Search

Automatically identifying struggling students learning to program can assist teachers in providing timely and focused help. This work presents a new deep-learning language model for predicting "bug-fix-time", the expected duration between when a software bug occurs and the time it will be fixed by the student. Such information can guide teachers' attention to students most in need. The input to the model includes snapshots of the student's evolving software code and additional meta-features. The model combines a transformerbased neural architecture for embedding students' code in programming language space with a time-aware LSTM for representing the evolving code snapshots. We evaluate our approach with data obtained from two Java development environments created for beginner programmers. We focused on common programming errors which differ in their difficulty and whether they can be uniquely identified during compilation. Our deep language model was able to outperform several baseline models that use an alternative embedding method or do not consider how the programmer's code changes over time. Our results demonstrate the added value of utilizing multiple code snapshots to predict bug-fix-time using deep language models for programming. [For the complete proceedings, see ED630829.]

This paper studies the use of Reinforcement Learning (RL) policies for optimizing the sequencing of online learning materials to students. Our approach provides an end to end pipeline for automatically deriving and evaluating robust representations of students' interactions and policies for content sequencing in online educational settings. We conduct the training and evaluation offline based on a publicly available dataset of diverse student online activities used by tens of thousands of students. We study the influence of the state representations on the performance of the obtained policy and its robustness towards perturbations on the environment dynamics induced by stronger and weaker learners. We show that 'bigger may not be better', in that increasing the complexity of the state space does not necessarily lead to better performance, as measured by expected future reward. We describe two methods for offline evaluation of the policy based on importance sampling and Monte Carlo policy evaluation. This work is a first step towards optimizing representations when designing policies for sequencing educational content that can be used in the real world. [For the full proceedings, see ED623995.]