Exploring features and classifiers for dialogue act segmentation

Abstract

This paper takes a classical machine learning approach to the task of Dialogue Act segmentation. A thorough empirical evaluation of features, both used in other studies as well as new ones, is performed. An explorative study to the effectiveness of different classification methods is done by looking at 29 different classifiers implemented in WEKA. The output of the developed classifier is examined closely and points of possible improvement are given. © 2008 Springer-Verlag Berlin Heidelberg.