Searching for people’s opinions via surveys and polls has been an expensive and time-consuming task. In partnership with the University of Antwerp, we performed a series of sentiment visualization experiments on live Political Tweets -produced during the run-up to the Dutch local-election in April 2015 to determine whether a tweet is positive, negative or neutral. The aim of the project was to process Twitter data both automatically with sentiment analysis algorithms, and also manually using annotation and qualitative judgements. There are several key aspects of this project:
- The infrastructure aspect focused (JStart, Streams, and Cloud Computing), on building and testing a platform for processing high volumes of data from the twitter firehose.
- The natural language processing aspect designed sentiment analysis techniques to deal with the prevalence of sarcasm on Twitter in the context of politics.
The tweets data we used were collected using the Twitter Streaming API for 8 weeks leading to the local-election. Search criteria specified include the mention of political parties such as The Labour Party (PVda), Conservative Party (VVD),…etc.; the mention of candidates and; the use of the hash tags such as #election2015, #Labour etc.; and the use of certain words such as “election”.
We also provide a content website to present the result of content visualization by introducing an interactive slider theme to represent the “Hot issues”: Healthcare, Sustainable energy, crime, etc.