Argumentative relation classification for argumentative dialogue systems

Abstract

Yet, many argumentative dialogue systems are capable of talking about a limited set of specific topics. In the herein presented work, we retrieve arguments from an argument search engine regarding any topic and organize them automatically in an argumentation structure that can serve as a database for such systems. In order to structure the arguments, we utilize methods from the field of argument mining and propose different approaches for each step. Our main contribution is the implementation of an argumentative relation classification pipeline that queries the argument search engine for arguments regarding a specific topic, performs argumentative relation classification and outputs the resulting structure. It is evaluated in an annotation study with a new combination of annotation guidelines. The carried out study yields promising results for the quality of the generated argumentation structure.