A Preliminary Study of Tweet Summarization using Information Extraction
Wei Hong Xu, Ralph Grishman, Adam Meyers, Alan Ritter · 2013
Although the ideal length of summaries differs greatly from topic to topic on Twitter, previous work has only generated summaries of a pre-fixed length. In this paper, we propose an event-graph based method using information extraction techniques that is able to create summaries of variable length for different topics. In particular, we extend the Pageranklike ranking algorithm from previous work to partition event graphs and thereby detect finegrained aspects of the event to be summarized. Our preliminary results show that summaries created by our method are more concise and news-worthy than SumBasic according to human judges. We also provide a brief survey of datasets and evaluation design used in previous work to highlight the need of developing a standard evaluation for automatic tweet summarization task. 1