Electronic Theses and Dissertations

Identifier

6089

Date

2017-12-04

Date of Award

2017

Document Type

Thesis (Campus Access Only)

Degree Name

Master of Science

Major

Electrical and Computer Engr

Concentration

Computer Engineering

Committee Chair

Xiangen Hu

Committee Member

Philip Pavlik

Committee Member

Madhusudhanan Balasubramanian

Abstract

Knowledge tracing is one of the major focus of modern artificial intelligence (AI) enabled intelligent tutoring systems (ITS). There have been a large variety of methods applied to modeling student knowledge within the educational data mining community. Not all of the models perform equally well in all varieties of datasets. However, there is no guidance for selecting a method for a particular situation. In this work, we attempted to understand the aspects of datasets that make them suitable for a particular knowledge tracing model, focusing specifically on the widely used Performance Factor Analysis models and current state-of-the-art, Deep Knowledge Tracing (DKT) model. Through recent developments in deep learning, DKT was explored as a method to improve upon traditional methods of analysis, but this method performs poorly with smaller dataset collected from a controlled experiment. We propose structural changes in Long short-term memory (LSTM) model used in DKT. The new LSTM model with deep autoencoder and other important recent performance features as input outperforms current DKT models. This new model is proven to be better capable of tracking learners’ knowledge level.

Comments

Data is provided by the student.

Library Comment

Dissertation or thesis originally submitted to the local University of Memphis Electronic Theses & dissertation (ETD) Repository.

Share

COinS