Modelling e-learner comprehension within a conversational intelligent tutoring system

Holmes, M, Latham, AM, Crockett, K and O'Shea, J (2018) Modelling e-learner comprehension within a conversational intelligent tutoring system. In: 11th IFIP TC 3 World Conference on Computers in Education (WCCE 2017), 03 July 2017 - 06 July 2017, Dublin, Ireland.

Preview

Accepted Version Available under License In Copyright. Download (639kB) | Preview

Abstract

Conversational Intelligent Tutoring Systems (CITS) are agent based e-learning systems which deliver tutorial content through discussion, asking and answering questions, identifying gaps in knowledge and providing feedback in natural language. Personalisation and adaptation for CITS are current research focuses in the field. Classroom studies have shown that experienced human tutors automatically, through experience, estimate a learner’s level of subject comprehension during interactions and modify lesson content, activities and pedagogy in response. This paper introduces Hendrix 2.0, a novel CITS capable of classifying e-learner comprehension in real-time from webcam images. Hendrix 2.0 integrates a novel image processing and machine learning algorithm, COMPASS, that rapidly detects a broad range of non-verbal behaviours, producing a time-series of comprehension estimates on a scale from -1.0 to +1.0. This paper reports an empirical study of comprehension classification accuracy, during which 51 students at Manchester Metropolitan University undertook conversational tutoring with Hendrix 2.0. The authors evaluate the accuracy of strong comprehension and strong non-comprehension classifications, during conversational questioning. The results show that the COMPASS comprehension classifier achieved normalised classification accuracy of 75%.