Classification of Natural Language Descriptions for Bayesian Knowledge Tracing in Minecraft.

Application of Bayesian Knowledge Tracing (BKT) has primarily occurred in formal learning settings. This paper presents an integration of BKT in an informal learning context to assess the structure and skill level of learner scientific observations. We compare different approaches to text classification in a Minecraft science simulation. Our models were trained on data collected from two separate middle schools with students of different backgrounds. Experimental results demonstrate the effectiveness of several machine learning models to automatically label observations.