Headliner: An integrated headline suggestion system

Headline generation is a short-form variant of document summarization that has been studied in natural language processing. This paper presents a case study examining the application of several different headline generation models at The Washington Post. Currently for individual news articles, multiple different headlines are manually written in order to target different platforms such as the web site and social media networks. To reduce the amount of effort from the newsroom, our strategy is to automatically propose variable-length headlines for news stories during the process of article posting. After exploring several methods, we present promising initial results. We also describe the challenges encountered in this process and future directions needed for full application of the method.