Argument mining from Amharic argumentative texts using machine learning approach

Argument mining is an emerging science that deals with the automatic identification and extraction of arguments along with their relation from large, unstructured data that are useful for reasoning engines and computational models. Identification and extraction of argument relation from Amharic argumentative texts is a current challenge in Amharic language text processing. There have been many efforts tried on argument relation prediction for English and other European languages. This work is aimed at the design and implementation of an argument relation prediction for Amharic language using MLP, Naïve Bayes, and SVM algorithms. The study used 815 argumentative sentences collected from politically focused sources such as Amharic newspapers, weblogs, Facebook, and other social media, to evaluate argument relation prediction. The evaluation of this experiment was conducted using discourse markers, propositional semantic similarity, and a combination of these approaches, and resulted in the highest weighted average F-scores of 68%, 84%, and 88%, respectively, using Naïve Bayes, ANN and SVM. This shows that a combination approach with a SVM classifier is preferable for an Amharic argument relation prediction task. The authors tackled the problem of argument relation prediction (a subtask of argument mining) for the Amharic language, and annotated Amharic argumentative sentences (120 argument maps), making them publicly available for future research work.