I am sure that at some time or the other you have heard or had to deal with the word twitter sentiment. Many people are confused about what it means. Some think that it is easy to measure and some just dismiss it as a useless metric. Before anything, we need to understand what it is and what are the complexities involved in it. Let us try that today
Twitter in all its glory is a huge amalgamation of thoughts and sentiment. The nature of the platform lends itself easily for mass participation and uninhibited sharing. That being said, when someone tries to analyze what twitter users are thinking now, the very size of the sample immediately becomes an analyst’s challenge. The numbers are just too huge!
Add to that the fact that we need to involve a lot of logic to determine the sentiment behind a tweet. For example, when Justin-the-one-who-shant-be-named tweets that he has a cold, a lot of people retweet it or talk about it. Of these, there are starry eyed girls who wish him well, there are jealous boys who want him dead and then there are folks like me who do not give a hoot about it, but just poke fun at him. It is very clear that a mention or a retweet is not in the same sentiment from the example above. I actually would object vehemently and even sue if I am quoted as liking JB in any form 🙂
Jokes apart, what I am trying to say here is that there is a fair bit of NLP and complex keyword/statistical analysis required to track twitter sentiment. And this analysis needs to be applied on every single one of the millions of tweets that are sampled. Only then do we get close to even 70% of a fair sentiment approximation.
This is why it saddens me when I see mere dictionary match search and count tools selling themselves off in the analytics space. They DO NOT belong there and the inference is flawed even at it’s base assumption! There are many large companies that waste funds, personnel and time on nonsense like this.
So what to do when measuring sentiment? Here is a small checklist I put together. May be useful
- Check whether you have targeted the right demographic for sampling. This is half the battle won.
- Check the ability of the tool to handle NLP. Without this, the tool is useless
- Put the responsibility in the hands of someone who knows Social Media, not a drive-by-night quack with a twitter account.
- What are the possible sentiments that you are looking for? Enumerate them so that when the report is done, you can filter the noise
- Take any analytic tool with a pinch of salt and use your logic to decide in case of conflicts
Remember these and you should be OK.