Abstract
Argument Mining is a relatively new field that aims to detect and classify arguments and their relations from input text. Link prediction is a subtask of Argument Mining that attempts to determine the structure of arguments by predicting relationships between argument components. New research on augmentations to deep recurrent neural networks for Natural Language Processing, such as attention mechanism and external memory, has demonstrated potential in similar tasks such as those of Argument Mining. This project discusses the recent trends in deep recurrent neural networks for language modelling and explores whether attention and memory mechanisms can improve link prediction accuracy. The proposed model is implemented with Tensorflow and trained on a corpus of persuasive essays written by high school students. Experiments demonstrated that, although the overall model architecture is not optimal, the memory mechanism increased the link prediction accuracy in some cases.