Treffer: Temporal Skill Discovery for Modeling Student Learning Progression across STEM Domains

Intelligent Tutoring Systems (ITSs) have emerged as valuable systems to promote active learning. It is critical to build accurate student models to support the learning process. In order to provide efficient and effective personalized instructions for students, tracking a student's time-varying knowledge state is essential to an ITS. Prior research has often relied on domain experts and/or educators to identify problem-specific skills from correct solutions, ignoring the temporal information involved in the student learning process. In this dissertation work, we focus on modeling student learning progression by considering temporal skill discovery across two types of STEM domains. One type is well-defined ITSs including "tutor-driven physics" tutor Cordillera and probability tutor Pyrenees; the other is open-ended programming environments including "self-paced" iSnap for block-based programming and CodeWorkout for classic text-based Java. Our research was carried out in four thrusts described as follows: Thrust 1--Two well-defined domains. We explored the impact of a Skill Discovery (SK) method, which can both leverage expert-designed skills and also discover new skills from student performance data automatically. We incorporated SK with various student models including two classic Bayesian models Bayesian Knowledge Tracing (BKT), Intervention-BKT (IBKT), and one popular deep learning model, Long Short Term Memory (LSTM), and evaluated them on two important student modeling tasks: post-test and learning gain prediction, across two "well-defined domains"--Cordillera and Pyrenees. Thrust 2--One well-defined and one open-ended domain. We first investigated the prediction power of Recent Temporal Pattern Mining (RTP), which is designed to extract interpretable, meaningful temporal patterns from irregularly-sampled multivariate time series data; and then we built an effective learning approach by considering time-awareness to model the dynamics of student knowledge state in continuous time, for both Pyrenees and iSnap. Thrust 3--Two open-ended domains. We proposed a data-driven method named Temporal-ASTNN to address temporal skill discovery during the long-longitude learning process in open-ended programming domains. Temporal-ASTNN combines a novel neural network model based on abstract syntactic trees (AST), named ASTNN, and LSTM model. ASTNN handles the "linguistic" nature of student programming code, and LSTM handles the "temporal" nature of student learning progression. Thrust 4--Domain Adaption. We further proposed a domain adaptation framework to leverage the data from different programming domains using the cross-lingual adversarial domain adaptation framework (CrossLing). And our results showed that the proposed framework is able to improve the performance of the original ASTNN in the task of programming classification. Then we extended the proposed CrossLing to handle temporal sequences by combining it with LSTM and Time-aware LSTM (TLSTM) and compared them with other models with different skills. The contributions of this dissertation are fourfold. First, we incorporate an automatic temporal skill discovery method to various knowledge tracing methods in two well-defined ITSs and achieve improvements over the prediction especially the early prediction of post-test scores. Second, we investigate the effectiveness of a time-aware model, T-LSTM in both well-defined and open-ended domains, through two different prediction tasks. Third, we extensively explore student modeling in open-ended domains and our proposed Temporal-ASTNN is able to outperform other models in the early prediction task. Finally, we propose a new domain adaptation framework across different programming domains to leverage the common knowledge, which is shown to achieve better performance for both programming classification and the early prediction of student success. Therefore, our framework can potentially shed some light on better understanding student learning progression and providing timely interventions and educational treatments for students. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://www.proquest.com/en-US/products/dissertations/individuals.shtml.]

As Provided