Supporting cross-media analyses by automatically linking multiple collections

Introduction
Analysing media coverage across several types of media-outlets is a challenging task for Humanities researchers. Up until now, the focus has been on newspaper articles: being generally available in digital, computer-readable format, these can be studied relatively easily. Analyses of visual material like photos or television programs are however rarely undertaken. This poster presents the results of the PoliMedia project that aimed to showcase the potential of cross-media analysis by linking the digitised transcriptions of the debates at the Dutch Parliament with three media-outlets: 1) newspapers in its original format and lay-out of the historical newspaper archive at the National Library, 2) radio bulletins of the Dutch National Press Agency (ANP) and 3) newscasts and current affairs programmes from the Netherlands Institute for Sound and Vision (Kleppe et al., 2014; Kemman & Kleppe, 2013). The PoliMedia search user interface allows researchers to search through the debates and analyse the related media coverage via www.polimedia.nl. The main research question that can be addressed using PoliMedia is: What choices do different media make in the coverage of people and topics while reporting on debates in the Dutch parliament since the first televised evening news in 1956 until 1995? An advantage of PoliMedia is that the coverage in the media is incorporated in its original form, enabling analyses of both the mark-up of news articles as well as the photos in newspapers. PoliMedia demonstrates the application of Linked Open Data in the Digital Humanities: not only was a search interface developed for scholars, the data was published online and made publicly available via a SPARQL endpoint at data.polimedia.nl. This enables researchers to build customised tools that can support their specific research.

Fig. 1: Screenshot of the PoliMedia search results page

Method
The basis of PoliMedia lies in the minutes of the Dutch parliament from 1814-1995, containing circa 2.5 million pages of debates with speeches that have been OCR’d and thus allow full-text search. The minutes have been converted to structured data in XML form in previous research (Gielissen & Marx, 2009). For each speech (i.e. a fragment from a single speaker in a debate), we extract information to represent this speech; the speaker, the date, important terms (i.e. named entities) from its content and important terms from the description of the debate in which the speech is held. This information is then combined to create a query with which we search the archives of the newspapers, radio bulletins and television programmes. Media items that correspond to this query are retrieved, after which a link is created between the speech and the media item, using semantic web technologies (Juric, Hollink, & Houben, 2013). In order to navigate these links, a search user interface was developed, based on a requirements study with five scholars in history and political communication. During development, an initial version of this interface was evaluated in an eye tracking study with 24 scholars (Kemman, Kleppe, & Maarseveen, 2013).

Fig. 2: Screenshot of the PoliMedia debate page

Results
From an evaluation of a set of links to newspaper articles, it was found that the recall of the algorithm is approximately 62%, with a precision of 80% (Juric et al., 2013; Kleppe et al., 2014). However, no links to television programmes could be made. At this point we can make no conclusions about whether this was due to the size of the television dataset, the lack of full-text search or due to lack of suitability of the linking algorithm. Linking to television programs thus remains a question for future research. The combination of a search interface and a SPARQL endpoint resulted in PoliMedia becoming the finalist of the Semantic Web Challenge 2013 en winning the first prize in the LinkedUp Veni Competition.1 The poster presentation will be accompanied with a live demo of the system via www.polimedia.nl.

References
Gielissen, T., & Marx, M. (2009). Exemelification of parliamentary debates. In Proceedings of the 9th Dutch-Belgian Workshop on Information Retrieval (DIR 2009) (pp. 19–25).

Juric, D., Hollink, L., & Houben, G. (2013). Discovering links between political debates and media. In The 13th International Conference on Web Engineering (ICWE’13). Aalborg, Denmark. doi:10.1007/978-3-642-39200-9_30

http://challenge.semanticweb.org/2013/finalists.html & http://blog.okfn.org/2013/09/17/linkedup-open-education-veni- competition-the-winners/

Kemman, M., & Kleppe, M. (2013). PoliMedia - Improving Analyses of Radio, TV & Newspaper Coverage of Political Debates. In Research and Advanced Technology for Digital Libraries (pp. 401–404). Springer-Verlag Berlin Heidelberg. doi:10.1007/978-3-642-40501-3_46

Kemman, M., Kleppe, M., & Maarseveen, J. (2013). Eye Tracking the Use of a Collapsible Facets Panel in a Search Interface. In Research and Advanced Technology for Digital Libraries (pp. 405–408). Springer-Verlag Berlin Heidelberg. doi:10.1007/978-3-642-40501-3_47

Kleppe, M., Hollink, L., Kemman, M., Juric, D., Beunders, H., Blom, Oomen, J., Houben, G.-J. (2014). PoliMedia - Analysing Media Coverage of Political Debates By Automatically Generated Links to Radio & Newspaper Items. In M. D’Aquin, S. Dietze, H. Drachsler, M. Guy, & E. Herder (Eds.), Proceedings of the LinkedUp Veni Competition on Linked and Open Data for Education. Geneva, Switzerland: CEUR-WS.