HCII PhD Thesis Defense: Kexin Yang
When
-
Description
Towards more customized, adaptive and reflective learning environments through human-AI collaboration: Design and evaluation of analytic-driven teacher support tools
Kexin Bella Yang
HCII PhD Thesis Defense
Time and Location
Thursday, Aug 14, 2025 @ 10 am EDT
Newell-Simon Hall (NSH) 4305 + Zoom
Zoom Link:
Meeting ID: 998 820 7001
Passcode: 516516
Zoom link: https://cmu.zoom.us/my/kexiny?pwd=JkNablTadgbkUuUCUbmpydTPeeOZID.1&omn=96007344011
Thesis Committee
Dr. Vincent Aleven (HCII, CMU), Co-chair
Dr. Nikol Rummel (Educational Psychology and Technology, RUB & HCII, CMU), Co-chair
Dr. Kenneth Holstein (HCII, CMU)
Dr. Xu Wang (Computer Science and Engineering, University of Michigan)
Abstract
Classroom teaching is a complex, multi-phase process that involves careful preparation, real-time adaptation, and post-class reflection. Effective support for teachers should ideally span all these phases. Advances in educational technology and multimodal learning analytics (MMLA) hold promise to enhance teaching and learning by enabling teachers to focus on what they do best—adapting instruction to students’ needs. My PhD research explores how human-AI collaboration can be leveraged to design, develop, and evaluate tools and pipelines that support teachers in creating more customized, differentiated, and reflective learning environments.
Customizing Instructional Materials and Technology. To support instructional customization, I investigated both content (instructional hints) and technology (pairing algorithms). For instructional materials, I designed a crowdsourcing pipeline that enables scalable hint customization with minimal expert involvement. Findings revealed that crowd contributors needed better scaffolding to meet teachers’ specific instructional styles and goals. For technology, I developed SimPairing, using data simulation to assess the feasibility of dynamic learning group policies. Results showed that dynamic transitions between different activities have pedagogical value in teachers' eyes and are feasible. varied across classroom contexts, highlighting the importance of aligning algorithmic policies with real-world constraints.
Dynamically Differentiated Learning. While individual and collaborative learning each offer benefits, prior work lacked systems to support dynamic transitions between the two. My research contribute to a technology ecosystem that enables such transitions. Through classroom studies, I found that teachers valued the flexibility and individualization offered by dynamic pairing, though challenges remain around optimal timing, grouping, and alignment with instructional goals. I also discovered findings related to comparing dynamic pairing versus standard pairing, as well as teacher-students' respective preferred level of agency in such settings.
Supporting Reflective Teaching with MMLA. To promote reflective teaching, I studied how Reflecto, a teacher-facing tool that integrates multimodal data (e.g., location sensing, log data) may support teacher reflection. Prior literature has emphasized the value of teacher reflection for instructional improvement, but less is known about how to design reflection tools that align with teacher routines and privacy preferences. Through analyzing over 100 teachers' surveys, I investigated how teachers interact with MMLA-based tools collaboratively, what analytics teacher value the most, with whom they want to share data and privacy considerationss. Later through tool prototyping and eventually an exploratory classroom study in school, I studied how Reflecto impact teachers' understanding and behavior of their teaching, through collaborative reflection using Reflecto.
Across these threads, my dissertation uses mixed methodologies—ranging from pipeline design and data simulation to user-centered design and in-situ experimentation. The work advances understanding at the intersection of AIED and HCI by illustrating how human-AI systems can support teachers in three key areas: (1) customizing instructional materials and technologies, (2) enabling dynamically differentiated learning, and (3) fostering critical reflection through MMLA. Together, these contributions support the broader goal of creating more responsive and human-centered classroom environments.
Dissertation: https://drive.google.com/file/d/1yeGnZH6knoC68vycBh0iqanguFCwT2J3/view