WikiTopics: What is Popular on Wikipedia and Why

Proceedings of the Workshop on Automatic Summarization for Different Genres, Media, and Languages, pages 33–40, Portland, Oregon, June 23, 2011. c⃝2011 Association for Computational Linguistics WikiTopics: What is Popular on Wikipedia and Why Byung Gyu Ahn1 and Benjamin Van Durme1,2 and Chris Callison-Burch1 1Center for Language and Speech Processing 2Human Language Technology Center of Excellence Johns Hopkins University Abstract We establish a novel task in the spirit of news sum- marization and topic detection and tracking (TDT): daily determination of the topics newly popular with Wikipedia readers. Central to this effort is a new public dataset consisting of the hourly page view statistics of all Wikipedia articles over the last three years. We give baseline results for the tasks of: discovering individual pages of interest, clustering these pages into coherent topics, and extracting the most relevant summarizing sentence for the reader. When compared to human judgements, our system shows the viability of this task, and opens the door to a range of exciting future work. 1 Introduction In this paper we analyze a novel dataset: we have collected the hourly page view statistics1 for every Wikipedia page in every language for the last three years. We show how these page view statistics, along with other features like article text and inter-page hyperlinks, can be used to identify and explain popular trends, including popular films and music, sports championships, elections, natural disasters, etc. Our approach is to select a set of articles whose daily pageviews for the last fifteen days dramatically increase above those of the preceding fifteen day period. Rather than simply selecting the most popular articles for a given day, this selects articles whose popularity is rapidly in- creasing. These popularity spikes tend to be due to sig- nificant current events in the real world. We examine 100 such articles for each of 5 randomly selected days in 2009 and attempt to group the articles into clusters such that the clusters coherently correspond to current events and extract a summarizing sentence that best explains the rel- evant event. Quantitative and qualitative analyses are pro- vided along with the evaluation dataset. 1The data does not contain any identifying information about who viewed the pages. See http://dammit.lt/wikistats Barack Obama Joe Biden White House Inauguration ... US Airways Flight 1549 Chesley Sullenberger Hudson River ... Super Bowl Arizona Cardinals Figure 1: Automatically selected articles for Jan 27, 2009. We compare our automatically collected articles to those in the daily current events portal of Wikipedia where Wikipedia editors manually chronicle current events, which comprise armed conflicts, international re- lations, law and crime, natural disasters, social, political, sports events, etc. Each event is summarized with a sim- ple phrase or sentence that links to related articles. We view our work as an automatic mechanism that could po- tentially supplant this hand-curated method of selecting current events by editors. Figure 1 shows examples of automatically selected ar- ticles for January 27, 2009. We would group the arti- cles into 3 clusters, {Barack Obama, Joe Biden, …

Content truncated to 500 words