Comparison of learning analytics and educational data mining using a topic modeling approach
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
The digitalization of education has led to an unprecedented accumulation of data, offering unique opportunities for understanding and optimizing the learning process.This bachelor thesis deeply investigates two related, yet distinct, interdisciplinary fields that leverage this data abundance: Learning Analytics (LA) and Educational Data Mining (EDM).The purpose of this study is to highlight the differences and synergies between them, while simultaneously focusing on the central role of Topic Modeling in the analysis of unstructured educational data.Initially, the thesis delineates each field separately.LA is defined as the measurement, collection, analysis, and reporting of data about learners and their environments, with the ultimate goal of immediate understanding and optimization of learning and teaching.It focuses on providing actionable insights to stakeholders (students, educators, administrators), enabling early intervention and the adaptation of instructional strategies.In contrast, EDM is presented as a field that applies data mining techniques to educational datasets, aiming to discover new, unexpected knowledge, develop sophisticated student models, and systematically improve educational systems.Their fundamental distinction lies in LA's focus on "decision support" and EDM's focus on "knowledge discovery."Particular emphasis is placed on the analysis of Topic Modeling, specifically the Latent Dirichlet Allocation (LDA) algorithm.LDA is presented as a powerful, unsupervised tool capable of revealing the abstract topics pervasive in large collections of unstructured textual data, such as online forum discussions, open-ended exam responses, or written student assignments.The thesis explains how LDA can serve the distinct goals of both LA (e.g., visualizing the dominant discussion topics in a class for immediate pedagogical intervention or identifying topics suggesting a risk of attrition) and EDM (e.g., extracting semantically rich features from texts used to construct student knowledge models or developing adaptive educational content recommendation systems).The practical aspect of the work includes a detailed description of the implementation in a Python environment.The critical steps of text preprocessing are presented-including cleaning, tokenization, stop word removal, and lemmatization.The process of creating a dictionary and corpus using the Gensim library, the training of the LDA model, and Institutional Repository -Library & Information Centre -University of Thessaly 31/03/2026 10:14:38 EEST -67.227.16.72 -x-the evaluation of the extracted topics through metrics such as coherence score and visualization tools like pyLDAvis are also analyzed.In conclusion, the thesis demonstrates that, despite their distinct philosophies, LA and EDM are complementary fields that can significantly benefit from the utilization of topic modeling.The ability to extract structured and interpretable knowledge from unstructured textual data provides a more holistic and profound understanding of the learning process.This critically contributes to the formation of more effective, personalized, and equitable educational systems.The work concludes with suggestions for future research, including the application of the methodologies to large, real-world datasets, comparison with more modern algorithms, and the further investigation of the ethical implications of using educational data.
Fields of Study & Subjects
Online Learning and Analytics
Intelligent Tutoring Systems and Adaptive Learning
Educational and Technological Research
Publication details
- Journal / source
- University of Thessaly Institutional Repository (University of Thessaly)
- Publication date
- 1 January 2026
- Language
- EL
- Document type
- Thesis