Time will Tell: Temporal Linking of News Stories

Abstract

Readers of news articles are typically faced with the problem of getting a good understanding of a complex story covered in an article. However, as news articles mainly focus on current or recent events, they often do not provide sufficient information about the history of an event or topic, leaving the user alone in discovering and exploring other news articles that might be related to a given article. This is a time consuming and non-trivial task, and the only help provided by some news outlets is some list of related articles or a few links within an article itself. What further complicates this task is that many of today's news stories cover a wide range of topics and events even within a single article, thus leaving the realm of traditional approaches that track a single topic or event over time. In this paper, we present a framework to link news articles based on temporal expressions that occur in the articles, following the idea "if an article refers to something in the past, then there should be an article about that something". Our approach aims to recover the chronology of one or more events and topics covered in an article, leading to an information network of articles that can be explored in a thematic and particular chronological fashion. For this, we propose a measure for the relatedness of articles that is primarily based on temporal expressions in articles but also exploits other information such as persons mentioned and keywords. We provide a comprehensive evaluation that demonstrates the functionality of our framework using a multi-source corpus of recent German news articles.