AUTHOR=Chen Su , Fang Ying , Shi Genghu , Sabatini John , Greenberg Daphne , Frijters Jan , Graesser Arthur C. TITLE=Automated Disengagement Tracking Within an Intelligent Tutoring System JOURNAL=Frontiers in Artificial Intelligence VOLUME=3 YEAR=2021 URL=https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2020.595627 DOI=10.3389/frai.2020.595627 ISSN=2624-8212 ABSTRACT=

This paper describes a new automated disengagement tracking system (DTS) that detects learners’ maladaptive behaviors, e.g. mind-wandering and impetuous responding, in an intelligent tutoring system (ITS), called AutoTutor. AutoTutor is a conversation-based intelligent tutoring system designed to help adult literacy learners improve their reading comprehension skills. Learners interact with two computer agents in natural language in 30 lessons focusing on word knowledge, sentence processing, text comprehension, and digital literacy. Each lesson has one to three dozen questions to assess and enhance learning. DTS automatically retrieves and aggregates a learner's response accuracies and time on the first three to five questions in a lesson, as a baseline performance for the lesson when they are presumably engaged, and then detects disengagement by observing if the learner's following performance significantly deviates from the baseline. DTS is computed with an unsupervised learning method and thus does not rely on any self-reports of disengagement. We analyzed the response time and accuracy of 252 adult literacy learners who completed lessons in AutoTutor. Our results show that items that the detector identified as the learner being disengaged had a performance accuracy of 18.5%, in contrast to 71.8% for engaged items. Moreover, the three post-test reading comprehension scores from Woodcock Johnson III, RISE, and RAPID had a significant association with the accuracy of engaged items, but not disengaged items.