Sentiment analysis experiment ends
I’ve ditched sentiment analysis as a metric on the broadbandrating.com website. The idea that we might be able to rate broadband providers according to social media sentiment was a good one but in reality most people commenting on their broadband provider were slagging them off (to use the venacular). Most positive vibes were being generated by positive marketing initiatives by the ISPs themselves, and then mostly by TalkTalk who seemed to be extensive users of Twitter. Nothing generally to do with customers unilaterally praising their ISP.
It is interesting to see sentiment analysis creeping in to more and more places in our lives (ish). During the general election it was being used to gauge how parties were doing. In fact for it to be anything other than random guesswork we found that the results needed to be 100% human generated.
In other words we couldn’t leave it to a computer to decide whether a tweet was praising or condemning an ISP. The English language has too much scope for misunderstanding. In theory this manual activity could be outsourced to somewhere with low cost labour but then that ran the risk came with a very high probability of it being done by someone whose first language was not English and who would therefore be facing some of the problems of the computer in identifying the tone of a tweet.
We bore the cost of getting this work done in the UK for some months but in the end decided the feature wasn’t worth it. It was an interesting experiment.
In the meantime we still have the Customer Service metric – how long it takes an ISP to answer the phone and where their call centres are typically based (ie UK or India). We will also no doubt introduce others but in the meantime are concentrating on pure marketing activities such as the pig racing and the bulletproof broadband videos (both also featured on this site if you’re interested).
So there ya go. The sentiment analysis experiment was an interesting toy but in the end didn’t cut it. You have to move on.