Année : 2018
Lieu de publication de l'article :

Résumé de l'article

The effectiveness of learning inmassive open online courses (MOOCs)can be significantly enhanced by introducing personalized interven-tion schemes which rely on building predictive models of studentlearning behaviors such as some engagement or performance indi-cators. A major challenge that has to be addressed when buildingsuch models is to design handcrafted features that are effective forthe prediction task at hand. In this paper, we make the first attemptto solve the feature learning problem by taking the unsupervisedlearning approach to learn a compact representation of the rawfeatures with a large degree of redundancy. Specifically, in order tocapture the underlying learning patterns in the content domain andthe temporal nature of the clickstream data, we train a modifiedauto-encoder (AE) combined with the long short-term memory(LSTM) network to obtain a fixed-length embedding for each inputsequence. When compared with the original features, the new fea-tures that correspond to the embedding obtained by the modifiedLSTM-AE are not only more parsimonious but also more discrim-inative for our prediction task. Using simple supervised learningmodels, the learned features can improve the prediction accuracyby up to 17% compared with the supervised neural networks andreduce overfitting to the dominant low-performing group of stu-dents, specifically in the task of predicting students’ performance.Our approach is generic in the sense that it is not restricted to aspecific supervised learning model nor a specific prediction taskfor MOOC learning analytics.Permission to make digital or hard copies of all or part of this work for personal orclassroom use is granted without fee provided that copies are not made or distributedfor profit or commercial advantage and that copies bear this notice and the full citationon the first page. Copyrights for components of this work owned by others than theauthor(s) must be honored. Abstracting with credit is permitted. To copy otherwise, orrepublish, to post on servers or to redistribute to lists, requires prior specific permissionand/or a fee. Request permissions from permissions@acm.org.LAK19, March 4–8, 2019, Tempe, AZ, USA© 2019 Copyright held by the owner/author(s). Publication rights licensed to ACM.ACM ISBN 978-1-4503-6256-6/19/03. . . $15.00https://doi.org/10.1145/3303772.3303795CCS CONCEPTS•Computingmethodologies→Unsupervised learning;Neu-ral networks;Learning latent representations; •Applied com-puting → E-learning;KEYWORDSFeature Learning, Learning Behavior, Unsupervised Learning, Di-mensionality Reduction, Autoencoder, Long Short-Term MemoryACM Reference Format:Mucong Ding, Kai Yang, Dit-Yan Yeung, and Ting-Chuen Pong. 2019. Ef-fective Feature Learning with Unsupervised Learning for Improving thePredictive Models in Massive Open Online Courses. In The 9th Interna-tional Learning Analytics & Knowledge Conference (LAK19), March 4–8, 2019,Tempe, AZ, USA. ACM, New York, NY, USA, 10 pages. https://doi.org/10.1145/3303772.33037951

Mots-clés

Feature selection,