Using Argumentative Semantic Feature for Summarization

Abstract

The last decade has witnessed digitization of many government organization’s data. Text summarization of the political discourse, particularly the parliamentary proceedings is relatively a lesser explored area of research. In this paper, we investigate the role of semantics especially theory of argumentation in debate summarization and use it to design a semi automatic pipeline for generating these summaries. The proposed approach considers the topic-relevance, argumentative nature, sentiment and context features. We test our approach on the dataset of debates mined from Lok Sabha, the elected house of representatives in India. Our proposed methodology and pipeline show significant improvement over the high performing popular systems for ROUGE-1, ROUGE-2 and ROUGE-L metrics.