Unsupervised behavioral modeling of an E-learning domain based on timed automata

Abstract

In this work an original unsupervised methodology for user behavior modeling and recognition based on timed automata is introduced. The key idea is that user actions are represented by timed strings which can be parsed and recognized by a timed automaton. The timed automata induced by the timed observation of user logs embeds the different behaviors by halting into different automaton final states. Experiments has been held on a dataset extracted by the authors from the student logs of an e-learning domain based on the platform moodle. The automaton induced by the session log sequences has been implemented by the UPAAL automaton construction library. Experimental results shows the effectiveness of the proposed user behavior model based on timed automaton, for classifying the user behavior recognition, Since the behaviors are initially un-labelled, the validation of the automata is obtained by evaluating the similarity of the distribution of different behaviors of the same users in the training/tests sets. © Research India Publications.