It Takes Two To Tango: Modification of Siamese Long Short Term Memory Network with Attention Mechanism in Recognizing Argumentative Relations in Persuasive Essay

Abstract

We propose a novel approach in a dataset of argumentation relations. This task is intended to analyze the presence of a support relation between two sentences. To be able to identify relations between two sentences or arguments, one is obliged to understand the nuance brought by both sentences. Our models are modification of siamese network architectures, in which we replace the feature extractor into Long Short Term Memory and implement cosine distance as the energy function. Our models take a pair of sentences as their input and try to identify whether there is a support relation between those two sentences or not.The primary motivation of this research is to prove that a high degree of similarity between two sentences correlates to sentences supporting each other. This work will focus more on the modification of siamese network and the implementation of attention mechanism. Due to the difference in dataset setting, we cannot arbitrarily compare our results with the prior research results. Therefore, this work will not highlight the comparison between deep learning and traditional machine learning algorithm per se, but it will be more of an exploratory research. Our models are able to outperform the baseline score of accuracy with a margin of 17.33% (67.33%). By surpassing the baseline performance, we believe that our work can be a stepping stone for deep learning implementation in argumentation mining field.