Dynamic Logic Reasoning Model for Argumentation Relation Classification

Argumentation relation classification (ARC) aims to identify the relations between argumentation component (AC) pairs. Existing studies have achieved impressive results on public datasets through the utilization of AC-level features, fine-grained attributes, and structural information. However, they overlook the logical relations and structure within AC pairs and fall short in performing dynamic logical reasoning for ARC. Therefore, in this paper, we propose a <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">D</u>yn<underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">A</u>mic <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">L</u>ogic r<underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">E</u>asoning model (DALE) with logic graph construction and dynamic logical reasoning pattern learning for ARC. Specifically, we first construct a logic graph to establish the logical structure within individual ACs and between AC pairs. Then we propose dynamic reasoning modules comprising four distinct types of graph cells to explore different logical reasoning operations. Each cell incorporates a dynamic router and interconnects others to establish a dynamic routing network. This design enables the model to autonomously adapt and select distinct operational paths based on the specific nature of the input instance. Our comprehensive experiments, conducted on three public datasets, reveal that our model outperforms strong baselines.