Start Date
11-8-2016
Description
The rapid growth of online communication has dramatically changed the manner in which arguments take place. Participants in virtual teams and communities continue to face challenges to detect and make sense of arguments when the constituent elements of an argument are scattered in prolonged online discussions. Few methods or tools are available for the detection of arguments from available sources. This paper develops a theory-based argument detection model. Drawing on the argumentation theory, we propose a model for argument detection. It is composed of features that reflect five categories of argumentation functions, including announcement, reasoning, modality, transition, and affect, and another language features that are informative for recognizing argument. The evaluation results show that the model achieves higher accuracy and recall in detecting arguments in message sets, compared to baseline models. The paper also presents an illustrative example to show application of the model in practice.
Recommended Citation
Zhang, Guangxuan; Zhou, Yilu; Purao, Sandeep; and Xu, Heng, "Argument Detection in Online Discussion: A Theory Based Approach" (2016). AMCIS 2016 Proceedings. 18.
https://aisel.aisnet.org/amcis2016/Decision/Presentations/18
Argument Detection in Online Discussion: A Theory Based Approach
The rapid growth of online communication has dramatically changed the manner in which arguments take place. Participants in virtual teams and communities continue to face challenges to detect and make sense of arguments when the constituent elements of an argument are scattered in prolonged online discussions. Few methods or tools are available for the detection of arguments from available sources. This paper develops a theory-based argument detection model. Drawing on the argumentation theory, we propose a model for argument detection. It is composed of features that reflect five categories of argumentation functions, including announcement, reasoning, modality, transition, and affect, and another language features that are informative for recognizing argument. The evaluation results show that the model achieves higher accuracy and recall in detecting arguments in message sets, compared to baseline models. The paper also presents an illustrative example to show application of the model in practice.