Generating multiple summaries based on computational model of perspective

Abstract

Every story about an event offers a unique perspective about the event. A popular sporting event, such as a Major League Baseball game, is followed by several summary articles that show different points of view. The goal of this research is to build a computational model of perspective and build a system for automatically generating multiple summary articles showing different perspectives. My approach is to take a neutral summary article, reorder the content of that summary based on event features extracted from the description of the game, and produce two new summaries showing the local team perspectives. I will present an initial user survey that validated the hypothesis that content ordering has a significant effect on the users' perception of perspective. I will also discuss collecting and analyzing a parallel corpus of baseball game data and summary articles showing local team perspectives. I will then describe the reordering algorithm, the implementation of the system, and a user study to evaluate the output of the system.