehensively calculate the attention duration score A(t), visual concentration score V a (t) and cognitive engagement score E(t).

[0221] Attention duration score A(t):

[0222] A(t)=0.18R p (t)+0.22T b (t)+0.17R l (t)+0.20F m (t)+0.23C(t)+0.02;

[0223] Visual concentration score V a (t):

[0224] V a(t)=0.19ρ(x,y)+0.18V(t)+0.21S(t)+0.21C t (t)+0.21S t (t)+0.02;

[0225] Cognitive engagement score E(t):

[0226] E(t)=0.15A(t)+0.18V a (t)+0.16F r (t)+0.13D c (t)+0.15R k (t)+0.13S l (t)+0.02;

[0227] According to step S70, the course group applies the above-mentioned concentration evaluation model to the real-time eye movement data to obtain the concentration score in each sliding time window. For example:

[0228] At the 100th second of learning, the concentration score of short window 3 was 0.79, the score of medium window 4 was 0.82, and the score of long window 2 was 0.76.

[0229] At the 200th second, the concentration score of short-term window 5 was 0.83, that of medium-term window 6 was 0.85, and that of long-term window 3 was 0.80.

[0230] At 300 seconds, the concentration score of short window 1 was 0.77, the score of medium window 2 was 0.81, and the score of long window 4 was 0.78.

[0231] Finally, according to step S80, the course group calculated the overall concentration score using the concentration summary equation:

[0232]

[0233] Throughout the learning process, the students' overall concentration scores showed a trend of first rising and then falling. Figure 1 In the first half of the study, the concentration level was maintained at a high level, with an average score of about 0.80. However, in the second half, the concentration level gradually decreased, with the average score dropping to around 0.75.

[0234] The course team fed back the results to the teachers and made corresponding teaching intervention suggestions. For example, in the second half of the learning, teachers can appropriately increase the interactive sessions to guide students to maintain a high level of attention. At the same time, the difficulty of the teaching content can also be appropriately adjusted to better match the students' cognitive level. Through these targeted measures, it is believed that students' learning effects can be further improved.

[0235] It should be noted that the variables involved in the present invention are explained in detail as shown in Table 2 below.

[0236] Table 2 Variable explanation table

[0237]

[0238]

[0239] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for evaluating online learning concentration combined with formative assessment, characterized in that: The following steps are involved: S10, obtaining learners’ eye movement data during online learning; S20, extracting features from the eye movement data to obtain eye movement feature parameters; S30, performing time series analysis on the eye movement feature parameters using a multi-scale sliding time window method to obtain eye movement feature time series data; S40, obtaining formative assessment data of the learner and normalizing the formative assessment data, and generating a formative assessment adjustment factor using an adjustment factor empirical equation; S50, using the concentration evaluation equation group to calculate the attention duration score, visual concentration score and cognitive engagement score; S60, constructing a concentration evaluation model using a deep neural network based on the attention duration score, the visual focus score, and the cognitive engagement score; S70, applying the concentration evaluation model to the real-time eye movement feature time series data to obtain a concentration score within each sliding time window; S80: Calculate the overall concentration level using a concentration summary equation based on the concentration level scores in each sliding time window, and output the result.

2. The method for evaluating online learning concentration combined with formative evaluation according to claim 1, characterized in that: The eye movement data includes pupil size data, blink frequency data, gaze point position data, eye movement speed data, pupil light reaction data, gaze point jump data, binocular coordination data, and eye micro-movement frequency data.

3. The method for evaluating online learning concentration combined with formative evaluation according to claim 2, characterized in that: The eye movement characteristic parameters include pupil size change rate, blink interval duration, fixation point spatial distribution density, eye movement speed change rate, pupil light reaction rate, fixation point jump pattern characteristic value, binocular coordination index, and eye micro-movement frequency change rate.

4. The method for evaluating online learning concentration combined with formative evaluation according to claim 3, characterized in that: The multi-scale sliding time window includes a short-time window, a medium-time window and a long-time window, corresponding to attention features of different time scales.

5. The method for evaluating online learning concentration combined with formative evaluation according to claim 4, characterized in that: The formative evaluation data includes the learning task completion quality score, the learning process participation score, the knowledge mastery degree score, the learning content difficulty coefficient, the knowledge point relevance score, the learning task time management score, the learning strategy use frequency score, and the error correction ability score.

6. The method for evaluating online learning concentration combined with formative evaluation according to claim 5, characterized in that: The deep neural network includes a time series attention mechanism module, an adaptive weight adjustment module and a multimodal feature fusion layer.

7. The method for evaluating online learning concentration combined with formative evaluation according to claim 6, characterized in that: The concentration evaluation equation group includes an attention duration calculation equation, a visual focus calculation equation, and a cognitive engagement calculation equation.

8. The method for evaluating online learning concentration combined with formative evaluation according to claim 7, characterized in that: The attention duration calculation equation inputs the pupil size change rate, blink interval duration, pupil light reaction rate, eyeball micro-movement frequency change rate and binocular coordination index, and outputs an attention duration score.

9. The method for evaluating online learning concentration combined with formative evaluation according to claim 8, characterized in that: The visual concentration calculation equation inputs the spatial distribution density of the gaze point, the rate of change of the eye movement speed, the characteristic value of the gaze point jump pattern, the complexity of the gaze trajectory and the gaze stability index, and outputs a visual concentration score.

10. The method for evaluating online learning concentration combined with formative evaluation according to claim 9, characterized in that: The cognitive engagement calculation equation inputs the attention duration score, visual concentration score, formative assessment adjustment factor, content difficulty coefficient, knowledge relevance and learning strategy score, and outputs a cognitive engagement score.

Images