Abstract. We present machine-learned models that detect two forms of middle school students ’ systematic data collection behavior, designing controlled experiments and testing the stated hypothesis, within a virtual phase change inquiry learning environment. To generate these models, we manually coded a

Abstract. This research investigates the detection of student meta-cognitive planning processes in real-time using log tracing techniques. We use fine and coarse-grained data distillation, in combination with coarse-grained text replay coding, in order to develop detectors for students ’ planning of experiments in

As novel forms of educational software continue to be created, it is often difficult to understand a priori which ensemble of interaction behaviours is conducive to learning. In this paper, we describe a user modeling framework that relies on interaction logs to identify different types of learners, as well as their characteristic interaction behaviours and how these behaviours relate to learning. This information is then used to classify new learners, with the long term goal of providing adaptive interaction support when behaviours detrimental to learning are detected. In previous research, we described a proof-of-concept version of this user modeling approach, based on unsupervised clustering and class association rules. In this paper, we describe and evaluate an improved version, implemented in a comprehensive user-modeling framework that streamlines the application of the various phases of the modeling process.