Classifying argumentative stances of opposition using Tree Kernels

The approach proposed in this study aims to classify argumentative oppositions. A major assumption of this work is that discriminating among different argumentative stances of support and opposition can facilitate the detection of Argument Schemes. While using Tree Kernels for classification problems can be useful in many Argument Mining sub-tasks, this work focuses on the classification of opposition stances. We show that Tree Kernels can be successfully used (alone or in combination with traditional textual vectorizations) to discriminate between different stances of opposition without requiring highly engineered features. Moreover, this study compare the results of Tree Kernels classifiers with the results of classifiers which use traditional features such as TFIDF and n-grams. This comparison shows that Tree Kernel classifiers can outperform TFIDF and n-grams classifiers.