Argument identification in chinese editorials

Abstract

In this paper, we develop and evaluate several techniques for identifying argumentative paragraphs in Chinese editorials. We first use three methods of evaluation to score a paragraph's argumentative nature: A relative word frequency approach; a method which targets known argumentative words in our corpus; and a combined approach which uses elements from the previous two. Then, we determine the best score thresholds for separating argumentative and non-argumentative paragraphs. The results of our experimentation show that our relative word frequency approach provides a reliable way to identify argumentative paragraphs with a F1 score of 0.91, though challenges in accurate scoring invite improvement through context-aware means.