Page 87 - Undergraduate Research Opportunities Program Proceedings 2017-18

School of Engineering
Department of Computer Science & Engineering

who need help because of the lack of face-to-face communication. On one hand, online instructors have to know
students’ learning states to offer help to them. On the other hand, developers need a data visualization scheme to
design better prediction and analysis models. In this paper, a data visualization framework is developed to cater for both
needs. It visualizes students’ learning trajectories through all chapters in a 2D plot, which gives an intuitive
representation of their learning state transition.

Supervisor: YEUNG Dit Yan/ CSE
Student: WANG Ziyu / COSC Course: UROP1100, Fall
The growing popularity of online courses has been continuously attracting research interests in developing the E-
Learning Platforms. Recently, most platforms provide identical service for all students, regardless of the difficulties they
may encounter during study. As a result, many students give up and cannot pass the course they enrolled. Personalized
E-Learning Platforms, which provide particular aids for students based on their performances, aim at improving their
learning motivation and efficiency. This report will focus on student performance prediction on Open edX platform. The
features reflecting students’ activeness will be extracted continuously throughout the entire semester. Thus, a recurrent
neural network with Long short-term memory (LSTM) cells is considered as a suitable prediction model for this project.

Machine Learning for Modeling Student Behavior on e-Learning Platforms
Supervisor: YEUNG Dit Yan/ CSE
Student: LU Xiuyuan / COMP Course: UROP1100, Summer
With the development and popularity of computer and internet, e-learning platforms have greatly affected today’s
education pattern, which can be treated as a revolution in the education area. Students can study at home conveniently
with the help of such platforms. However, studying at home means that there will be no teacher helping them and
students can only go through the same materials by themselves and do some practice which greatly limits the outcome.
To provide students with some personalized education in accordance with their own study pattern, we need to
reconstruct their thinking process and offer personalized education to them. But because of the lack of teachers, we can
only restore students’ thinking process based on their action on the computer. This project focuses on designing
distance formula between students’ event sequences and clustering students into different groups.

Machine Learning for Building AI Teaching Assistants

Supervisor: YEUNG Dit Yan/ CSE
Student: YANG Handi / CPEG-HR Course: UROP1000, Summer

For the task of building a robust AI teaching assistant, effective algorithms that automatically recognize a duplicate
question from the question bank have a crucial role to play in labour saving. In this report, several state-of-the-art
approaches to the duplicate question identification are presented and analysed. Starting from BoW model that doesn’t
care about word orders, inverse term frequency helps raise its performance a lot. Then the best result from the Siamese
architecture, test accuracy of 0.8687, is achieved by the bidirectional GRU with attention mechanism. A non-Siamese
approach that enables sentence interaction in the forward phase is also tested with CNN and gets an accuracy of 0.8132.
Possible future explorations are discussed in the end.

Supervisor: YEUNG Dit Yan/ CSE
Student: YANG Hsuan-ru / COMP Course: UROP1100, Summer

77