Multidisciplinary learning effect intelligent evaluation method and equipment based on electroencephalogram signal characteristics

Abstract

The invention discloses a multidisciplinary learning effect intelligent evaluation method and equipment based on electroencephalogram signal characteristics. A learning experiment scheme based on multidisciplinary video materials is designed, and in the process that a subject watches a teaching video and completes an answering task, a forehead lobe electroencephalogram signal of the subject is collected. The method comprises the following steps: extracting features of electroencephalogram signals in different frequency bands; and constructing a learning effect feature set in combination with learning behavior data such as answering accuracy and answering time. Fusing the behavior characteristics by adopting a clustering algorithm, and establishing a comprehensive learning effect index; furthermore, the learning effect is predicted based on the model, and intelligent evaluation of learning performance under a multi-disciplinary level is realized. According to the method, neural response differences among tasks of different subjects can be identified, personalized and adaptive analysis of learning effects is comprehensively realized, and the method has the characteristics of non-invasion, low cost and high timeliness, and can provide scientific and objective learning effect evaluation and personalized teaching decision basis for smart class, online education and learning state monitoring systems.

Description

Technical Field

[0001] This invention belongs to the field of electroencephalography (EEG) technology, specifically relating to an intelligent evaluation method and device for multidisciplinary learning performance based on EEG signal characteristics. Background Technology

[0002] Currently, the deep integration of education and technology has become an important direction for higher education. With the rapid development of intelligent learning analytics, brain-computer interface educational applications, and educational neuroscience, understanding and evaluating learners' learning processes and cognitive characteristics using objective physiological signals and intelligent algorithms has become a crucial research topic in the field of educational technology.

[0003] Traditional online or blended learning assessments primarily rely on behavioral data (such as learning duration and test scores), which struggle to reflect learners' deeper cognitive activities and attention allocation. This type of data is limited in scope, exhibits significant time lag, and lacks real-time and objective descriptions of the learning process, making it difficult to accurately predict and dynamically adjust individual learning outcomes. Therefore, singular behavioral data cannot fully reveal learners' neural activity patterns across different subject tasks.

[0004] Electroencephalography (EEG), with its high temporal resolution, non-invasiveness, and wearable nature, can capture changes in neural activity at the millisecond level, enabling real-time characterization of workload, attention levels, and cognitive processing states during learning tasks. Research indicates significant differences in EEG signal characteristics across different subject-specific learning tasks, particularly changes in activity intensity in the prefrontal cortex. These changes can serve as important physiological indicators of learning status and provide crucial evidence for intelligent evaluation of multidisciplinary learning outcomes.

[0005] However, most existing EEG-based learning research focuses on single-discipline or single-task experimental scenarios, lacking a systematic modeling and prediction framework for multidisciplinary learning conditions. Therefore, how to intelligently predict learning outcomes using EEG signal features combined with behavioral data in multidisciplinary learning scenarios is a pressing issue in the integration of educational neuroscience and artificial intelligence. Thus, it is necessary to establish an intelligent evaluation method for multidisciplinary learning outcomes that integrates EEG signals and learning behavior characteristics to improve the objectivity and accuracy of learning outcome analysis. Summary of the Invention To address the aforementioned problems in the existing technology, this invention provides a method and device for intelligent evaluation of multidisciplinary learning performance based on electroencephalogram (EEG) signal characteristics.

[0006] The technical problem to be solved by this invention is achieved through the following technical solution: This invention provides an intelligent evaluation method for multidisciplinary learning performance based on electroencephalogram (EEG) signal features, comprising: The test subjects' electroencephalogram (EEG) signals were collected when they were learning teaching materials for multiple subjects; the multiple subjects included N different subjects, where N is a positive integer greater than or equal to 2; each subject had multiple teaching courses, and each teaching course had multiple teaching materials; A raw EEG signal from the subject learning teaching materials from multiple disciplines was preprocessed to obtain a preprocessed signal. The frequency domain and statistical features of the preprocessed signal are extracted to obtain a multidisciplinary EEG feature. The multidisciplinary EEG features are input into a trained learning performance prediction model to predict the subject's learning performance in the multidisciplinary subjects. The trained learning performance prediction model is trained using an EEG feature dataset. Each sample in the EEG feature dataset represents the multidisciplinary EEG features of a participant, and the labels of the multidisciplinary EEG features are used to represent the participant's learning performance in each of the N different disciplines.

[0007] The present invention also provides an intelligent evaluation device for multidisciplinary learning performance based on EEG signal features, comprising a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; The memory is used to store computer programs; When the processor executes the program stored in the memory, it implements the steps of the above-described intelligent evaluation method for multidisciplinary learning effects based on EEG signal features.

[0008] Compared with the prior art, the beneficial effects of the present invention are as follows: 1) Multidimensional modeling by fusing brain signals and behavioral data: This invention integrates EEG signal features with learning behavior data to establish a comprehensive learning effect labeling system, which more accurately reflects learners’ cognitive input and learning efficiency. Compared with traditional evaluation methods that rely solely on grades or duration, this method is more scientific and objective. 2) Wearable devices support natural learning scenarios: Data collection is carried out using portable prefrontal EEG devices, which have the advantages of high timeliness, convenient operation and low cost. They are suitable for various scenarios such as classroom, online and blended learning, and can realize real-time monitoring of the learning process. 3) Constructing an individualized learning effect prediction model: Through multi-algorithm fusion modeling, this invention achieves high prediction accuracy at different subject and task levels, enabling hierarchical identification of students' learning status and individualized ability assessment, providing data support for personalized teaching; 4) Innovation in EEG applications in multidisciplinary learning scenarios: This invention analyzes the temporal and frequency characteristics of EEG during the learning process of test subjects in different disciplines, reveals the differences in neural activity between soft and hard disciplines as well as pure and applied disciplines, and achieves an objective comparison of cross-disciplinary learning effects; 5) Neuroscience support for promoting intelligent education: This invention has broad application value in the fields of teaching evaluation, intelligent education platforms and cognitive monitoring. It can provide teachers with objective neurophysiological indicators and personalized learning suggestions, and promote the scientific and intelligent nature of education evaluation.

[0009] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating an intelligent evaluation method for multidisciplinary learning performance based on EEG signal features provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the multidisciplinary learning EEG experimental process provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the EEG experimental data processing and learning effect prediction modeling process provided in the embodiments of the present invention; Figure 4 This is a schematic diagram of the EEG signal preprocessing flow provided in an embodiment of the present invention. Detailed Implementation

[0011] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0012] This invention aims to address the problems of existing technologies in evaluating learning outcomes, such as limited dimensions, lack of brain signal feature fusion, and limited prediction accuracy. It provides an intelligent evaluation method for multi-disciplinary learning outcomes based on EEG signal features. This method utilizes portable EEG devices to collect students' EEG signals during multi-disciplinary learning processes, combines this with learning behavior characteristics to establish a learning outcome prediction model, and analyzes the differences in neural activity under different subject learning tasks through comparative analysis of EEG features, achieving personalized and objective evaluation of learning outcomes. Electroencephalography (EEG) is a non-invasive technique that records neural electrical activity with millisecond-level temporal resolution, reflecting an individual's cognitive states during learning, including attention, working memory, and information processing. Different subject tasks exhibit significant differences in cognitive load, abstraction level, and knowledge processing methods, leading to stable and identifiable neural differences in learners' EEG spectral characteristics. This invention utilizes prefrontal EEG signals to analyze the neural mechanisms underlying differences in multi-disciplinary learning outcomes, and combines this with learning behavior data to construct a comprehensive learning outcome label, establishing a learning outcome prediction model based on EEG signal features to achieve intelligent evaluation of multi-disciplinary learning outcomes.

[0013] Figure 1 This is a flowchart illustrating an intelligent evaluation method for multidisciplinary learning outcomes based on electroencephalogram (EEG) signal features provided in an embodiment of the present invention. Figure 1 As shown, the method includes: S101. Collect EEG signals from the subject when learning teaching materials for multiple disciplines; multiple disciplines include N different disciplines, where N is a positive integer greater than or equal to 2, each discipline has multiple teaching courses, and each teaching course has multiple teaching materials.

[0014] For example, the N different disciplines can be at least two of the four disciplines selected according to the Biglan discipline classification model: social sciences, humanities, natural sciences, and applied sciences. Among them, social sciences and humanities are soft disciplines, while natural sciences and applied sciences are hard disciplines.

[0015] For example, each subject has two different teaching courses, and each teaching course has two different teaching materials. The teaching materials can be played in the form of videos, courseware, or PPTs, etc., and this invention does not limit them.

[0016] S102. A raw EEG signal from the subject learning teaching materials from multiple disciplines is preprocessed to obtain the preprocessed signal.

[0017] Here, a raw EEG signal when a test subject learns teaching materials from multiple disciplines refers to a raw EEG signal generated when the subject sequentially learns each teaching material from each of N different disciplines and completes a set of questions corresponding to each teaching material after learning it. In addition, the test subject also has the accuracy rate and time taken to answer each set of questions; each teaching material has a corresponding set of questions, and each test subject has a rest period before starting to learn the first teaching material of the first discipline, and a rest period after learning each teaching material or completing each set of questions; each test subject's raw EEG signal includes the EEG signals from their rest period, learning each teaching material, completing each set of questions, and each rest period. It should be noted that the total number of questions in each set of questions is the same; for example, each set of questions contains 10 different questions.

[0018] S103. Extract the frequency domain and statistical features of the preprocessed signal to obtain a multidisciplinary EEG feature.

[0019] Here, the frequency domain and statistical characteristics include: the average power characteristics of the δ EEG band (0.5–4Hz), θ EEG band (4–8Hz), α EEG band (8–13Hz), β EEG band (13–30Hz), and γ EEG band (30–40Hz), as well as the average amplitude, Shannon entropy, and sample entropy.

[0020] S104. Input the multidisciplinary EEG features into the trained learning effect prediction model to predict the test subject's learning effect on multiple subjects; wherein, the trained learning effect prediction model is trained using an EEG feature dataset, each sample in the EEG feature dataset is a test subject's multidisciplinary EEG features, and the labels of the multidisciplinary EEG features are used to represent the test subject's learning effect on each of N different subjects.

[0021] It should be noted that the trained learning performance prediction model is obtained by training any one of the following networks—Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest (RF), and Shallow Fully Connected Neural Network—on an EEG feature dataset. Since the model structures, loss functions, and training methods of SVM, KNN, RF, and Shallow Fully Connected Neural Networks are existing technologies, they will not be elaborated upon here. Furthermore, during the training phase, 10-fold cross-validation can be used to evaluate model performance, and the SMOTETomek method can be used to alleviate the class imbalance problem. Since the specific principles are existing technologies, they will not be elaborated upon here.

[0022] In some embodiments, the method for constructing the above-mentioned EEG feature dataset includes S10-S13: S10. Acquire raw EEG signals from multiple subjects, where each subject sequentially learns each teaching material for each of N different subjects, and completes a set of questions corresponding to each teaching material after learning it. Also acquire the accuracy rate and time taken for each subject to answer each set of questions. Each teaching material has a corresponding set of questions, and each subject has a rest period before starting to learn the first teaching material of the first subject, and a rest period after learning each teaching material or completing each set of questions. The raw EEG signals of each subject include the EEG signals of the subject's rest period, learning each teaching material, completing each set of questions, and each rest period.

[0023] For example, when there are 4 different subjects, each with 2 different teaching courses, each teaching course with 2 different video teaching materials, and each set of questions has 10 questions, the multi-disciplinary learning EEG experimental process designed in this invention is as follows: Figure 2 As shown, the EEG experimental data processing and learning effect prediction modeling process is as follows: Figure 3 As shown, the experimental stimuli were selected according to the Biglan subject classification model, covering four major categories: social sciences, humanities, natural sciences, and applied sciences, distinguishing between soft and hard sciences. A pre-test questionnaire was designed based on the learning content to assess students' familiarity with the knowledge points, ensuring they had no prior knowledge of the material. Students followed on-screen instructions for self-study; each video segment was approximately 1-3 minutes long and explained basic concepts; followed by multiple-choice questions corresponding to the video content. Throughout the experiment, students wore portable EEG signal acquisition devices, with data collected from their left and right foreheads to record EEG signals (i.e., EEG sampling data) during three phases: resting, watching videos, and answering questions. Simultaneously, the Matlab Psychotoolbox recorded learning behavior data, including answer time and accuracy, for constructing learning effectiveness labels. After preprocessing, quality screening, and feature extraction, the EEG sampling data was used to construct the EEG dataset and to analyze differences in learning effectiveness across multiple disciplines. The constructed learning effect labels and EEG dataset are then used to train the learning effect prediction model. For example, in the EEG experiment designed in this embodiment, each participant rests for a period of time before the first learning video. They then watch eight videos sequentially, taking a break after each video, followed by completing a set of questions corresponding to that video. This process is repeated until all 16 videos have been watched and all 16 sets of questions have been completed. For example, the total duration of one experiment is approximately 45 minutes.

[0024] S11. Preprocess and extract frequency domain and statistical features from one raw EEG signal for each subject to obtain a multidisciplinary EEG feature sample corresponding to the subject. A multidisciplinary EEG feature sample refers to a multidisciplinary EEG feature as a sample. The multidisciplinary EEG feature sample contains N sets of EEG features that correspond one-to-one with N different disciplines.

[0025] Here, combined Figure 4 As shown, the method for preprocessing and extracting frequency domain and statistical features from a single raw EEG signal for each participant to obtain a multidisciplinary EEG feature sample for each participant includes steps S201-S210: S201. Remove the signal from an invalid channel in one of the original EEG signals of the test subject and perform electrode localization to preserve the signal of the prefrontal electrode channel of the test subject and make the lead position correspond to the brain region.

[0026] S202. The signals from the retained prefrontal electrode channels are sequentially subjected to ocular de-ocularization, rereference, and filtering processes to obtain the filtered EEG signal of the subject.

[0027] Specifically, firstly, invalid or unused channels in the original EEG signal were removed according to the experimental design, retaining only the target prefrontal cortex electrodes. Next, electrode localization was performed on the six retained prefrontal cortex leads to ensure consistency between lead locations and brain region correspondences. Then, an adaptive artifact removal algorithm (AAR) combined with dual-channel EOS signals was used to remove the EOS component from the EEG signal. Subsequently, the EEG signal was rereferenced using the Reference Electrode Standardization Technique (REST). Finally, a 0.1-40Hz finite impulse response bandpass filter was used to filter the entire EEG signal. To avoid instability in ICA decomposition due to insufficient lead numbers, this invention did not use independent component analysis (ICA) for further artifact separation.

[0028] S203. Extract the resting EEG signal and the EEG signal of each teaching material stage under each teaching lesson of each subject from the filtered EEG signal of the test subject.

[0029] Specifically, based on the experimental behavioral event markers recorded by the Matlab Psychtoolbox experimental program, the EEG signals of the subject were extracted during the resting phase, as well as during each phase of learning teaching materials for each subject in each lesson.

[0030] S204. Perform Detrend baseline removal processing on the EEG signals of each teaching material learning stage, and segment the EEG signals of each teaching material learning stage after Detrend baseline removal processing to obtain multiple signal segments of fixed length.

[0031] Specifically, after obtaining the EEG signals for each stage of learning the teaching materials, to ensure the reliability of subsequent feature extraction, this invention further segments the EEG signals for each stage of learning the teaching materials. Specifically, the Detrend method is used to remove linear trends from the EEG signals for each stage of learning the teaching materials, eliminating the influence of low-frequency drift. Then, the detrended EEG signals for each stage of learning the teaching materials are divided into multiple signal segments of fixed length according to a preset window length and a preset overlap rate. For example, the preset window length is 10 seconds (s), and the preset overlap rate is 0.7.

[0032] S205. Signal quality assessment and screening are performed on multiple fixed-length signal segments to obtain multiple effective signal segments from the EEG signals of the subject during each stage of learning the teaching materials.

[0033] For example, for each 10-second signal segment, let the signal value of a certain sampling point in the signal segment be y, the mean of the sampling points in the signal segment be μ, and the standard deviation be σ. If |y-μ|>3*σ, then the sampling point is determined to be an outlier. If the proportion of outliers in the signal segment exceeds a threshold (e.g., 50%), then the signal segment is considered of poor quality and is discarded. If the proportion of outliers does not exceed the threshold, the filloutliers function in MATLAB is used to replace the outliers in the signal segment using nearest neighbor interpolation to ensure signal continuity and smoothness. Finally, the amplitude of the interpolated signal segment is checked. If the absolute value of the signal at any sampling point in the signal segment is greater than ±150μV, then the signal segment is discarded; otherwise, the signal segment is considered of good quality and can be retained. All the signal segments retained from the EEG signals of the subject during the learning phase of a certain teaching material are the valid signal segments in the EEG signals of the subject during the learning phase of that teaching material.

[0034] It should be noted that the preprocessing principle described in S201 to S205 here is the same as the preprocessing principle in S102 above. The present invention will not repeat the preprocessing principle in S102 above.

[0035] S206. Extract the frequency domain and statistical features of each effective signal segment to obtain the EEG features of each effective signal segment during each stage of the subject's learning of teaching materials.

[0036] Specifically, the method for extracting the frequency domain and statistical features of each effective signal segment is as follows: The power spectral density of the effective signal segment is obtained using the Welch method. Then, the average power characteristics of typical EEG frequency bands δ (0.5–4Hz), θ (4–8Hz), α (8–13Hz), β (13–30Hz), and γ (30–40Hz) are calculated on the power spectral density to obtain the average power characteristics of the δ, θ, α, β, and γ EEG bands of the effective signal segment. Next, the average absolute amplitude of the effective signal segment is obtained by averaging the absolute values ​​of the amplitudes of each channel. Finally, the Shannon entropy and sample entropy of the effective signal segment are calculated separately to obtain their respective values. The Welch method is a robust spectrum estimation algorithm based on the discrete Fourier transform. By segmenting the signal, windowing, and averaging the power of each segment, it can effectively reduce the variance of the spectrum estimation.

[0037] S207. Combine the EEG characteristics of multiple effective signal segments during each stage of the subject's learning of the teaching materials to obtain the EEG characteristics of the subject during each stage of the subject's learning of the teaching materials.

[0038] The EEG characteristics of the subject during each stage of learning the teaching material were a multidimensional feature vector.

[0039] S208. Based on the EEG characteristics of the test subject during all stages of learning the teaching materials under each teaching course, determine the EEG characteristics of the test subject for each teaching course under each subject.

[0040] Specifically, by averaging the EEG characteristics of the test subject across all stages of learning a particular course, the EEG characteristics of that test subject learning that particular course can be obtained.

[0041] S209. Based on the EEG characteristics of the test subject learning all the teaching courses under each subject, determine a set of EEG characteristics of the test subject learning each subject.

[0042] Specifically, by averaging the EEG characteristics of all the courses taught by the subject in a certain subject, a set of EEG characteristics of the subject in that subject can be obtained.

[0043] S210. The EEG feature vector composed of N sets of EEG features from N different subjects learned by the test subject is used as a multidisciplinary EEG feature sample corresponding to the test subject.

[0044] S12. Based on the test subject's answer accuracy and answer time for each set of questions, and the clustering algorithm, generate a multi-disciplinary label for the multi-disciplinary EEG feature samples. The multi-disciplinary label contains N labels corresponding to N sets of EEG features, and each label is used to characterize the learning effect of a subject.

[0045] Specifically, the time allotted for each participant to answer each set of questions includes the time allotted for each question. Based on this, S12 is implemented through steps S301~S306: S301. Sort the completion times of all participants who answered each question in each set of questions in ascending order, and then perform K-Means clustering to divide all participants who answered each question in each set of questions in each set of questions into multiple different clusters that represent different answering speeds.

[0046] For example, using K-Means clustering, all participants who answered each question correctly in each set of questions can be divided into three clusters: "fast", "medium", and "slow" answering speeds.

[0047] S302. For each test subject, count the total number of questions in the cluster of the fastest answering speed in each set of questions completed by the test subject, and take the ratio of the total number of questions in the cluster of the fastest answering speed in each set of questions to the total number of questions in each set of questions as the test subject's fast and correct answering rate for each set of questions.

[0048] For example, if each set of questions contains 10 questions, and 4 questions in a certain set of questions belong to the cluster of "fast" answering speed, then the fast and correct answering rate of that set of questions for the experimenter is 4 / 10 = 0.4.

[0049] S303. Based on the test subject's accuracy rate and speed of answering questions for each set of questions, determine the test subject's comprehensive learning effect label for the teaching materials corresponding to each set of questions.

[0050] Specifically, the test subject's correct answer rate and quick correct answer rate for each set of questions were weighted and summed to obtain the test subject's comprehensive learning effect label for the teaching materials corresponding to each set of questions. Among them, the weight of the correct answer rate was greater than the weight of the quick correct answer rate.

[0051] For example, the expression for the comprehensive learning effectiveness label Q of the test subject for a certain set of questions and corresponding teaching materials is: Q = 0.7 × Accuracy + 0.3 × FastAcc, where Accuracy is the test subject's correct answer rate for that set of questions, and FastAcc is the test subject's fast and accurate answer rate for that set of questions. 0.7 and 0.3 are the weights of the correct answer rate and the fast and accurate answer rate, respectively. This label reflects both learning accuracy and learning efficiency, making the evaluation of learning outcomes more comprehensive and objective. It should be noted that the correct answer rate for a certain set of questions refers to the ratio of the total number of correctly answered questions to the total number of questions in that set. For example, if the total number of correctly answered questions in a certain set is 6, and the total number of questions in that set is 10, then the correct answer rate for that set of questions is 6 / 10 = 0.6.

[0052] S304. The average of the comprehensive learning effect labels of the test subject for each of the four teaching materials is taken as the comprehensive learning effect label of the test subject for each subject.

[0053] S305. Using subjects as categories, perform K-Means clustering on the comprehensive learning effect labels of all testers for each subject, so as to divide the comprehensive learning effect labels of all testers for each subject into multiple clusters that represent various different learning effects, and obtain the learning effect label of each tester for each subject.

[0054] For example, by performing K-Means clustering on the comprehensive learning performance labels of all participants for each subject, the comprehensive learning performance labels of all participants for each subject can be divided into three clusters with learning performance of "good", "average" and "poor". In this way, the learning performance label of each participant for each subject can be obtained.

[0055] S306. For each subject, four learning effect labels for four subjects are used to form a multidisciplinary label for a multidisciplinary EEG feature sample of the subject.

[0056] S13 uses a set of multidisciplinary EEG feature samples with multidisciplinary labels corresponding to multiple test subjects as an EEG feature dataset.

[0057] In some embodiments, after S103, steps S105 to S112 are further included: S105. From the multidisciplinary EEG characteristics of the test subject, obtain the EEG characteristics of two different teaching material stages for learning each teaching course, and obtain two first data columns.

[0058] S106. Analyze and compare the two first data columns to obtain the differences in EEG characteristics when the test subject learns two different teaching materials for each course.

[0059] Specifically, we analyze whether both first data columns satisfy a normal distribution. If either first data column does not satisfy a normal distribution, we perform a Mann-Whitney rank-sum test on the two first data columns to obtain the test results. The test results are used to characterize the differences in EEG characteristics when the test subject learns two different teaching materials for each course. If both first data columns satisfy a normal distribution, we first perform a homogeneity of variance test on the two first data columns to determine whether homogeneity of variance holds. If homogeneity of variance holds, we perform an unpaired t-test on the two first data columns to obtain the test results. If homogeneity of variance does not hold, we perform a Welch t-test on the two first data columns to obtain the test results.

[0060] S107. The average value of the EEG characteristics of the subject during two different stages of learning each teaching course is taken as the EEG characteristics of the subject during learning each teaching course.

[0061] S108. The EEG characteristics of the subject learning two different courses in each subject are used as two second data columns.

[0062] S109. Analyze and compare the two second data columns to obtain the differences in EEG characteristics of the test subject when learning two different teaching courses for each subject.

[0063] It should be noted that the principle of analyzing and comparing two second data columns is the same as the principle of analyzing and comparing two first data columns described above, and will not be repeated here.

[0064] S110. The average value of the EEG characteristics of the test subject for two different teaching courses in each subject is taken as the EEG characteristics of the test subject for learning each subject.

[0065] S111. The EEG characteristics of the subject learning each of two subjects are used as two third data columns.

[0066] S112. Analyze and compare the two third data columns to obtain the differences in EEG characteristics when the test subject studies two different subjects.

[0067] It should be noted that the principle of analyzing and comparing two third data columns is the same as the principle of analyzing and comparing two first data columns described above, and will not be repeated here.

[0068] The test results obtained through the above statistical analysis can be used to identify differences in cognitive load under different subject tasks, providing a feature basis for subsequent learning effect prediction models.

[0069] In some embodiments, after S103, steps S113 to S116 are further included: S113. Extract key features from the EEG characteristics of the test subject learning each subject and aggregate them to obtain the EEG characteristic spectrum of the test subject learning each subject.

[0070] Specifically, for each subject, the average power, average amplitude, Shannon entropy, and sample entropy of each frequency band (δ, θ, α, β, γ) are extracted from the EEG characteristics of the test subject learning that subject, and then combined to obtain the EEG characteristic spectrum of the test subject learning that subject. The EEG characteristic spectrum can be used to measure students' attention level, cognitive load changes, and information processing ability in different subject tasks.

[0071] S114. Based on the EEG characteristic spectra of the test subject learning N different subjects, determine the test subject's attention level index, cognitive load index, and interdisciplinary adaptation index respectively.

[0072] Here, the attention level index of the test subject is determined based on the stability and differences in the average power of the alpha EEG band in the EEG characteristic spectrum of different subjects. Specifically, the average power of the alpha EEG band in the EEG characteristic spectrum of N different subjects is calculated to obtain the average power of the alpha EEG band. Then, the standard deviation of the average power of the alpha EEG band in the EEG characteristic spectrum of N different subjects is calculated based on the average value. If the standard deviation exceeds the standard deviation threshold, the test subject is considered to have a low attention level index and large attention fluctuations, and can be marked as "attentional distraction risk". If the standard deviation does not exceed the standard deviation threshold, and the average power of the alpha EEG band in the EEG characteristic spectrum of N different subjects exceeds the preset average power threshold, the test subject is considered to have a high attention level index, that is, the test subject has high alpha power and stable attention level, and can be marked as "good attention".

[0073] Here, the cognitive load index of the test subject is determined based on the increase in the average power of the β-EEG band in the EEG characteristic spectrum of the subject learning different subjects. Specifically, the average power of the β-EEG band of a segment of EEG signal in a stable state during the resting phase is obtained as the judgment benchmark. The average power of the β-EEG band in the EEG characteristic spectrum of the subject learning each subject is then subtracted from this benchmark. If the differences obtained for all N subjects are greater than 0 and greater than a preset value, it indicates that the average power of the β-EEG band in the EEG characteristic spectrum of the subject learning N different subjects has increased significantly, indicating that the subject's cognitive load index is high, and the subject can be marked as "high cognitive load". If the differences obtained for some subjects are greater than 0 but the differences obtained for other subjects are less than 0, it indicates that the average power of the β-EEG band of the subject has abnormal fluctuations, indicating that the subject's cognitive load index is low, and the subject can be marked as "high task stress".

[0074] Here, the interdisciplinary adaptation index of the test subject is determined based on the consistency of changes in EEG characteristic spectra when the test subject learns different subjects. Specifically, the average value of the EEG characteristic spectra of the test subject learning all subjects belonging to soft sciences is calculated to obtain the first average characteristic spectrum, and the average value of the EEG characteristic spectra of the test subject learning all subjects belonging to hard sciences is calculated to obtain the second average characteristic spectrum. The similarity and difference between the frequency domain features in the first and second average characteristic spectra are compared. If the similarity is greater than the similarity threshold and the difference is less than the difference threshold, it indicates that the difference in cognitive activation patterns between soft and hard sciences is small, indicating that the test subject's interdisciplinary adaptation index is high, and the test subject can be marked as "good interdisciplinary adaptation". If the similarity is less than the similarity threshold and the difference is greater than the difference threshold, it indicates that the difference in cognitive activation patterns between soft and hard sciences is large, indicating that the test subject's interdisciplinary adaptation index is low, and the test subject can be marked as "weak interdisciplinary adaptation".

[0075] S115. Based on the test taker's accuracy rate and time taken to answer each set of questions, calculate the test taker's comprehensive learning effect label for each subject. Based on the test taker's comprehensive learning effect label for N different subjects, determine the test taker's learning efficiency index.

[0076] Specifically, the accuracy rate and time taken by multiple test subjects for each set of questions can be obtained. The test subject is then used as a new test subject. The methods described in S301-S304 above are then used to determine the test subject's overall learning performance label for N different subjects. Alternatively, a time threshold can be set to determine which questions the test subject answered correctly for each set of questions had a completion time shorter than the threshold. Finally, the ratio of the total number of questions answered correctly for a set of questions with a completion time shorter than the threshold to the total number of questions in that set is taken as the rapid accuracy rate for that set of questions. Then, the methods described in S303-S304 above are used to determine the test subject's overall learning performance label for N different subjects.

[0077] Specifically, after obtaining the comprehensive learning effect labels of the test subject for N different subjects, each comprehensive learning effect label is compared with a label threshold. If the comprehensive learning effect labels of the test subject for N different subjects are all greater than the label threshold, it indicates that the test subject has a high learning efficiency index and is a stable and efficient learner, and can be marked as "fast processing speed". If the comprehensive learning effect labels of some subjects are all greater than the label threshold, but the comprehensive learning effect labels of other subjects are all less than the label threshold, it indicates that the test subject has a low learning efficiency index and is a learner with large efficiency fluctuations, and can be marked as "unstable learning speed". If the comprehensive learning effect labels of the test subject for N different subjects are all less than the label threshold, it indicates that the test subject has a low learning efficiency index, and can be marked as "slow processing speed".

[0078] S116. Based on the test subject's attention level index, cognitive load index, interdisciplinary adaptation index, and learning efficiency index, and the predicted learning effect of the test subject on multiple subjects, a personalized learning ability profile of the test subject is obtained.

[0079] After obtaining the test subject's attention level index, cognitive load index, interdisciplinary adaptation index, and learning efficiency index, and predicting the test subject's learning performance in multiple subjects, these indices and predictions constitute a personalized learning ability profile of the test subject.

[0080] In some embodiments, after S116, step S117 is further included: S117. Based on the individualized learning ability profile of the test subject, generate personalized learning suggestions.

[0081] For example, when the attention level index is below a preset threshold, attention-related suggestions are generated, such as "shortening video viewing time and increasing spaced review"; when the cognitive load index remains high, cognitive load adjustment suggestions are generated, such as "breaking learning materials into smaller knowledge blocks and learning with examples"; when the learning efficiency index is low and accompanied by a low learning outcome prediction level, efficiency improvement suggestions are generated, such as "increasing the number of practice questions and using timed quiz training"; and when the interdisciplinary adaptation index is weak, learning path optimization suggestions are generated, such as "starting with soft science content and then transitioning to hard science." Thus, this invention can automatically generate personalized learning reports and actionable learning suggestions based on different students' EEG response characteristics and behavioral performance, achieving continuous, targeted, and adaptive feedback in the learning process. It should be noted that more specific details can be set according to actual needs, and this invention does not limit them.

[0082] The present invention also provides an intelligent evaluation device for multidisciplinary learning performance based on EEG signal characteristics, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; The processor, when executing the program stored in the memory, implements the above-described method steps for intelligent evaluation of multidisciplinary learning effects based on EEG signal features.

[0083] This invention has the following advantages: 1) This invention synchronously collects students' electroencephalogram (EEG) signals during learning tasks, recording learners' neural activities during rest, learning, and answering phases, and combines this with behavioral data to comprehensively model learning outcomes. This invention can reveal differences in learning states across different subject tasks at the neural activity level, providing new objective evidence for analyzing attention, cognitive load, and information processing mechanisms during the learning process. Unlike traditional one-dimensional learning evaluations that rely on questionnaires or grades, this invention achieves objective learning outcome assessment based on EEG signals, reflecting the immediacy and dynamic changes in the learning process.

[0084] 2) This invention employs a multidisciplinary EEG experimental design based on the Biglan subject classification model. Experimental materials are derived from real online course video tasks. Compared to traditional paradigms that present static images or single stimuli, this design combines video, audio, and interactive question-and-answer sessions to more realistically simulate the actual learning environment, thereby improving the ecological validity of the research and its adaptability to educational applications. By collecting EEG signals in natural learning scenarios, this invention can capture the neural response characteristics of learners in their real-world state, providing highly reliable data support for the analysis of cognitive differences across different subjects.

[0085] 3) This invention effectively improves data quality and analytical accuracy through a series of signal preprocessing steps, including filtering, artifact removal, REST rereference, and outlier interpolation. In the feature extraction stage, this invention comprehensively employs power spectral density analysis, amplitude variation calculation, and entropy feature extraction, fully utilizing the temporal and frequency domain information of EEG signals, thereby enabling more sensitive identification of cognitive differences and workload changes in learning tasks. Compared to single-band power analysis methods, this feature extraction method has higher robustness and interpretability.

[0086] 4) By comparing the power differences of various EEG bands (δ, θ, α, β, γ) under different subject tasks, the neural mechanisms underlying the differences in learning outcomes across different subjects can be analyzed. The study found that the increased power in the θ and α bands during soft subject tasks is associated with deep semantic processing and attention retention; while the increased power in the β band during hard subject tasks reflects improved logical reasoning and working memory load. These results demonstrate that this invention not only enables quantitative prediction of learning outcomes but also reveals cognitive differences in learning across different subjects from a neural perspective, providing new empirical support for educational neuroscience research.

[0087] 5) This invention constructs a comprehensive learning performance label by standardizing and fusing behavioral data such as students' answer accuracy and response time. This label is then used in conjunction with EEG features for predictive model training. This invention employs a multi-model fusion strategy, including algorithms such as Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest (RF), and shallow fully connected neural networks, to model and predict learning performance. This method can hierarchically identify students' learning states based on EEG and behavioral characteristics, thereby achieving intelligent prediction of individual learning performance.

[0088] 6) This invention constructs a personalized learning ability profile by combining EEG feature distribution, behavioral performance, and prediction results, and automatically generates targeted learning suggestions. This method reflects learners' focus, information processing efficiency, and cognitive load from both neurophysiological and behavioral perspectives, thus providing a dynamic and sustainable feedback mechanism for teachers and students. Through this mechanism, students can adjust their learning strategies based on individual feedback, and teachers can optimize their instructional design to further improve teaching effectiveness and learning outcomes.

[0089] It should be noted that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined as "first" or "second" may explicitly or implicitly include one or more features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0090] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0091] In this specification, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple instances. While different embodiments may describe certain measures, this does not mean that these measures cannot be combined to produce a good effect.

[0092] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A multidisciplinary learning performance intelligent evaluation method based on EEG signal features, characterized in that, include: Collect EEG signals from subjects while they are learning teaching materials from multiple disciplines; The multidisciplinary field includes N different disciplines, where N is a positive integer greater than or equal to 2; Each subject has multiple teaching courses, and each teaching course has multiple teaching materials; A raw EEG signal from the subject learning teaching materials from multiple disciplines was preprocessed to obtain a preprocessed signal. The frequency domain and statistical features of the preprocessed signal are extracted to obtain a multidisciplinary EEG feature. The multidisciplinary EEG features are input into a trained learning performance prediction model to predict the subject's learning performance in the multidisciplinary subjects. The trained learning performance prediction model is trained using an EEG feature dataset. Each sample in the EEG feature dataset represents the multidisciplinary EEG features of a participant, and the labels of the multidisciplinary EEG features are used to represent the participant's learning performance in each of the N different disciplines.

2. The intelligent evaluation method for multidisciplinary learning performance based on EEG signal features according to claim 1, characterized in that, The method for constructing the EEG feature dataset includes: The experiment involved acquiring raw EEG signals from multiple participants. Each participant sequentially studied each teaching material for each of the N different subjects, and completed a set of questions corresponding to each teaching material after studying it. The experiment also recorded the accuracy rate and time taken for each participant to answer each set of questions. Each teaching material had a corresponding set of questions, and each participant had a rest period before starting to study the first teaching material of the first subject, and a rest period after studying each teaching material or completing each set of questions. Each participant's raw EEG signal included the EEG signals from their rest period, studying each teaching material, completing each set of questions, and each rest period. A raw EEG signal from each participant is preprocessed and subjected to frequency domain and statistical feature extraction to obtain a multidisciplinary EEG feature sample for each participant. A multidisciplinary EEG feature sample refers to one of the multidisciplinary EEG features as a sample. The multidisciplinary EEG feature sample contains N sets of EEG features that correspond one-to-one with the N different disciplines. Based on each participant's correct answer rate and answer time for each set of questions, and a clustering algorithm, a multidisciplinary label is generated for the multidisciplinary EEG feature samples. The multidisciplinary label contains N labels corresponding to the N sets of EEG features, and each label is used to characterize the learning effect of a subject. The set consisting of multiple EEG feature samples with the multidisciplinary labels corresponding to the multiple test subjects is used as the EEG feature dataset.

3. The intelligent evaluation method for multidisciplinary learning performance based on EEG signal features according to claim 2, characterized in that, The process involves preprocessing and extracting frequency domain and statistical features from a single raw EEG signal for each participant to obtain a multidisciplinary EEG feature sample for each participant, including: The invalid channels in one raw EEG signal of each subject were removed, and electrode localization was performed to preserve the signals of the prefrontal electrode channels of each subject and to make the lead positions correspond to the brain regions. The signals from the retained prefrontal electrode channels were sequentially subjected to ocular de-ocularization, rereference, and filtering to obtain the filtered EEG signals for each subject. The resting phase EEG signal and the teaching material phase of each subject's learning were extracted from the filtered EEG signal of each subject. The EEG signals of each learning stage were processed by Detrend baseline removal, and the EEG signals of each learning stage after Detrend baseline removal were segmented to obtain multiple signal segments of fixed length. The signal quality of the multiple fixed-length signal segments is evaluated and screened to obtain multiple effective signal segments from the EEG signals of each subject during each stage of learning the teaching material. The frequency domain and statistical features of each effective signal segment were extracted to obtain the EEG features of each effective signal segment for each subject during each stage of learning each teaching material; The EEG characteristics of each subject during each learning stage of the teaching material are combined to obtain the EEG characteristics of each subject during each learning stage of the teaching material. Based on the EEG characteristics of each participant during the learning of all teaching materials for each subject, the EEG characteristics of each participant during the learning of each subject's teaching course were determined. Based on the EEG characteristics of each participant learning all the teaching courses under each subject, a set of EEG characteristics for each participant learning each subject was determined. Each participant learns N sets of EEG features from N different disciplines to form an EEG feature vector, which serves as a multidisciplinary EEG feature sample for each participant.

4. The intelligent evaluation method for multidisciplinary learning performance based on EEG signal features according to claim 2, characterized in that, The time each participant takes to answer each set of questions includes the time taken to complete each question; the generation of a multidisciplinary label for the multidisciplinary EEG feature samples based on each participant's answer accuracy and answer time for each set of questions, and a clustering algorithm, includes: The completion times of all participants who answered each question correctly in each set of questions were sorted in ascending order and then K-Means clustering was performed to divide all participants who answered each question correctly in each set of questions into multiple different clusters that represent different answering speeds. The total number of questions in the fastest answering cluster in each set of questions completed by each participant was counted. The ratio of the total number of questions in the fastest answering cluster in each set of questions to the total number of questions in each set of questions was taken as the fast and correct answering rate of each participant for each set of questions. Based on each participant's accuracy rate and speed of answering questions for each set of questions, a comprehensive learning effect label for each participant's teaching materials for each set of questions is determined. The average of the comprehensive learning effect labels for each participant across the four teaching materials for each subject was used as the comprehensive learning effect label for each participant across the subject. K-Means clustering was performed on the comprehensive learning effect labels of all testers for each subject, based on the subject as the category, to divide the comprehensive learning effect labels of all testers for each subject into multiple clusters that represent various different learning effects, thus obtaining the learning effect label of each tester for each subject. Each participant was assigned a multidisciplinary label for a multidisciplinary EEG feature sample, consisting of four learning outcome labels for four subjects.

5. The intelligent evaluation method for multidisciplinary learning performance based on EEG signal features according to claim 4, characterized in that, The method involves determining the overall learning effectiveness label for each participant's teaching materials for each set of questions based on their correct answer rate and rapid correct answer rate for each set of questions. This label includes: The correct answer rate and the rapid correct answer rate for each subject for each set of questions are weighted and summed to obtain the comprehensive learning effect label for each subject for the teaching materials corresponding to each set of questions. The weight of the correct answer rate is greater than the weight of the rapid correct answer rate.

6. The intelligent evaluation method for multidisciplinary learning performance based on EEG signal features according to claim 2, characterized in that, Each subject has two different teaching courses, and each teaching course has two different teaching materials. The method also includes: From the multidisciplinary EEG characteristics of the subject, the EEG characteristics of two different teaching material stages for learning each teaching course are obtained, resulting in two first data columns; By analyzing and comparing the two first data columns, the differences in EEG characteristics of the test subject when learning two different teaching materials for each teaching course were obtained; The average value of the EEG characteristics of the subject during two different teaching material stages of learning each teaching course was taken as the EEG characteristics of the subject during learning each teaching course. The EEG characteristics of the subject learning two different courses in each subject were used as two second data columns. By analyzing and comparing the two second data columns, the differences in EEG characteristics of the test subjects when learning two different courses in each subject were obtained; The average value of the EEG characteristics of the subject learning two different courses in each subject is taken as the EEG characteristics of the subject learning each subject. The subjects learned the EEG characteristics of each of two subjects, which were then used as two third data columns. By analyzing and comparing the two third data columns, the differences in EEG characteristics of the test subjects when learning two different subjects were obtained.

7. The intelligent evaluation method for multidisciplinary learning performance based on EEG signal features according to claim 6, characterized in that, The differences in EEG characteristics between the two first data columns, two second data columns, and two third data columns were obtained by analyzing and comparing them. These differences included: Analyze whether both first data columns, both second data columns, and both third data columns follow a normal distribution; If any data column does not satisfy a normal distribution, then the Mann-Whitney rank-sum test is performed on the two first data columns, the two second data columns, and the two third data columns to obtain the test results. The test results are used to characterize the differences in EEG characteristics of the test subject when learning two different teaching materials for each teaching course, learning two different teaching courses for each subject, and learning two different subjects. If both first data columns, two second data columns, and two third data columns all satisfy a normal distribution, then first perform a homogeneity of variance test on the two first data columns, two second data columns, and two third data columns to determine whether the homogeneity of variance holds. If the homogeneity of variance of the two first data columns / two second data columns / two third data columns holds, then perform an unpaired t-test on the two first data columns / two second data columns / two third data columns to obtain the test results; If the homogeneity of variance of the two first data columns / two second data columns / two third data columns is not met, then Welch's t-test is performed on the two first data columns / two second data columns / two third data columns to obtain the test results.

8. The intelligent evaluation method for multidisciplinary learning performance based on EEG signal features according to claim 6, characterized in that, The method further includes: Key features are extracted and aggregated from the EEG features of the subject learning each subject to obtain the EEG feature spectrum of the subject learning each subject. Based on the EEG characteristic spectra of the test subject learning N different subjects, the attention level index, cognitive load index, and interdisciplinary adaptation index of the test subject are determined respectively. Based on the test subject's answer accuracy and answer time for each set of questions, calculate the test subject's comprehensive learning effect label for each subject. Based on the test subject's comprehensive learning effect labels for N different subjects, determine the test subject's learning efficiency index. Based on the subject's attention level index, cognitive load index, interdisciplinary adaptation index, and learning efficiency index, as well as the predicted learning performance of the subject across multiple subjects, a personalized learning ability profile of the subject is obtained.

9. The intelligent evaluation method for multidisciplinary learning performance based on EEG signal features according to claim 8, characterized in that, The step of determining the subject's attention level index, cognitive load index, and interdisciplinary adaptation index based on the EEG characteristic spectra of the subject learning N different subjects includes: The attention level index of the test subject is determined based on the stability and differences in the average power of the α-brainwave band in the EEG characteristic spectrum of the test subject learning different subjects; The cognitive load index of the test subject is determined based on the increase in the average power of the β-brainwave band in the EEG characteristic spectrum of the test subject learning different subjects. The interdisciplinary adaptation index of the test subjects was determined based on the consistency of the changes in EEG characteristic spectra of the test subjects when they learned different subjects.

10. A multidisciplinary learning performance intelligent evaluation device based on EEG signal characteristics, comprising a processor, a communication interface, a memory, and a communication bus, characterized in that, The processor, the communication interface, and the memory communicate with each other via the communication bus; The memory is used to store computer programs; When the processor executes a program stored in the memory, it implements the steps of the method described in any one of claims 1-9.

Images