The tricky art of sentiment measurement

Sentiment is a notoriously difficult thing to automate, and unless a topic is polarising you have to allow a fairly wide margin of error. The most reliable way so far seems to analyse the words used, however that has its flaws. Imagine you had to write a programme to filter sentiment and were presented with

“Thanks to the guy who parked in my space at work today. Amazing. Really amazing.”

Tricky innit?

A Ghostbusters trailer came out yesterday, you may have seen…

…and the sentiment has been… well, tough going. Under the video, there’s 89k thumbs up to 154k thumbs down… Look under any post on Den of Geek or Digital Spy and there’s a LOT of hate for this film. Personally I’m quite looking forward to it, but that’s by the by.

At twentysix we use Sysomos quite a bit to monitor activity and sentiment on all sorts of things, so for a lark I put in Ghostbusters to see how the sentiment was performing… and these are the results.


Makes for interesting reading – perhaps the least surprising is that Twitter is by far the most negative place for… well just about everything. But more surprising is the overall sentiment being so positive. It would seem a big skew for this is the positive tone driven by news sites… The big test will of course be to see how this translates into people going to see the film at the box office.