Annotating Online Civic Discussion Threads for Argument Mining

Abstract

Argument mining techniques have become popular in online civic discussion thread analysis to understand an enormous amount of posts and flow of discussions for consensus building. However, the existing corpora and discussion thread analysis haven't discussed argument mining schemes sufficiently. This paper proposes a novel scheme for discussion thread analysis, annotates online civic discussions, and analyzes the annotated corpus. Our scheme consists of novel inner-and inter-post schemes. The inner-post scheme considers a post as a stand-alone discourse in a thread. We perform a micro-level annotation of argument components and relations in a post. The inter-post scheme provides a micro-level inter-post interaction to capture the argumentative reply-to relation. As a result, we have an annotated corpus including 399 threads and 5559 sentences of 204 citizens that is valid and argumentative. In addition, we analyze the annotated corpus to demonstrate statistical and linguistic properties of the corpus.