Parsing argued opinion structure in Twitter content

Abstract

In this paper, we address the opinion argumentation mining issue from Twitter data with the objective of further analyzing Twitter users’ preferences and motivations. After introducing the argued opinion definition and its different elements, we propose an argued opinion mining system called TOMAS where we present an end-to-end approach to parse the structure of the argued opinion in order to identify its elements. Our suggested system consists of four consecutive sub-tasks, namely: (1) opinion-topic detection, (2) argumentative opinions identification, (3) argument components detection, and (4) argumentative relation recognition. The proposed system optimizes the argued opinion structure using different classification models. The experimental study is conducted on the MC2 Lab CLEF2017 tweets corpus while considering various comparative baselines. We highlight that our system significantly outperforms the majority baselines and significantly outperforms challenging existing approaches.