Sebastian Faulks in the Guardian helps pull together a global visualisation (well, Global to the extent that the Globe is represented by Twitter) of sentiment.  Call me an old cynic, but I suspect someone in his publisher’s publicity team was probably more responsible than the grand old man himself.  Nevertheless, it stirs some interesting dirt from the bottom of our social media puddle.

The analysis simply pulls references to simple language – words – and then categorises them into key emotional categories – love, happiness, fear, longing and hope.  He suggests that people hashtag their tweets accordingly, and preferably add a picture so that the visualisation can be more colourful.  What this fails to capture, of course, is genuine sentiment as distinct from simply words.  Furthermore, it is a representation of objective expression, as opposed to context sensitive understanding.  Let me explain.

I may tweet “I hate Mondays”, which may genuinely be an expression of feeling hatred.  Or it may be a figure of speech, expressing more frustration, tiredness, or possibly no more than banter – consider the addition of an exclamation mark, in response to a colleague who tweets that they need to remember to submit the 45F form because today is Monday.  Further extending the example, I may tweet “I hate The Happy Mondays”, which is a statement of taste in music, but of course a rudimentary analysis might pick up the word happy in there as well, and the robot’s brain would begin to freeze.

An interesting little publicity gimmick for Faulks’ new book, and one that begins to hint at what might be possible.  IBM (with whom your correspondant is associated, but with whom this blog is most certainly not!) has been doing a huge amount of work in sentiment analytics, and in natural language processing.  Combining the two, Twitter and social media could genuinely tell us a lot.  For example, we could ask whether people in a particular geographic area (a country, for example) generally approved of a particular piece of legislation, a particular candidate, a particular party.  It could be measured over time, tweaked in real-time, and messaging could become dynamic.  Ultimately, predictive sentiment analytics could dictate policy.  Now that’s pretty scary.  Could we build a framework within which we could measure State Legitimacy?  Possibly…