Category: NC State

Four papers are accepted at EDM21 in Paris, France. The topics of the papers cover student sub-goal detection, student performance prediction, bug detection, and detection of progress/struggling moments in CS courses for novice programmers.

Yang will be presenting a paper “Toward Semi-Automatic Misconception Discovery Using Code Embeddings” this year at LAK. The paper uses deep learning methods to extract embedding vectors from student code submissions in an introductory programming class in a US university, and use these embeddings to detect possible misconceptions by clustering them in the vector space. The results suggest that code embeddings can be used to find misconceptions that cannot be detected by normal methods such as tree edit distances.

HINTS lab has received $6,000 research credits from Amazon Web Services for the research of Representation Learning from CS Education Data. This research fund, granted to Dr. Price and Yang will be used to support intensive computing in leveraging machine learning models for CS education research topics.

Emma will be presenting a paper “Crescendo: Engaging Students to Self-Paced Programming Practices” this year at SIGCSE. Crescendo is a programming practice system that help students practice isolated programming concepts one at a time, in a block-based programming environment (Snap!). It uses Parsons problems to reduce problem complexity, Use-Modify-Create task progressions to scaffold a programming concept, as well as automatic assessments to generate immediate feedback. Students from a Summer Camp and a CS0 university classroom used Crescendo as introductory practices and engaged in using the system.

We’ll be presenting two papers at EDM this year. The first, “Toward Data-Driven Example Feedback for Novice Programming,” explores generating adaptive example-based feedback, which presents a partial solution to a programming problem when a student is stuck. His results suggest that by leveraging student data, we can generate higher-quality, more adaptive examples than just using an expert solution, but the results may depend on a student’s ability to select which feature they want to see completed. The second, ”One minute is enough: Early Prediction of Student Success and Event-level Difficulty during a Novice Programming Task,” is a collaboration with the D3 and Game2Learn labs at NCSU, and presents a model for predicting whether a student will succeed at a given programming problem. The model is able to predict with impressively high accuracy with only 1 minute of data on a 20+ minute programming task.