Title: Integrating deep learning to improve text understanding in conversation-based ITS
Authors: Sheng Xu; Frank Andrasik; Zhiqiang Cai; Xiangen Hu
Addresses: Central China Normal University, Wuhan, China ' Department of Psychology, University of Memphis, Memphis, TN 38152, USA ' University of Wisconsin - Madison, Wisconsin, USA ' Department of Psychology, The University of Memphis, Memphis, TN 38152, USA
Abstract: In Conversation-based intelligent tutoring systems (CbITS), assessing learners' natural language input is a key factor for the system to be effective. When using AutoTutor, a well-known CbITS, assessments of this type are reduced to evaluating the semantic similarity between learners' inputs and pre-set expectations/misconceptions. Traditional semantic representation methods have prominent inherent limitations, while more advanced deep learning models require large amounts of labelled data which is expensive to obtain. We contend that using deep learning models in concert with an active learning training procedure can reduce the demand for labelled data, thus improving the effectiveness of natural language understanding in CbITS. We report findings from a series of experiments that document how our proposed model was able to significantly outperform traditional models with much fewer labelled data. These findings thus illustrate both the possibility and potential benefits that can be accrued by utilising more advanced semantic representation models.
Keywords: conversation-based ITS; semantic; deep learning; active learning; pretrained language model.
DOI: 10.1504/IJSMARTTL.2021.118908
International Journal of Smart Technology and Learning, 2021 Vol.2 No.4, pp.304 - 324
Received: 25 Apr 2021
Accepted: 02 Sep 2021
Published online: 10 Nov 2021 *