Get started
Work

Applying Machine Learning and User Experience Design to Solve the Assistance Dilemma in a Logic Tutor.

Year: 2020
Type: article
Abstract: MANIKTALA, MEHAK.Applying Machine Learning and User Experience Design to Solve the Assistance Dilemma in a Logic Tutor.(Under the direction of Tiffany Barnes.)Intelligent tutoring systems (ITSs) have the unique ability to provide individualized instruction to students at a large scale.However, personalized assistance in the form of hints or feedback requires appropriate scaffolding of instruction.In the domain of learning, scaffolding refers to the temporary support provided for the completion of tasks that learners otherwise might not be able to complete.Several researchers identify scaffolding as an interactive process, where the tutor is an active participant in the process of learning.Yet, most ITSs provide assistance upon student request.While some studies have shown improvement in student learning when ITSs provide unsolicited assistance, the field of proactive assistance is relatively unexplored.Providing proactive assistance, where ITSs provide unsolicited assistance in anticipation of future moves, is particularly crucial yet difficult in open-ended domains where the problem space is relatively wide open.To achieve optimal learning, a careful trade-off exists between information sharing and withholding.Providing less assistance may lead to frustration and wasted time, while providing more information may lead to lack of motivation to learn by oneself.This assistance dilemma is a well-recognized challenge in the field of ITS.In this thesis, I present three studies that seek to improve the proactive help usage in Deep Thought, a data-driven tutor for teaching open-ended logic proofs, using machine learning and user-experience methods.In the Hint Interface study, I use user-experience methods to address the problem of help avoidance in Deep Thought, where students rarely request hints and ignore unsolicited hints.The new hint interface for providing unsolicited assistance is designed to increase student attention towards these hints while reducing the cognitive load required to use them.This study is novel in that it shows that hint interface alone can address the problem of help avoidance.The impact of these new graphical hints is compared to the previously used message hints.The results suggest that the new graphical hints foster productive persistence and leads to improved performance for students with low prior knowledge.A generalizable principled-framework is also provided.In the next, Productivity and HelpNeed study, I present a data-driven method to determine step-level productivity for open-ended problem-solving and showcase how it can be used to solve the assistance dilemma in Deep Thought.This approach leverages and modifies the Markov decision processes in the Hint Factory, a data-driven hint generator.I introduce a HelpNeed classification that uses educational literature and the data-driven productivity metrics to determine steps where students need help learning efficient problem-solving strategies.I also present a HelpNeed predictor that predicts unproductivity at the start of each step, and an Adaptive hint policy that uses it to provide proactive hints.I present a controlled study where students in the Adaptive and Control conditions can request hints, but interventions using the predictor are given only in the Adaptive BIOGRAPHY Mehak Maniktala was born in New Delhi, India.With her family's support, she grew up to be a successful computer scientist.She secured a position in one of the top engineering colleges in India -Delhi Technological University and received her Bachelor of Technology in 2015.She worked at SAP for a year and then joined the Master of Computer Science program at North Carolina State University in 2016.During her Master's, she conducted research under Dr. Tiffany Barnes who frequently encouraged Mehak to join the Ph.D. program.Mehak started her Ph.D. in the Fall 2018 semester with a focus on enhancing an intelligent tutor.In her first Ph.D. semester, she was assigned to be the product manager as well as the lead programmer of the tutoring system.Her responsibilities included ensuring a successful end-to-end delivery of the system each semester at two universities: North Carolina State University and the University of North Carolina at Charlotte.She also carried out the initial analysis and processing that serves as the data foundation for all of the IUSE group's collaborative research.She also helped other graduate students integrate their machine learning policies into the system, and mentored several graduate and undergraduate students, enriching their ability to make contributions to their research lab.She has several published articles including two first-author journals in some of the top conferences on AI in Education.She successfully defended her Ph.D. defense in the Fall 2020 semester amidst the Covid-19 pandemic.While the pandemic brought hardships upon everyone, she believes that focusing on the positive can help us adapt and embrace life as is.She enjoys exploring life through new experiences and hopes to encourage others to do the same.deeply grateful that you encouraged me to join the Ph.D. program.I have learned so much through the course of my Ph.D. I would also like to thank Dr. Min Chi for
Language: English
Open Access status: green
OpenAlex ID: W3161922300
FWCI:
Cites: 116
Cited by:
Related to: 10