| Titre : | Deep learning techniques for improving MOOC systems: Toward Emotion-Aware Learning |
| Auteurs : | Nihel Fatima Baarir, Auteur |
| Type de document : | Thése doctorat |
| Editeur : | Samir, 2026 |
| Format : | 1 vol. (281 p.) / couv. ill. en coul |
| Langues: | Anglais |
| Langues originales: | Anglais |
| Résumé : |
The COVID-19 pandemic marked a major turning point in MOOCs. Despite their rapid expansion, MOOCs continue to face persistent challenges related to learner persistence. EDM has emerged as a central paradigm for analyzing behavioral data generated by E-learning platforms in order to predict student dropout and academic failure. However, the integration of emotional data intopredictive modeling remains largely underexplored.This thesis addresses this gap by investigating the role of emotional featuresin MOOC outcome prediction. The first contribution proposes a predictive framework based on LSTM model that integrates emotional measures collected using the AEQ with clickstream data extracted from the GDP MOOC. Findingsshow that both navigation and emotional data independently provide strong predictive performance. However, integrating both types of features does not significantly improve prediction performance, suggesting that emotions may already be implicitly reflected in learners’ navigation. Building upon this finding, the second contribution introduces a novel paradigm for extracting and predicting emotional states directly from clickstream data. A sequential pattern mining technique is employed for pattern extraction, followed by the design of LSTM-based model for emotion prediction. Two data representations are investigated: a sequential representation using a tailored embedding matrix and a time-series representation. This contribution introduces a new emotion acquisition approach that overcomes the limitaions of existing methods, while deep learning techniques help reduce the time and cost required for manual pattern extraction.This research contributes to adaptive learning systems supporting early psychological intervention and aims to reduce dropout rates while enhancing the effectiveness of MOOCs and E-learning environments. |
| Sommaire : |
Abstract i ڲڪٌۘ i Résumé ii Acknowledgements iii Publications vi List of Figures xi List of Tables xiii Abbreviations xiv 1 General Introduction 1 1.1 Context and Motivation . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.3 Overview of the Contributions . . . . . . . . . . . . . . . . . . . 5 1.4 Thesis Structure . . . . . . . . . . . . . . . . . . . . . . . . . . 6 2 Preliminaries and Basic Concepts 7 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 2.2 Educational Data Mining . . . . . . . . . . . . . . . . . . . . . 7 2.3 Computer-Based Education . . . . . . . . . . . . . . . . . . . . 8 2.3.1 Distance Learning . . . . . . . . . . . . . . . . . . . . . 9 2.3.2 E-Learning . . . . . . . . . . . . . . . . . . . . . . . . . 9 2.3.3 Online Learning . . . . . . . . . . . . . . . . . . . . . . 10 2.3.4 Massive Open Online Courses . . . . . . . . . . . . . . . 10 2.4 Learning Analytics . . . . . . . . . . . . . . . . . . . . . . . . . 13 2.4.1 Cronbach’s Alpha . . . . . . . . . . . . . . . . . . . . . . 13 2.4.2 Pearson Correlation Coeffcient . . . . . . . . . . . . . . 14viii 2.5 Data Mining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 2.5.1 Sequential Pattern Mining . . . . . . . . . . . . . . . . . 15 2.5.2 SPAM Algorithm . . . . . . . . . . . . . . . . . . . . . . 15 2.5.3 Co-occurrence Map SPAM . . . . . . . . . . . . . . . . 16 2.6 Machine Learning . . . . . . . . . . . . . . . . . . . . . . . . . . 17 2.6.1 Classical Supervised Methods . . . . . . . . . . . . . . . 17 2.6.2 Ensemble Learning . . . . . . . . . . . . . . . . . . . . . 19 2.6.2.1 Bagging . . . . . . . . . . . . . . . . . . . . . . 19 2.6.2.2 Stacking . . . . . . . . . . . . . . . . . . . . . 19 2.6.2.3 Boosting . . . . . . . . . . . . . . . . . . . . . 20 2.6.3 Deep Learning . . . . . . . . . . . . . . . . . . . . . . . 20 2.6.3.1 Artificial Neural Networks . . . . . . . . . . . . 21 2.6.3.2 Recurrent Neural Networks . . . . . . . . . . . 25 2.6.3.3 Long Short-Term Memory . . . . . . . . . . . . 26 2.6.4 Sequence Modeling Tasks . . . . . . . . . . . . . . . . . 27 2.6.4.1 NLP and Embedding Layer . . . . . . . . . . . 28 2.6.4.2 Attention Mechanism . . . . . . . . . . . . . . 29 2.6.4.3 Time series Prediction . . . . . . . . . . . . . . 30 2.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 3 Prediction in moocs: State of the Art 32 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 3.2 Student Performance Prediction . . . . . . . . . . . . . . . . . . 33 3.2.1 Dropout risk . . . . . . . . . . . . . . . . . . . . . . . . 36 3.2.2 Academic performance . . . . . . . . . . . . . . . . . . . 37 3.2.3 Learning Style . . . . . . . . . . . . . . . . . . . . . . . 37 3.2.4 Emotion . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 3.3 Features used . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 3.3.1 Demographic Data . . . . . . . . . . . . . . . . . . . . . 40 3.3.2 Clickstream Data . . . . . . . . . . . . . . . . . . . . . . 41 3.3.3 Emotion Data . . . . . . . . . . . . . . . . . . . . . . . . 43 3.3.3.1 Textual data . . . . . . . . . . . . . . . . . . . 43 3.3.3.2 Physiological and visual signals . . . . . . . . . 43 3.3.3.3 Self-reported data . . . . . . . . . . . . . . . . 43 3.3.3.4 Challenges in Emotion Data Collection . . . . 44 3.3.3.5 Cultural and Contextual Influences on Emotion 46 3.4 Machine Learning and Deep Learning in prediction . . . . . . . 47 3.5 Student Behaviour Analysis . . . . . . . . . . . . . . . . . . . . 52ix 3.6 Student Grouping . . . . . . . . . . . . . . . . . . . . . . . . . . 54 3.7 Synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 3.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 4 Investigating the impact of emotions on learners’ success and dropout prediction in a MOOC 58 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 4.2 Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 4.2.1 GDP Dataset . . . . . . . . . . . . . . . . . . . . . . . . 59 4.2.2 OULAD Benshmark . . . . . . . . . . . . . . . . . . . . 60 4.3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 4.3.1 Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . 62 4.3.1.1 Clickstream Cleaning . . . . . . . . . . . . . . 62 4.3.1.2 Emotion Cleaning . . . . . . . . . . . . . . . . 66 4.3.1.3 Augmentation . . . . . . . . . . . . . . . . . . 68 4.3.1.4 Normalization . . . . . . . . . . . . . . . . . . 70 4.3.2 Parallel Phase . . . . . . . . . . . . . . . . . . . . . . . . 70 4.3.3 Combinatorial Phase . . . . . . . . . . . . . . . . . . . . 72 4.3.4 Training and Tuning . . . . . . . . . . . . . . . . . . . . 72 4.3.5 Evaluation Metrics . . . . . . . . . . . . . . . . . . . . . 74 4.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 4.4.1 Clickstream . . . . . . . . . . . . . . . . . . . . . . . . . 76 4.4.1.1 OULAD Clickstream . . . . . . . . . . . . . . 77 4.4.1.2 Tracking Dropout Risk Over Time . . . . . . . 78 4.4.2 Emotion . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 4.4.3 Clickstream + Emotion . . . . . . . . . . . . . . . . . . 82 4.4.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 82 4.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 5 LSTM-Based Prediction of Emotional Patterns in MOOCs from Behavioral Data 86 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 5.2 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 5.2.1 Pattern Extraction . . . . . . . . . . . . . . . . . . . . . 88 5.2.2 Clickstream Preparation . . . . . . . . . . . . . . . . . . 91 5.2.2.1 Cleaning and Standarization . . . . . . . . . . 92 5.2.2.2 Formatting Inputs for LSTM . . . . . . . . . . 92 5.2.2.3 Data Augmentation . . . . . . . . . . . . . . . 94 5.2.3 Model Training . . . . . . . . . . . . . . . . . . . . . . . 96x 5.2.3.1 Tuning . . . . . . . . . . . . . . . . . . . . . . 97 5.2.3.2 Evaluation Metrics . . . . . . . . . . . . . . . . 98 5.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 5.3.1 Positive Emotion . . . . . . . . . . . . . . . . . . . . . . 99 5.3.2 Negative Emotion . . . . . . . . . . . . . . . . . . . . . 105 5.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 5.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 6 General Conclusion 110 A Validation of Emotional Measures 113 A.1 Global Coherence . . . . . . . . . . . . . . . . . . . . . . . . . . 113 A.2 Individual Coherence . . . . . . . . . . . . . . . . . . . . . . . . 115 A.2.1 Internal Validation . . . . . . . . . . . . . . . . . . . . . 115 A.2.1.1 Cosine Similarity . . . . . . . . . . . . . . . . . 115 A.2.1.2 Range . . . . . . . . . . . . . . . . . . . . . . . 116 A.2.2 External Validation . . . . . . . . . . . . . . . . . . . . . 11 |
| Type de document : | Thése doctorat |
Disponibilité (1)
| Cote | Support | Localisation | Statut |
|---|---|---|---|
| TINF213 | Mémoire | bibliothèque sciences exactes | Consultable |
