IssueBrowser: Knowledge Acquisition via Multimedia Data
Jonathan Ventura, Angus Graeme Forbes, Benjamin Adams, Swapna Joshi, Karl E. Grossner, Monica Bulger, Tobias Höllerer, B. S. Manjunath · 2008
One can hardly overestimate the pervasiveness of digitalmultimedia in the 2008 United States presidential election.Anyone seeking to learn more about the election has attheir disposal a plethora of digital media content such asonline news articles, blogs, ‘informational’ and public candi-date videos. These data not only originate from traditionalsources of media\translatedto the web but also come froman increasingly large web campaign presence of candidatesthemselves and supporters. Navigating through all thesedata can be dicult for users as the information in variousmedia formats is scattered across the web. In addition, eachof these sources of data has implicit information (such astemporal or semantic relationships) that is not readily avail-able to a casual consumer, but which nevertheless could beof interest to someone seeking a more nuanced understand-ing of the data. Furthermore, these data are connected inways that only become apparent when analyzed together.For users wishing to combine the available data into a fullerpicture of a candidate and his or her position on particu-lar issues, methods for mining the multimedia data becomesmore and more important.Towards these ends, we propose\IssueBrowser,a multime-dia data mining application that accumulates and allows forexploration of multimedia data about the 2008 U.S. presi-dential election. A novel interactive visualization installa-tion is designed to aid users’ understanding of the complex,shifting tapestry of candidates and issues involved. Theanalysis and unique combination of multimedia content pre-sented by IssueBrowser help give the user insights into thevarious presidential candidates and their positions on issuesin a manner that might not be possible when observing theconstituent media in situ.Data is in the form of text, including speeches, debate tran-scripts, news reports and blog entries, as well as relevantvideo clips, all harvested automatically from various websources. This stored data is then analyzed for issue relation-ships and presented in a navigable weighted spring graphvisualization spanning four large plasma screens in a publicsetting. We use a Nintendo Wii remote to point and clickon the screen.