Can Twitter predict ‘the mood of the nation’?
A team of academics at Loughborough University in the UK have developed a system for analysing Tweets for emotion. The Emotive project detects emotion that people express on Twitter across the UK in order to identify and analyse the national psyche as expressed on the social network. There are many challenges with this kind of mass analysis of emotion, but the Loughborough project is an interesting step towards drawing meaningful analysis from the Tweets that people leave behind them online every day.
There are a number of challenges to analysing Tweets for emotion that the Emotive project seeks to overcome:
- Rather than a simple analysis of sentiment, the project looks to identify emotion expressed in eight areas: Anger, Disgust, Fear, Happiness, Sadness, Surprise, Shame and Confusion
- Rather than a simple binary distinction (such as negative or positive), each emotion is rated on a scale that captures differing intensities
- Rather than allocating sentiment to an entire Tweet, each string of text is segmented to identify the different elements discussed in it
Together, these three elements produce a more nuanced set of analyses than most sentiment and social media analytics tools and the Emotive project reports good success at analysing reaction to relatively extreme events – the murder of a British solider in London, for example.
The challenge for such emotion analysis is to really understand when it can be useful, and when it cannot. When dealing with any large linguistic corpus it can be difficult to spot extremes of emotion – for as many people expressing sadness, there are people expressing happiness. So the danger is that any analysis shows ‘the mood of the nation’ as always being in the middle – neither one thing nor the other.
In fact, a tool like Emotive is of most use (potentially only of use) when it is identifying extreme reaction to events. This, in itself, can be a very useful feature – understanding the emotion expressed during major events in the nation, for example, or even spotting when major events are emerging. Taken further, it would be interesting to see how a tool like Emotive can be used to analyse differing emotions not just by geography, but also by groups or communities of individuals. Perhaps we know that a certain group is likely to react to events more quickly, if we can isolate their accounts on Twitter, we can then start to spot when this group’s collective emotion changes more markedly than another group.
Overall, the work from Loughborough is a great step forward for those of us looking to better understand the value of analysing social media conversations. It starts to overcome some of the barriers of this analysis, and can be built on further to provide more nuanced understand.