Error Pattern-based similarity analysis of task performance in a virtual training environment: a meta graph clustering approach

Resumo

Intelligent Tutor Systems, Serious Games, and Simulations are user interaction-based instructional
technologies capable of adapting to the needs of learners. Tracking and logging data from user interactions during
the execution of a task allow for learning assessment and visualization of learner performance, which, in turn,
allows for the identification of learner mistakes. Instructional tasks can be mapped as a graph, and paths represent
the ordered sequences of task activities. Mining path patterns seek similar strategies and anomalous behaviors of
learners in performing the task. In this paper, graph similarity methods are applied to clustering tasks performed
in a virtual training system. Feasible paths on the graph represent the expected sequences for task execution, and
errors are deviations from them. Sequences of activities performed by the learners correspond to free walks on the
graph. Through task rules and reliability analysis, errors in learner walks were extracted and represented by vectors
and then clustered in error patterns. A meta clustering analysis of error pattern-based clusterings and similarities
clusterings reveals which of the former are closest to the error patterns. Based on the findings achieved, in future
work, a new similarity method that is sensitive to error patterns will be proposed.