Capturing Luxembourg Elections

“Television is no gimmick, and nobody will ever be elected to major office again without presenting themselves well on it.” – Roger Ailes, 1968.

Even though TV and print media still prevail in the Luxembourg political environment, the influence of online information is continually growing. The days when elections are either won or lost, based on the success of online campaigning may not be far away. Recently, Luxembourg has seen an exceptional time in political discourse, since the country held local elections in 2017 and national elections in 2018 (plus the European elections in 2019).

As a pilot project in targeted web crawls, the National Library of Luxembourg has archived over the course of two years: websites, social media profiles and online news media in relation with local and national election campaigns. While we were able to include Facebook and Twitter profiles of candidates and political parties, aiming at completeness in capturing all relevant content from social media was simply out of reach, due to technical hurdles and not being able to keep up with the pace of changing content.

For this reason, a focal point of the project is the online news media collection. After searching the archives and different sub-categories of each news site, we decided to check each news site daily and add individual articles to the collection. This approach not only helps with keeping up with current discussions and trends, but also allows to track changes to the media landscape: several new sites emerged over the course of the project, others changed their structure, address, or disappeared completely. Handpicking individual articles, instead of archiving the entire website, is very time consuming, but cost-effective in data consumption and also very reliable in terms of harvest quality. Moreover, each article is saved as a separate object in the collection, allowing us to go one step further in our data analysis.

We tagged each article in the following fields:

The fields “Political Party”, “Municipality” and “Topic” were tagged manually, by overlooking every article and skipping over the first few paragraphs of text. In some cases, the tags are very straightforward and can be determined only by looking at the title and image, but in some cases, a closer reading is necessary to determine the topic of the article or break down all the parties mentioned.

Same as the article selection, this manual procedure was completed by one person, with a single screening of each article and its resulting characterisation is therefore subjective.

In the context of the 2017 local elections, over 2600 news articles were harvested and tagged. News coverage of the 2018 national elections however produced over 8200 articles. While the “Topic” field was also tagged manually, we wanted to move on to automated tagging in the fields of “Political party”, “Electoral district”, “Publication date” and “Votes received”. The publication dates of articles were captured by a web scraping tool from the live web versions of the articles. By extracting the text fields from the archived WARC files, we would like to determine the names of political candidates mentioned in each article, as well as mentions of political parties and other regular expressions (such as nicknames or official titles for example). By determining the mentions of political parties and candidates, the electoral district can be deduced automatically. Since the results of the elections are in, the number of votes received for each candidate, or party mentioned in the article, can also be attributed to a field.

Filling out each field, allows for a variety of possible dissections: which news site published most articles on a certain subject, or which municipality was most often linked with a certain political party. You could find out which candidate had the highest number of mentions a certain date, or compare the number of mentions with the number of votes received. We could create a publication timeline, or geographic heat map for any combination of fields. To create and visualise these queries, we use Kibana, a data search engine and analysis tool.

From a customisable dashboard, we can explore different constellations of news articles we are interested in. For example, we can find out which party’s candidates had the most media coverage in Luxembourg City and know the URLs of the respective articles are.

Since this method of processing news articles is very time-consuming, this form of examination might go beyond the goal of collecting and preserving websites and will most likely not be a regular exercise for the Luxembourg Web Archive. The main objective of this project was to present the possibilities and benefits of working with archived web materials, with a relatable subject, in an easily accessible way. Moreover, we are able to share insights about our activities, without accessing the actual web archive.

Looking back at two years of harvesting and tagging, we also have some exciting news from the National Library of Luxembourg: We are currently moving to our newly constructed building, along with 1,8 million documents, previously stored in 7 different locations. A massive project with lots of challenges along the way towards our new reading room with up to 200.000 books. If you are interested in the relocation process , we are sharing our progress on the newly installed BnL Twitter page: twitter.com/bnluxembourg.

In addition, we are putting the finishing touches to the website for the Luxembourg Web Archive, which will serve as a platform to:

• Raise awareness about the necessity of web archiving.
• Inform website owners about the technical process behind web harvests.
• Present our collections and selection policy.
• Enable site suggestions to the web archive, get feedback on special collections and foster community participation.

We are looking forward to publishing webarchive.lu in the summer of 2019.