Detecting conflicts in collaborative learning through the valence change of atomic interactions

Abstract

Naturally, every collaboration will bring conflicts that can affect the performance of a team. The earlier a conflict is detected and managed in a collaborative group, the better. Detecting and tracking conflicts in Computer-Supported Collaborative Learning (CSCL) is laborious work. If the teacher does it, the intervention may be out of time. Although written dialogues in groups having a conflict reveal the increment of negative emotions in comparison to non-conflict dialogues, a classifier that only uses statistics of the valence of consecutive messages in a window of the talk shows poor performance. This paper proposes to use features based on the valence change between a message and its response. In this way the algorithm focuses in the kind of interaction. We study different implementations of the bootstrap aggregating technique to detect conflicts. Results obtained show the viability of the proposed approach.