You are here

Modeling Learner Mood in Realtime through Biosensors for Intelligent Tutoring Improvements

Download pdf | Full Screen View

Date Issued:
2013
Abstract/Description:
Computer-based instructors, just like their human counterparts, should monitor the emotional and cognitive states of their students in order to adapt instructional technique. Doing so requires a model of student state to be available at run time, but this has historically been difficult. Because people are different, generalized models have not been able to be validated. As a person's cognitive and affective state vary over time of day and seasonally, individualized models have had differing difficulties. The simultaneous creation and execution of an individualized model, in real time, represents the last option for modeling such cognitive and affective states. This dissertation presents and evaluates four differing techniques for the creation of cognitive and affective models that are created on-line and in real time for each individual user as alternatives to generalized models. Each of these techniques involves making predictions and modifications to the model in real time, addressing the real time datastream problems of infinite length, detection of new concepts, and responding to how concepts change over time. Additionally, with the knowledge that a user is physically present, this work investigates the contribution that the occasional direct user query can add to the overall quality of such models. The research described in this dissertation finds that the creation of a reasonable quality affective model is possible with an infinitesimal amount of time and without (")ground truth(") knowledge of the user, which is shown across three different emotional states. Creation of a cognitive model in the same fashion, however, was not possible via direct AI modeling, even with all of the (")ground truth(") information available, which is shown across four different cognitive states.
Title: Modeling Learner Mood in Realtime through Biosensors for Intelligent Tutoring Improvements.
7 views
4 downloads
Name(s): Brawner, Keith, Author
Gonzalez, Avelino, Committee Chair
Boloni, Ladislau, Committee Member
Georgiopoulos, Michael, Committee Member
Proctor, Michael, Committee Member
Beidel, Deborah, Committee Member
University of Central Florida, Degree Grantor
Type of Resource: text
Date Issued: 2013
Publisher: University of Central Florida
Language(s): English
Abstract/Description: Computer-based instructors, just like their human counterparts, should monitor the emotional and cognitive states of their students in order to adapt instructional technique. Doing so requires a model of student state to be available at run time, but this has historically been difficult. Because people are different, generalized models have not been able to be validated. As a person's cognitive and affective state vary over time of day and seasonally, individualized models have had differing difficulties. The simultaneous creation and execution of an individualized model, in real time, represents the last option for modeling such cognitive and affective states. This dissertation presents and evaluates four differing techniques for the creation of cognitive and affective models that are created on-line and in real time for each individual user as alternatives to generalized models. Each of these techniques involves making predictions and modifications to the model in real time, addressing the real time datastream problems of infinite length, detection of new concepts, and responding to how concepts change over time. Additionally, with the knowledge that a user is physically present, this work investigates the contribution that the occasional direct user query can add to the overall quality of such models. The research described in this dissertation finds that the creation of a reasonable quality affective model is possible with an infinitesimal amount of time and without (")ground truth(") knowledge of the user, which is shown across three different emotional states. Creation of a cognitive model in the same fashion, however, was not possible via direct AI modeling, even with all of the (")ground truth(") information available, which is shown across four different cognitive states.
Identifier: CFE0004822 (IID), ucf:49734 (fedora)
Note(s): 2013-08-01
Ph.D.
Engineering and Computer Science, Electrical Engineering and Computing
Doctoral
This record was generated from author submitted information.
Subject(s): Artificial Intelligence -- Affective -- Cognitive -- Real-time Systems -- Computer-managed Instruction -- Intelligent Tutoring Systems
Persistent Link to This Record: http://purl.flvc.org/ucf/fd/CFE0004822
Restrictions on Access: public 2013-08-15
Host Institution: UCF

In Collections