使用腦電波訊號預測英文學習狀態

Abstract

由於腦波儀設備的科技進步，採用腦電波方式探討語言學習的研究逐漸增加，但對於英語（第二語言）不同的學習教材刺激，採用腦電波訊號作為學習歷程，並進一步採用機器學習預測學習狀態研究甚少。為此，本研究建置特徵導向資料科學模式（Feature-oriented data science model，FDSM），其主要將特徵工程技術嵌入資料科學，將腦電波的特徵訊號與英文學習表現進行分析。 FDSM包括四個階段：第一，資料選擇階段：本研究設計實驗蒐集資料，一共有徵求100位（男女各半）受試者。每位受試者進行教材（閱讀、聽力、視聽）*四個單元，一共是12次實驗。過程中進行腦電波蒐集。第二，資料預處理階段：刪除15位受試者腦電波雜訊，採用過抽樣與生成網路模擬解決學習表現不平衡資料。第三，特徵學習階段：資料直接進行機器學習或者先進行特徵篩選，接著特徵創建，最後進行特徵萃取。第四，解釋評估階段：根據結果提出相關建議。 本研究總共有五項結果：第一，英文整體學習歷程偏向聽覺與動覺感官學習風格；在英文研讀當下則是偏向視覺與動覺感官學習風格；對於英文訊息在接收與處理則又偏向視覺/非語言感官學習風格；第二，感官學習風格與學習成效關係：閱讀教材學習表現與團隊、視覺/觸覺-正向的影響。聽力教材學習表現與動覺、聽覺正向的影響。視聽教材學習表現與聽覺、視覺正向的影響。第三，腦電波與學習表現相關：三種教材刺激下，高分群與低分群，在額葉、中央區、頂葉、枕葉區，產生不同的結果。這些結果作為感官接收與處理的狀態的差異比較。第四，腦電波預測學習成效：閱讀教材正確率76%、聽力教材正確率63%、視聽教材正確率62%，特徵萃取後降至59%。第五，腦電波或感知學習風格量表分別預測七種合適教材，量表正確率70%，腦電波正確率100%。 根據上述結果，本研究針對英語學習提出相關學習策略建議外，FDSM系統作為腦電波作為預測學生學習狀態與教材合適度將不失為一套實用工具。

With advances in EEG technology, research using EEG to study language learning has steadily increased. However, few studies have leveraged EEG signals to analyze the learning process in response to stimuli from different English (second language) learning materials and applied machine learning to predict learning outcomes. This study established a Feature-oriented Data Science Model (FDSM), which primarily embeds feature engineering techniques into data science to analyze EEG signals and English learning performance. The FDSM includes four stages: (1) Data Selection Stage: This study designed an experiment to collect data, recruiting a total of 100 subjects (50 males and 50 females). Each subject participated in four units of teaching materials (reading, listening, and audio-visual), conducting a total of 12 experiments. EEG data were collected throughout the process. (2) Data Preprocessing Stage: EEG noise from 15 subjects was removed, and SMOTE oversampling and GAN simulation were used to address the imbalanced learning performance data. (3) Feature Learning Stage: The data were either directly subjected to machine learning or first underwent feature screening, followed by feature creation and feature extraction. (4) Interpretation and Evaluation Stage: Based on the results, relevant recommendations were made. This study produced five key findings: (1) The overall English learning process is more inclined towards auditory and kinesthetic learning styles. However, when studying English, it shifts towards visual and kinesthetic learning styles. For receiving and processing English information, there is a preference for a visual/non-verbal learning style. (2) Relationship Between Sensory Learning Style and Learning Effectiveness: Learning performance in reading textbooks is positively impacted by visual and tactile learning styles. Listening materials positively impact learning performance in kinesthetic and auditory learners, while audio-visual materials positively impact performance related to both auditory and visual senses. (3) EEG are linked to learning performance: under stimulation from the three types of learning materials, the high-scoring and low-scoring groups exhibited different brainwave activity in the frontal lobe, central region, parietal lobe, and occipital lobe. These results highlight the differences in sensory reception and processing states between the two groups. (4) EEG Prediction of Learning Outcomes: The accuracy of predicting learning outcomes was 76% for reading textbooks, 63% for listening-based textbooks, and 62% for audiovisual textbooks. After feature extraction, the accuracy dropped to 59%. (5) Predicting suitable teaching materials using EEG or perceptual learning style scales: the sensory learning style scale achieved a 70% accuracy rate, while EEG data reached 100% accuracy in predicting suitable teaching materials. Based on these findings, this study provides relevant learning strategy recommendations for English learning. The FDSM system, integrating EEG analysis, will be a practical tool for predicting students' learning status and the suitability of teaching materials.