Performance tasks in virtual environments result in rich data about the test takers’ behavior. The more realistic the task, the more difficult it is to identify meaningful signals in these data from the noise and the artifacts. Several approaches that have been considered in educational assessments have been inspired by the work conducted in the data mining and NLP communities. In this presentation several approaches are described: a data model for creating a structure for the log files; a visualization method using networks, and a scoring method using NLP and clustering methods. The methods will be illustrated with data from a collaborative task from a collaborative game from ACTNext, ACT.

See more: IMPS 2017 AVD PC ACTNext ACT 07202017 and IMPS_2017_Talks_w_Cover_C_Reduced_0