Argument Mining with Structured SVMs and RNNs

Abstract

We propose a novel factor graph model for argument mining, designed for settings in which the argumentative relations in a document do not necessarily form a tree structure. (This is the case in over 20% of the web comments dataset we release.) Our model jointly learns elementary unit type classification and argumentative relation prediction. Moreover, our model supports SVM and RNN parametrizations, can enforce structure constraints (e.g., transitivity), and can express dependencies between adjacent relations and propositions. Our approaches outperform unstructured baselines in both web comments and argumentative essay datasets.

Argumentative Link Prediction using Residual Networks and Multi-Objective Learning

“…The application of neural network architectures in argumentation mining is relatively recent. A study most closely related to ours was presented by Niculae et al (2017) and will be described in greater detail in Section 4. The authors propose a structured learning framework based on factor graphs.…”

“…The state of the art for the CDCP corpus is the work described by the corpus authors themselves (Niculae et al, 2017). They use a structured learning framework to jointly classify all the propositions in a document and determine which ones are linked together.…”

What Changed Your Mind: The Roles of Dynamic Topics and Discourse in Argumentation Process

“…As a fast growing sub-field of computational argumentation mining [35,41], previous work in this area mostly work on the identification of convincing arguments [13,44] and viewpoints [14,19] from varying argumentation genres, such as social media discussions [37], political debates [4], and student essays [6]. In this line, many existing studies focus on crafting hand-made features [37,44], such as wordings and topic strengths [43,53], echoed words [2], semantic and syntactic rules [15,30], participants' personality [42], argument interactions and structure [29], and so forth. These methods, however, require labor-intensive feature engineering process, and hence have limited generalization abilities to new domains.…”

Argument Component Classification for Classroom Discussions

“…Like other areas of natural language processing, argument mining is experiencing an increase in the development of neural network models. Niculae et al (2017) used a factor graph model which was parametrized by a recurrent neural network. Daxenberger et al (Daxenberger et al, 2017) investigated the different conceptualizations of claims in several domains by analyzing in-domain and cross-domain performance of recurrent neural networks and convolutional neural networks, in addition to other models.…”

“…After propagating a complete argument move through the LSTM network, the resulting hidden state is the feature vector used as input to a softmax layer which outputs the predicted label. Recurrent neural networks have also been used in the context of argument mining (Daxenberger et al, 2017;Niculae et al, 2017). We set the size of the hidden state to 75 based on several factors.…”