Research Projects

This project is funded as part of the Digital Learning Platforms Network (IES # R305N240049) in partnership with with OpenStax-Rice University and researchers from the University of Pennsylvania and the University of Denver. This research investigates how interactive tools in digital textbooks may enhance student learning by developing scalable interventions that help students use self-regulated learning (SRL) strategies more effectively, improving their comprehension and knowledge retention.

We will analyze learner data from multiple digital textbooks and build SRL detectors using cutting-edge educational data mining and machine learning techniques. Co-design sessions with students and instructors will guide the development of scaffolds aimed at enhancing SRL behaviors. The project will include rapid rounds of A/B testing to assess the impact of these interventions on learning outcomes, with a focus on equity and mitigating algorithmic bias. Findings will inform the design of digital learning tools and offer practical insights for stakeholders across educational platforms.

The Coding Like a Data Miner project is a data science-based computer science curriculum funded by the National Science Foundation (NSF #2137708) that uses culturally relevant and responsive pedagogies to constructively center diverse learners in STEM. The curriculum was co-created through participatory curriculum design sessions with educators and youth stakeholders, and has been adapted by the Philadelphia Science Center.

The curriculum emphasizes student access to datasets from the social media platform Twitter/X, providing access to data sources that are both familiar to social media natives and expansive across topics and opinions, enabling users to engage with personally and culturally relevant interests. The curriculum was structured around the use of Twitter/X's Application Programming Interface (API), providing an authentic sandbox-like environment (Sandbox Data Science) that retains material, personal and disciplinary agency for learners to go beyond their typical roles as consumers of information to actively serve as producers of knowledge on their own terms with real, complex, and messy data.

Large Language Models (LLMs) are transforming the landscape of qualitative research by enhancing how we analyze data. While natural language processing methods have supported qualitative analysis for years, LLMs offer new levels of precision and scalability. Our team of researchers at the Penn Center for Learning Analytics aims to foster a community of researchers exploring the use of LLMs for qualitative coding, focusing on how these tools can be integrated into existing methodologies to streamline processes like developing thematic codebooks and applying codes to data. Areas of focus include ensuring reliability and validity and exploring how LLMs can improve the efficiency of qualitative analysis without losing the depth and nuance that human researchers bring.

ChatGPT for Automated Qualitative Coding

Automating Inductive Codebook Development

In Press: Prompting Techniques for Automated Coding

In Preparation: LAK25 Workshop - LLMs for Qualitative Analysis in Education

The goal of What-If Hypothetical Implementations in Minecraft (WHIMC) is to develop computer simulations that engage, excite, and generate interest and engagement in STEM. WHIMC leverages Minecraft Java Edition as a learning environment for learners to interactively explore the scientific consequences of alternative versions of Earth via “what if?” questions, such as “What if the earth had no moon?” Learners explore these worlds as aspiring scientists and engineers on an interactive server.

Researchers at the University of Pennsylvania (NSF #2301173) have partnered with the University of Illinois at Urbana-Champaign to use machine learning techniques to detect moments of engagement or disengagement in real-time. Using the Quick Red Fox paradigm to trigger alerts for researchers when students exhibit behaviors tied to interest, we can interview learners about their experiences before they are forgotten or reinterpreted. We will triangulate findings from interviews, activity logs, and surveys of STEM knowledge and interest to better understand how diverse students engage with STEM content and develop sustained interest in games for learning.

The work is part of research in the Penn Center for Learning Analytics on the MOOC Replication Framework (MORF). Our goal is to build tools that allow researchers to explore patterns in qualitative data from Massively Open Online Courses (MOOCs) using Quantitative Ethnographic techniques such as Epistemic Network Analysis while ensuring data security and privacy.

Resources in progress:

MORF ENA Web tool

MORF ENA User Guide

In Press: The Impacts of COVID-19 on Student Interaction in a Massive Open Online Course

Researchers in the Games and Learning in Interactive Digital Environments (GLIDE) lab at Drexel University partnered with researchers in the Epistemic Analytics Group at the University of Wisconsin-Madison to redesign the virtual environment Land Science to enhance middle school students' exploration of science and STEM career identities. Three design-based iterations of Philadelphia Land Science and supportive curriculum were implemented through Science Leadership Academy programming at The Franklin Institute in Philadelphia, PA. Student identity exploration was assessed using Quantitative Ethnographic approaches.

My dissertation research involved similar assessments of player exploration of science identities through discussion board posts made the the commercial model rocketry Simulation game Kerbal Space Program.

STEM Identity Exploration in Philadelphia Land Science

STEM Identity Exploration in Kerbal Space Program