GetHelp

AI-Powered Help-Seeking in Programming Education

About This Project

This project is an educational data mining research project that investigates how students seek help from AI-powered tutoring systems in NUS’s introductory programming courses. Drawing on rich interaction logs from Coursemology, an AICET-built learning platform, the project examines the frequency, timing, and sequencing of help-seeking behaviours, tracing how students navigate confusion, formulate questions, and respond to AI-generated guidance across the arc of a programming task.

A key methodological contribution is the development of a multi-dimensional analytical framework incorporating AST (Abstract Syntax Tree) distance and other behavioural metrics as proxies for learning progress: by measuring the structural gap between a student’s current code and a correct solution before and after each AI interaction, we can assess whether help-seeking is moving students toward genuine understanding or simply toward a working submission.

In doing so, this project aims to distinguish between AI that scaffolds real learning and AI that merely offloads the cognitive work that learning requires.

Research Questions

What patterns of help-seeking behaviour emerge when students interact with AI tutoring systems in introductory programming, and how do these patterns vary across student profiles?

To what extent does help-seeking lead to genuine learning progress, as measured by AST distance metrics, versus surface-level task completion?

What distinguishes productive from unproductive AI-assisted help-seeking episodes, and how can the tutoring system be redesigned to promote the former?

Key Methodological Contribution

A central contribution of this project is a new way of measuring learning progress using multiple data points. In particular, it uses Abstract Syntax Tree (AST) distance and other behavioural metrics as proxies for learning progress: by measuring the structural gap between a student’s current code and a correct solution before and after each AI interaction, we can assess whether help-seeking is moving students toward genuine understanding or simply toward a working submission. In doing so, this project aims to distinguish between AI that scaffolds real learning and AI that merely offloads the cognitive work that learning requires.

Team Members:

Lead PI:

Research Period:

Liu Liu, Gladia Hotan, Markus Yeo, Lucius Khor

Ben Leong Wing Lup

Aug 2025 – Present