Penn Center for Learning Analytics

Transforming Educational Technology Through Convergence

MOOC Replication Framework

The MOOC Replication Framework (MORF) is being developed to enable the investigation of research questions on MOOCs across multiple data sets. MORF tests whether previously published findings on engagement and completion in MOOCs replicate to new data sets. Currently, MORF has access to over 150 MOOC data sets and is able to test 21 previously published findings on new data sets. We are currently working on improving MORF's user interface and increasing the findings and data sets analyzed.

Identifying Malleable Factors in Blended Learning Environments Using Automated Detectors of Engagement

This project is focused on building automated detectors of behavioral and affective engagement for students using LearnBop educational software. We are exploring how various system and contextual factors relate to engagement and investigating how detected engagement relates to student performance on both within-system assessment and external assessment (e.g., state tests).

Customizing a Digital Learning Platform for Rapid, Low-Cost Research

We are working with the ASSISTments platform to determine new ways to validate the effectiveness of an educational research platform in terms of its scientific impact.

Affective Supports in Physics Playground

This project, in partnership with Physics Playground, examines the impact of affective supports on learning and interest in science, embedded in PP. Their goal is to fuel motivation when students succeed and encourage persistence when they fail. We’ll investigate the effects of various types of affective supports in PP including motivational messages, studying the effects of cognitive and affective supports individually and combined, and different support delivery methods (i.e., game- vs. student-controlled).

Collaborative Research: AquaLab 9: Developing an online game to support science practice learning using adaptive learning progressions

This project, in partnership with the Field Day Lab, explores the production of learning progressions of science practice learning using an educational game context, to understand what order learning experiences should occur in. Using these learning progressions, we can determine how personalized learning interventions can be developed for digital games, using machine learning and educational data mining approaches. This work occurs in the context of the game (under development) AquaLab 9.

Understanding and Enhancing Self-Regulated Learning in Introductory CS Courses

This project will create a fully connected, instrumented introductory Computer Science course that will enable a better understanding and modeling of the connections between conceptual materials, textbook materials, and a programming IDE, and use the resultant data to model learner trajectories between these materials and behavior within them, to understand SRL and metacognition in introductory CS courses better. The project will examine how these behaviors map to students' development of self-efficacy and confidence, and how each of these constructs predicts the development of CS knowledge and skill across the course, as well as longer-term course-taking.

Cyber Infrastructure for Shared Algorithmic and Experimental Research in Online Learning

This project is creating a cyberinfrastructure that will enable external researchers to run large-scale field experiments to improve adaptive learning in both the K-12 ASSISTments platform and in edX and Coursera courses at the University of Pennsylvania. We are also creating tools that will support external researchers in conducting privacy-protected analyses of research data.

Student Affect Detection and Intervention with Teachers in the Loop

This project, in partnership with the ASSISTments platform, will create new approaches where machine learning algorithms that give real-time information to teachers can ask teachers for advice in cases where the algorithm cannot make a decision, using Active Learning machine learning algorithms. The data being used in this project is available here.

Making learning visible: scalable, multisystem detection of self-regulation related to EF

This project, in partnership with the Gold Lab, and the Affect, Cognition, & Computational Lab, will create automated detectors of self-regulated learning behaviors associated with differences in executive function, for multiple learning platforms.

Collaborative Research: Using Educational Data Mining Techniques to Uncover How and Why Students Learn from Erroneous Examples

This project is studying the learning that takes place within erroneous examples, focusing on the role played by confrustion (and how long it occurs for) in whether a student learns from an erroneous example. This work is being conducted in partnership with Decimal Point.

Collaborative Research: Exploring Algorithmic Fairness and Potential Bias in K-12 Mathematics Adaptive Learning

This project is studying how to develop models of student knowledge and affect that are not algorithmically biased. It asks the question of whether current demographic categories are the right frame for thinking about algorithmic bias, or whether groups not specifically considered in the U.S. census may also be highly impacted by algorithmic bias. This work is being conducted in partnership with Nigel Bosch and Carnegie Learning.

Instrumenting the Realizeit Platform.

This project will instrument the Realizeit platform for research on behavioral science interventions, creating an infrastructure that can be used by external researchers at scale. We will enhance support for delivering automated interventions, and for studying their effectiveness when delivered in different situations. As part of this infrastructure, we will develop automated detectors of students' affect as they are using the platform, as well as data analysis tools that facilitate sophisticated statistical and causal inference. This work is being conducted in partnership with Rene Kizilcec.

Collaborative Research: Using Data Mining and Observation to derive an enhanced theory of SRL in Science learning environments.

This project developed a handheld app, QRF, for researchers conducting interviews in classrooms -- when an event of interest involving student self-regulated learning or changes in student affect, the app notifies the interviewer so that they can speak to that student immediately. This enables qualitative researchers to focus their time on the phenomena they are most interested in studying. We used this app to study the interrelationships between self-regulated learning, affect, and students' interactions with the Betty's Brain learning system. This work was conducted in partnership with Luc Paquette and Gautam Biswas. Project Webpage

Linguistic Analysis and a Hybrid Human-Automatic Coach for Improving Math Identity (National Science Foundation, Cyberlearning and Future Learning Technologies)

This project studies an existing hybrid human-automatic learning system used at scale: the GenieMail system within Imagine Learning’s Reasoning Mind platform. We are studying how students’ behaviors in the Reasoning Mind system, demographics, and mathematics skill relate to their math identity. This work will enable researchers to develop proxy measures for math identity that can be used to drive interventions.

Natural Language Processing in Teacher Authored Content

We are researching natural language processing in digital learning spaces - how do teachers author content, and how do students experience and engage with this content? This work centers around studying the features of problems in mathematics contexts, using part-of-speech, bag-of-words, and semantic analysis tools such as WMatrix, CohMetrix, and TAALES, and analyzing the relationships that these features have to student affective states, student performance, and student learning outcomes.

The Downside of Perseverance –Investigating and Moving Students Beyond Unproductive Persistence

This project investigates the role of persistence in an online math learning platform, ASSISTments. The goal of this project is to develop automated detectors that can differentiate between students’ productive and unproductive struggle, in order to better understand when persistence is beneficial. Findings of this project can help inform classroom practices and the design of educational technologies, towards supporting struggling young learners.