An Adaptive Learning System Based on Real-Time User Emotions
Total Page:16
File Type:pdf, Size:1020Kb
SSStttooonnnyyy BBBrrrooooookkk UUUnnniiivvveeerrrsssiiitttyyy The official electronic file of this thesis or dissertation is maintained by the University Libraries on behalf of The Graduate School at Stony Brook University. ©©© AAAllllll RRRiiiggghhhtttsss RRReeessseeerrrvvveeeddd bbbyyy AAAuuuttthhhooorrr... Sensemo: An Adaptive Learning System Based on Real-Time User Emotions A Thesis presented by Karan Joshi to The Graduate School in Partial Fulfillment of the Requirements for the Degree of Master of Science in Computer Engineering Stony Brook University December 2016 Stony Brook University The Graduate School Karan Joshi We, the thesis committe for the above candidate for the Master of Science degree, hereby recommend acceptance of this thesis Shan Lin - Thesis Advisor Assistant Professor, Department of Electrical and Computer Engineering Timothy J. Driscoll - Second Reader Adjunct Faculty, Department of Electrical and Computer Engineering This thesis is accepted by the Graduate School Charles Taber Dean of the Graduate School ii Abstract of the Thesis Sensemo: An Adaptive Learning System Based on Real-Time User Emotions by Karan Joshi Master of Science in Computer Engineering Stony Brook University 2016 Learning and education is an important part of everyone's life. E-learning systems and smart classrooms have recently become very popular because of the continuous effort to improve learning experience of students. This work studies a new emotion sensing based real-time adaptive learning system using Microsoft Band smartwatch based sensing system. We use three bio-sensors, namely photoplethysmogram (PPG) sensor, galvanic skin response (GSR) sensor and skin temperature sensor, which are available in Microsoft Band 2 to help the system read the bio-signals from a user in real time and unobtrusively. The system is able to detect two indicators of human emotion (valence and arousal) with 79.1% - 84.7% accuracy. We present two usage scenarios for our system- online lecture system and classroom system. In order to improve the learning experience of a user, we propose two types of adaptation/suggestion methods: adapting the speed of the session and adapting the content, depending upon the emotion response of a user in real-time along with student profile. We perform experiments to find the effectiveness of the system. We test our adaptive learning system with 22 participants, and show that the autonomous feedback methods used are effective way of improving user learning experience evident from the minimal user interactions with system and higher score for quizzes after the lessons. iii To my Mom, Dad, Sisters and Family iv Contents Abstract iii List of Figures vii List of Tables viii Abbreviations ix Acknowledgementsx 1 Introduction1 2 Related Work4 2.1 Emotion Recognition..............................4 2.2 Emotion based Adaptive Systems.......................5 2.3 Adaptive Tutoring Systems..........................5 3 Background6 3.1 Electrodermal Activity.............................7 3.2 Blood Volume Pulse and HRV.........................7 3.3 Skin Temperature...............................8 4 Usage Scenarios9 4.1 Online Lecture System.............................9 4.2 Classroom Lecture System........................... 10 5 System Design 11 5.1 Emotion Recognition Subsystem....................... 12 5.2 Feedback Control Subsystem......................... 13 6 System Implementation 15 6.1 Sensor system.................................. 15 6.2 Feature extraction............................... 16 6.3 Emotion Recognition.............................. 17 6.4 Feedback Control................................ 17 6.4.1 PI Controllers.............................. 18 6.4.2 Decision Model............................. 20 v 7 System Evaluation 22 7.1 Platform Evaluation.............................. 22 7.2 Emotion Recognition Evaluation....................... 23 7.2.1 Experimental Setup.......................... 23 7.2.2 Accuracy................................ 24 7.2.3 Reliability and Latency........................ 24 7.3 Control System Evaluation.......................... 26 7.3.1 Response Time............................. 26 7.3.2 Fitness of Model............................ 26 7.3.3 System Acceptance........................... 27 8 Real Deployment 28 8.1 Online Learning System............................ 28 8.1.1 Experimental Setup.......................... 28 8.1.2 Result.................................. 29 8.2 Classroom System............................... 29 8.2.1 Experimental Setup.......................... 30 8.2.2 Result.................................. 30 9 Conclusion 33 9.1 Future Work.................................. 33 A Appendix- Quiz Mergesort 35 B Appendix- Quiz Quicksort 36 C Appendix- Survey 37 D Appendix- Student Profile 38 Bibliography 39 vi List of Figures 5.1 System Model.................................. 12 5.2 Learning Emotional Space........................... 13 5.3 Feedback Control Model............................ 14 6.1 System Components.............................. 15 6.2 Feedback Model................................. 18 6.3 PI Controller.................................. 18 7.1 Physiological response of users to video stimuli............... 25 7.2 Skin temperature response of user to video stimuli............. 25 7.3 Model Fitness.................................. 27 8.1 System Comparison.............................. 29 8.2 Emotional response to First Session..................... 31 8.3 Emotional response to second session..................... 31 8.4 Quiz results for two sessions.......................... 32 vii List of Tables 7.1 Power usage breakdown of application running on Android device..... 23 7.2 Valence Levels................................. 24 7.3 Arousal Levels................................. 24 viii Abbreviations ANS Autonomic Nervous System BVP Blood Volume Pulse EDA Electro Dermal Activity GSR Galvanic Skin Response SCL Skin Conductance Level SCR Skin Conductance Response SVM Support Vector Machine ix Acknowledgements This work would not have been possible without the guidance and support of the fol- lowing people. I would like to thank my adviser Dr. Shan Lin for giving me an opportunity to work with him on this exciting project. It was his continuous guidance and support that kept me motivated throughout the course of my thesis. I would also like to thank Dr. Shahriar Nirjon for his guidance throughout this project. His continuous support has been an essential driver for this work. He and Shan helped make this difficult task really easy and smooth. They have shown me how to approach a problem in a methodical way. I would also like to thank my Thesis second reader Prof. Timothy Driscoll for his time and valuable feedback to improve this work. Lastly but not least, I would like to take this opportunity to thank my parents for their moral support and my sisters for supporting me throughout my thesis. Special thanks to Ritika Joshi for helping with the initial system fabrication, to Arunjot Singh for moral and technical support at times and to Suhas Budhiraja for proofreading this document. This thesis has been a great journey because of all these people and deserve to be sincerely thanked. x Chapter 1 Introduction E-learning systems and smart classrooms have recently become very popular. E-learning systems have provided the means for everyone to learn new things at the comfort of their homes or on the go. It has proved to be essential for people who do not have access to the classrooms or are unable to attend the classes due to their other commitments. This has led to a widespread increase in the usage of the e-learning systems. At the same time there has been increasingly growing interest in improving the current classrooms through the means of technology. The online web based learning systems have come far from their initial conceptual ver- sions. These systems have become more user-friendly and more user-aware as compared to the earlier versions of dictator style teaching. These systems are now more aware of the fact that if the user is being able to absorb the information or not. This is mostly done through the use of quizzes at the end of the lessons to know if the user is able to answer the questions right or not. Today's modern learning systems have used this fea- ture to introduce some degree of adaptivity by using the fact that if the user is not able to get the right answers to the question, that means he is not being able to understand the topic and thus the system now can adapt the level of difficulty of the future content accordingly. This however, does not involve the use of the real-time feedback from the student as it is possible in real world classroom, for example, a student can ask the instructor to explain a topic using an alternative example and can give instructor a verbal/intuitive feedback for him/her to know how to proceed with the lecture. Thus, the instructor is able to see in real-time how the student is feeling about the current lecture. This idea to make the online learning systems at par the traditional classroom systems introduces the possibility to explore use of real-time user feedback to help the system 1 adapt as a classroom instructor would. An emotion aware learning system thus can act as a real world classroom instructor and adapt based on the student's real-time emotions. At the same time such exploration also opens the doors for introducing such feedback from students in a traditional classroom, in more collaborative way, to help the instructor