ABSTRACT
Socrates once stated: “Education is the kindling of a flame, not the filling of a vessel.” Intelligent tutoring systems use computational models to efficiently “fill the vessel.” However, research is limited on how these systems can enable “kindling of a flame.” To explore motivation-adaptive learning, this study assesses the learning impact of an individual’s sentiments (emotional associations) on three learning outcomes: information recall, intrinsic motivation, and engagement. Seventy volunteers took two computer-based tutors and provided self-report measures throughout their learning. The learning impacts of topic sentiment and learning-medium sentiment were measured separately and compared. For both topic and learning-medium sentiment, results showed positive linear relationships between net sentiments and intrinsic motivation and net sentiments and engagement. A negative linear relationship between negative sentiments and information recall was also identified. Findings were summarized, and four computer-based instruction design recommendations were provided. Recommendations include: to use learner emotion data in the form of sentiments to better understand learning outcomes, to align sentiment measurement strategies with the tutor’s purpose, to account for the impact of prior sentiments, and to monitor sentiment change.
CHAPTER 1: INTRODUCTION
When preparing to take a standardized test such as the GRE, some people are willing to pay hundreds and possibly thousands of dollars for their own private tutor. This is reasonable, as one-on-one tutoring has been shown to be more effective than classroom instruction (Bloom,
1984). A one-on-one tutor can personalize lessons to a student’s personal needs and adapt instruction methods to a student’s mental state. The effectiveness of one-on-one tutoring has sparked advancement into intelligent tutoring systems (ITSs) research. ITSs are computer-based tutors that provide immediate and customized instruction or feedback to users. ITSs have been shown to be effective in many contexts and have even been implemented in schools (Kulik and Fletcher, 2016).
Modern ITSs seek to replace human tutors (Bloom, 1984; Ma, Adesope, Nesbit, & Liu, 2014), but can an ITS compete with a human teacher in affective areas like emotion sensing, motivating, and engaging? Today, ITSs can identify and adapt to user emotions due to advances in emotion detection software and devices (Arroyo et al., 2009; Woolf, Arroyo, Cooper, Burleson, and Muldner, 2010). These technologies include facial expression sensors, mouse pressure sensors, and posture sensors. Additionally, through applying insights such as attitudes, personality, and goals from user profiles to a theory of emotion formation, tutors are also able to predict potential emotional states in different learning situations (Sottilare, Graesser, Hu, and Holden, 2013).
Despite advances in emotion recognition and adaptation, ITSs have not learned to recognize, predict, and adapt to learner intrinsic motivation and engagement. A human tutor can sense if a student will be intrinsically motivated and engaged in a certain lesson. An ITS cannot do this currently and, therefore, cannot adapt accordingly to foster intrinsic motivation and engagement. Motivation-adaptive learning presents a major opportunity for ITSs to improve the student’s learning performance, learning experience, and drive to continue learning the material.
Fortunately, progress in emotion sensing may have unlocked insight into predicting intrinsic motivation and engagement. When discussing emotion-adapting ITS features in her talk Building an Affective Learning Companion, Rosalind Picard predicted that emotion-adaptation features “can contagiously excite learners with passion for a topic, leading to greater efforts on the part of the learner to master the topic” (Picard, 2006). Picard’s prediction invites investigation into how learner emotions relate to intrinsic motivation and engagement.
In a computer-based instruction context, this research aims to explore how sentiments, which are long-term emotional associations based on prior emotional experiences, relate to intrinsic motivation and engagement theories. I begin by investigating current ITS design recommendations and emotion sensing capabilities. Next, I define sentiment and distinguish its qualities among similar constructs. Finally, I discuss current motivation and engagement research in the context of self-determination theory (Ryan and Deci, 2002) and flow theory (Nakamura and Csikszentmihalyi, 2014). Through a computer-based tutoring experiment, I then test whether learner sentiments have a linear relationship with intrinsic motivation, engagement, and information recall ability.
IMPACTS OF USER SENTIMENT ON INFORMATION RECALL, INTRINSIC MOTIVATION, AND ENGAGEMENT IN THE CONTEXT OF INTELLIGENT TUTORING SYSTEMS