140 likes | 315 Views
Development of an Affect-Sensitive Agent for an Intelligent Tutor for Algebra. Thor Collin S. Andallaza August 4, 2012. Introduction. Computer Agent
E N D
Development of an Affect-Sensitive Agent for an Intelligent Tutor for Algebra Thor Collin S. Andallaza August 4, 2012
Introduction • Computer Agent • A program embedded within a certain environment capable of autonomous action to achieve its design objectives (Padgham and Winikoff 2004, Wooldrige and Jennings 1995) • Embodied Conversational Agent (ECA) • A computer interface which exhibits humanlike conversational behavior (Cassell et al. 2000) • Intelligent Tutoring System (ITS) • A computer application that is capable of providing individualized instruction to learners through the use of artificial intelligence, thereby supporting the learner and facilitating the learning process (Nwana 1990)
Introduction • Aplusix • An intelligent tutor for algebra • Features an advanced editor that allows for step-by-step solutions to problems • Provides visual feedback on student progress • Domain-based agents for hints or final answer to the problem 3
Research Objective • To have a significant influence in enhancing the learning experience of students when using an ITS such as Aplusix • To determine what considerations will be needed in order to design, implement, develop, and test a motivational agent that can interact with the student on a real time basis • Significant features for detection • Integration of models and responses • Evaluating learning experience, especially motivation, of students when using the agent
Previous Work • First version of the ECA for Aplusix (Andallaza and Jimenez 2012) • Based from previous work • Affective and Learning Profiles (Lagud 2010) • Detecting Off-task Behavior (Bate 2010) • Framework for Developing Motivational Agents (Lim 2010) • Third-party application ran alongside Aplusix • Real time analysis of student affect using student models • Agent avatar, script of responses, and text-to-speech capability • Initial test run with high school students • Able to evaluate student affect • Not able to effectively motivate students
Current Work • Modeling Affective States • Initial attempt to improve the existing student models used by the Aplusix ECA more effective at motivating students • A refined analysis of student interaction logs using linear regression • Results • None of the models were usable • average no. of steps used in four of the five remaining models • Consistent with findings from previous work (Lagud 2010) • More steps taken is indicative of boredom or confusion, while less steps is indicative of engagement
Future Work • Student Models • Better feature engineering techniques • Generating fine-grained data • Splitting raw data into smaller time windows for more timely evaluations • Application of Bayesian Knowledge Tracing (Corbett and Anderson 1995) on data analysis • Agent Interface • Actual Field Test
Acknowledgements • DOST PCIEERD, “Development of Affect-Sensitive Interfaces” grant • Dr. Ma. Mercedes T. Rodrigo and the Ateneo Laboratory for the Learning Sciences • Department of Information Systems and Computer Science, Ateneo de Manila University
Development of an Affect-Sensitive Agent for an Intelligent Tutor for Algebra Thor Collin S. Andallaza August 4, 2012