Recent developments in computer linguistics make it possible to automatically determine what a text is about and whether its sentiment is more positive or more negative. In the political barometer we are interested in what sentiment is expressed about politicians, parties (average of all politicians of that party), and news files.
To determin the subject of a text, we look for keywords that are used to communicate on specific political themes such as housing and the environment. To determine the sentiment of a text, we look at the context in which a party, politician or file is mentioned and determine the positive or negative charge of the words in that context using dictionaries.
In this way we can automatically track how often specific politicians, parties and themes are discussed in the press and on Twitter (the "buzz"), and in which way (positive or negative). We can also check how often certain political themes are associated with politicians and parties. By indicating events in political news on the timeline, we can also show its effect on sentiment and "buzz".
The political barometer can be useful for politicians, parties, journalists, political scientists and anyone with an interest in politics.
However, it is important to know the limitations of the method. The entire process is automated and is not perfect. For example, the accuracy of automatic linguistic analysis of sentiment is around 90%. This means that the trends will be correctly detected, but individual tweets and press releases can be regularly analyzed incorrectly. Also, previous research has shown that sentiment on specific media such as Twitter does not provide sufficient information to, for example, predict the outcome of elections. Finally, the barometer does not take into account "bots" that automatically post messages on Twitter.
The political barometer is part of the NWS data project of the University of Antwerp and was developed by the CLiPS research group. Do you have any questions or remarks about the Political Barometer? You can contact us via firstname.lastname@example.org.