Auto-Highlighter: Identifying Salient Sentences in Text

To help analysts sift through large numbers of documents, we suggest an auto-highlighting system that computationally identifies the topmost salient sentences in each document as a form of summary and rapid comprehension aid. We conducted a user study to gather data about the types of sentences people highlight when reading and comprehending text. Our study focuses not only on the comparison between expert and non-expert users for different document types, but also the comparison between users and common algorithmic metrics for sentence selection. We analyze user-defined categories for describing the variations in the types of highlighted sentences as well as insight concerning rhetoric and language that could strengthen future algorithms.