Summarizing microblogs during emergency events: A comparison of extractive summarization algorithms

Abstract

Microblogging sites, notably Twitter, have become important sources of real-time situational information during emergency events. Since hundreds to thousands of microblogs (tweets) are generally posted on Twitter during an emergency event, manually going through every tweet is not feasible. Hence, summarization of microblogs posted during emergency events has become an important problem in recent years. Several summarization algorithms have been proposed in the literature, both for general document summarization, as well as specifically for summarization of microblogs. However, to our knowledge, there has not been any systematic analysis on which algorithms are more suitable for summarization of microblogs posted during disasters. In this work, we evaluate and compare the performance of 8 extractive summarization algorithms in the application of summarizing microblogs posted during emergency events. Apart from comparing the performances of the algorithms, we also find significant differences among the summaries produced by different algorithms over the same input data.