Prediction of Student Achievement Goals and Emotion Valence during Interaction with Pedagogical Agents

There 1 is evidence that Pedagogical Agents (PA) can influence students' emotions while learning with Intelligent Tutoring Systems, and that this influence is modulated by the students' achievement goals for learning. This suggests that students may benefit from personalized PAs that could rectify episodes of negative affect depending on their achievement goals. To ascertain the possibility of devising such personalized PAs, this paper investigates the real-time prediction of both students' achievement goals and affective valence while interacting with MetaTutor, an agent-based intelligent tutoring system. We train classifiers using eye-tracking data to make such prediction, and show that these classifiers can outperform a majority-class baseline at predicting both achievement goals and emotion valence.