May 31, 2016
The social sentiment monitoring of the upcoming Brexit vote is our very first attempt to test the SSIX technology live. Originally, we had not envisioned that politics would be its very first real world test. However, Brexit is a unique event not to be missed, due to its global economic impact, as well as political and social implications. For the SSIX consortium, this is an ideal event to test SSIX behaviour and the limits of our first platform prototype.
These days, news and social media channels are getting ‘red hot’ with comments pro and against Brexit. From regular British citizens, political parties and various other EU citizens to Bank of England, FMI head Christine Lagarde and many financial institutions worldwide – everybody has an opinion. The debate has heated up so much that even Napoleon Bonaparte and Adolf Hitler were brought into the debate by the former mayor of London – Boris Johnson, in order to support his stance.
Regularly updated polls show the vote is going to be a close call:
According to Financial Times Brexit Poll Tracker (Figure 1) 46% would vote to stay in the EU, 41% would vote to leave the EU, while 12% are undecided as of 25th of May.
Figure 1 Financial Times Brexit Poll Tracker (source: Financial Times, May 2016)
Bloomberg Brexit Watch (Figure 2) comes to a similar result. 47.5% would vote to remain in the EU, 42.4% would vote to leave and 10% are undecided as of 24th of May.
Figure 1 Bloomberg Brexit Watch (source: Bloomberg, May 2016)
With such a narrow difference, anything can happen. And therefore, here is our challenge: can we infer from the past sentiment patterns where the tide is going to push? Are there unseen early signals within social media that, combined, would define the most probable outcome of the vote, before it really happens?
For the SSIX consortium, this is a high value exercise as once the process is consolidated with lessons learned from BREXIT, we can extrapolate it to other domains of high interest for us such as Finance, Health and eGovernment.
Without entering in specific technical details, the SSIX team have started for more than a month already to assemble all pieces of the technical puzzle required for the BREXIT sentiment monitoring. One of the puzzle pieces was to build a dedicated gold standard for Brexit. While doing it, several interesting aspects appeared when analyzing Twitter data feeds:
- Political tweets are more challenging to interpret due to the amount of irony, sarcasm, and double meaning;
- Machine based interpretation of tweets with very specific local political particularities is much more difficult;
- Many tweets try to persuade other readers by solely referencing 3rd parties opinions and links;
- Most twitter users feel the need to validate their point by providing 3rd parties links. While such links in many cases don’t exhibit a full line of argumentation (with the exception of newspaper articles) they are intended to be considered valid arguments of their point;
- Some tweets are invoking sentiments such as fear and greed without a reasoning behind;
- Tagging tweets is widely used and in some cases they are barely relevant to the tag context (#Brexit);
- There can be an overuse of “@” and “#” that sometime makes it hard for a human user to get the clear message of the tweet;
- Many tweets are context dependant: a clear sentiment cannot be inferred unless the context they point to is explored and understood;
Throughout our raw Twitter data analysis, any of the above situations or combinations thereof were frequently met This shows the complexity of such an analysis and the fact that a simple ‘stay’/’leave’ type analysis is not sufficient: there are nuances at play that show the strength or weaknesses of a sentiment. Consequently, SSIX will target measuring the strength of such sentiments with the view to infer meaningful patterns in advance.
One can imagine in various ways how such a technology could be used in the future for electoral events: polling techniques may become redundant as online social media will provide enough data to infer in real time what citizens’ latest preferences are.
And then…why not measure politicians’ effectiveness and popularity in real time as they execute their duties?
More updates to come as Brexit gets closer!
This blog post was written by SSIX partner Laurentiu Vasiliu at PERACTON.
For more information on SSIX, visit our website ssix-project.eu.