Context-aware Argument Mining and Its Applications in Education

Context is crucial for identifying arguments and argumentative relations in text, but existing argument studies have not addressed context dependence adequately. In this thesis, we propose context-aware argument mining that makes use of contextual features extracted from writing topics and context sentences to improve state-of-the-art argument component and argumentative relation classifications. The effectiveness as well as generality of our proposed contextual features is proven through its application in different argument mining tasks in student essays. We further evaluate the applicability of our proposed argument mining models in automated persuasive essay scoring tasks.