Motivation is important for learners. As the provision of education moves increasingly towards online delivery, keeping motivation high is a key challenge. The lack of personalized approaches traditionally delivered by human tutors increases drop out rates. This thesis investigates how a conversational agent, taking the role of a virtual tutor, could deliver personalised feedback on performance to learners. It also investigates the most effective emotional support to incorporate in this feedback in order to maintain learner motivation. How learners respond to feedback depends on many factors. We focus on learner personality as our adaptation criterion, in particular Generalized Self Efficacy (GSE) and the Five Factor model (FFM). First, based on the literature, the thesis establishes how the core concepts of motivation, affective state, feedback and personality relate to one another, and discusses other research into learner motivation and feedback. Our main methodology is the User-as-Wizard method where we model how people adapt to personality in the real world when giving feedback to learners. An algorithm is then developed to encapsulate these adaptations, and is then evaluated. To achieve this, we required a way to express different learner personalities in a controlled way. Personality can be described by many models, but one of the most popular and reliably validated is the Five-Factor Model. In this model, the personality of an individual is described by a score for each of the five dimensions or 'traits'. There are numerous self-report questionnaires for these traits, indicated by a measure of agreement with certain adjectives on a scale. To express the traits at polarized levels, we produced and validated ten stories, two for each of the traits, expressing only that trait at high and low level. These stories were based on the adjectives used in the IPIP-NEO questionnaire. We also created two polarized stories for Generalized Self Efficacy. Subsequently, we investigated how people use different slants (or bias) in performance feedback, depending on learner personality. We designed two experiments, in which participants took the role of a teacher. Participants were shown the learner's personality (through the personality story) and a set of test scores on a range of topics. We provided different ways of describing the learner's performance which could result in positive, neutral or negative slants (e.g. “you are slightly below my expectations” on a score of 10% is positive, and “you are substantially below my expectations” is neutral). The type of slant was established by a focus group of experts prior to the experiments. We found some evidence that slanting was used for very low test scores for students with low GSE, and positive slanting for conscientious students who had only just failed a test. Next, we investigated supplementing slant with emotional support. A set of emotional support statements was generated and categorized. A series of five experiments was run where participants gave feedback to students with differing personality traits (using the FFM stories) and test scores. Participants could provide the same performance feedback (with associated slant) from the previous experiments, and could choose to supplement this with the validated emotional support statements. The type of emotional support given did indeed vary between different personalities (e.g. neurotic individuals with poor grades received more emotional reflection), and an algorithm was created to describe these adaptations. Finally, we ran a qualitative study with teachers and students to investigate the algorithm's effectiveness. During the course of the thesis we also developed a methodology for generating, validating and investigating the use of Emotional Support for other domains. This has already been applied in research to persuade older adults to participate in social interactions and to support Community First Responders and carers when experiencing various stressors.
Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:633286 |
Date | January 2014 |
Creators | Dennis, Matthew Gordon |
Publisher | University of Aberdeen |
Source Sets | Ethos UK |
Detected Language | English |
Type | Electronic Thesis or Dissertation |
Source | http://digitool.abdn.ac.uk:80/webclient/DeliveryManager?pid=215571 |
Page generated in 0.0021 seconds