Citation-Based Scientific Paper Summarization Using Game Theory

Given the vast amount of research-oriented literature, any new research work asks for a comprehensive literature review. This process is typically very time-consuming. Citation information of a paper is a reflection of its contributions to the researcher's group. We use the citations of the paper to generate a comprehensible and informative summary, which will reduce the time taken to review the paper. In this paper, we propose a novel evolutionary game-theoretic approach to summarize scientific papers. A set of two-player games is played to choose the most relevant sentences in the paper that should be a part of the summary. Our method doesn't require an annotated dataset with gold summaries, thus levitating the need for a large, manually created dataset, which would be a primary requirement for any supervised approach. We have evaluated our results against human-expert generated summaries and state-of-the-art methods.