Political Compass

Idea

The so called “political compass” is a multi-axis political model, used to label political thought on two dimensions, which are Left-Right and Authoritarian-Libertarian. While the political parties and well known influencers are already mapped on the political compass we were the first to also map newspapers on the compass according to their political attitude.
The idea was to automatically crawl as many articles as possible from a given website and to do a semantic analysis on how the different articles write about certain politicians. Because they are the representatives of their party we can then later determine on how close one newspaper is to a given party according to how positively they write about its representatives.

Results

Political Compass with German Newspapers

To find out in which way a certain newspaper writes about a given party we decided to also crawl a list of the most known German Politicians und to see in which way the different newspapers report about them. According to their political party we are then going to conclude that if they write about them, as representatives in a good way, then the newspaper is also closer to their political party. Of course we also used the political party names and tried to find out how the newspapers talked about the parties directly, but by also looking at the politicians we could gather much more data.



 An interesting new approach of systematic analysis of newspapers reaffirming academic opinions as well as showing current boundaries of natural language processing by misjudgements of various stylistic devices. 

David Nau

B.Sc. Media Science and Politics





Dataset

Each politician or party name we call “entity” and each entity we gave a “sentiscore” , between -1 and on 1, each time they were found in an article. To do the semantic analysis we used from IBM Bluemix the Alchemy API which is basically a block box that gives you a number f entities found in the text and their sentiment.

If you want, you can take a look at their demo and see how it works. You will see that in some cases it works really well, but in other cases it also misclassifies a lot. Our main emphasis was therefore on the data analysis, where we had to find out those outlies and eliminate them.

Analysis

One of our first and quick analysis when we looked at the data was, the moving mean of Angela Merkel’s sentiscore over time, regarding all newspapers, to see if it would match with our intuition of her being more criticized recently.

Angela Merkel popularity varied in the past months

Since the refugee crisis the moving average of her sentiscore kept decreasing almost constantly. In January 2016 however, it dropped rapidly. Presumably this was caused by the New Years Eve event in Cologne resulting in a critical debate on her current refugee policy. After her public television appearence on 28.February in the talkshow "Anne Will" her popularity rating started increasing again.

Further Results

Treemap

As our results we like to present our political compass with the political newspapers, which we already showed at the top of our page. As another result we programmed a tool, called tree map, which you can use to get a deeper understanding of our data. There you can see the sentiscores on an entity level, which means that you can see how each newspaper writes about every politician individually for example. To show this we decided to give three different kinds of perspectives:

  1. You can go from Newspaper to Political Party to Entity
  2. You can go from Political Party to Entity to Newspaper
  3. You can go from Political Party to Newspaper to Entity

Political Compass

For each newspaper, we know, whether it is pulled towards a given political party or rather pushed away from it. We can use this information to determine the final position of the newspaper if we consider all political parties. To display the different directions of pushing and pulling we thought of a gravitational field, where an “uphill” stands for a positive attraction and a “downhill” for a negative one. We can now add all different newspapers layer by layer, and in the end get some kind of landscape with hills and valleys. As final location for the newspaper in the political compass we use the global maximum of the landscape.

Prominence level

The percentage reflects how many indirect mentions the politician had in comparison to total mentions. A politician can be seen as more present or influencial in media when his mentions in other articles are higher compared to mentions found by directly searching articles with the politicians name in it. Being a neutral indicator, the proportion distinguishes itself from populartiy rankings which are usally generated via surveys.

Display Data

For further analysis you can take a look at our actual data with an easy to use tool we made with Shiny R. Type in the name of a politician you are interested in and select one of the newspapers. You will see a Histogram of the Sentiscore and a full list of all URLs on which the histogram is based.