A conceptual framework for an affective tutoring system using unobtrusive affect sensing for enhanced tutoring outcomes

Abstract

Affect plays a pivotal role in influencing the student’s motivation and learning achievements. The ability of expert human tutors to achieve enhanced learning outcomes is widely attributed to their ability to sense the affect of their tutees and to continually adapt their tutoring strategies in response to the dynamically changing affect throughout the tutoring session. In this thesis, I explore the feasibility of building an Affective Tutoring System (ATS) which senses the student’s affect on a moment-to-moment basis with the use of unobtrusive sensors in the context of computer programming tutoring. The novel use of keystrokes and mouse clicks for affect sensing is proposed here as they are ubiquitous and unobtrusive. I first establish the viability of using keystrokes and contextual logs for affect sensing first on a per exercise session level and then on a more granular basis of 30 seconds. Subsequently, I move on to investigate the use of multiple sensing channels e.g. facial, keystrokes, mouse clicks, contextual logs and head postures to enhance the availability and accuracy of sensing. The results indicated that it is viable to use keystrokes for affect sensing. In addition, the combination of multiple sensor modes enhances the accuracy of affect sensing. From the results, the sensor modes that are most significant for affect sensing are the head postures and facial modes. Nevertheless, keystrokes make up for the periods of unavailability of the former. With the affect sensing (both sensing of frustration and disengagement) in place, I moved on to architect and design the ATS and conducted an experimental study and a series of focus group discussions to evaluate the ATS. The results showed that the ATS is rated positively by the participants for usability and acceptance. The ATS is also effective in enhancing the learning of the students