Comparative analysis of supervised learning approaches for argument identification

In this paper, we have discussed the argument identification problem and the various solutions that have been presented over the years. We have tested various supervised learning algorithms for the argument component identification task using the features proposed by Stab and Gurevych on an annotated corpus of persuasive essays. SVM gave the best results. Argumentation is quite domain dependent. To fully exploit supervised learning for this task there is a need for more annotated corpora. Generalized approaches that can cater to argumentation of various domains have yet to be developed for the argument identification problem.