Five Reasons why Sentiment Analysis is Sick

There’s a lot of talk about Sentiment Analysis at Customer Experience and Marketing conferences at the moment.

We’ve written on the perils of Sentiment Analysis before.  We believe it will never work.

Next time you’re lucky enough to be sitting in an airless room with weak coffee and an unending supply of Danish pastries, listening to someone talk about their sentiment analysis tools, here are five things to bear in mind.

Reason 1: Sentiment Analysis is sick.

Put the phone down. 112 can’t help you here. Does “sick” mean “poorly”? “In bad taste”? Or are we saying sentiment analysis is, like, totally radical, dudes?

Social media is like normal language evolution on speed, and unless someone’s going to recode the English language every day before breakfast, sentiment analysis is bust, right down at the individual word level.

Reason 2: Sentiment analysis would be truly great but actually isn’t any good at the wonderful things it claims it can do.

How do you analyse the sentiment of the above sentence? There are 3 positive sentiment words and only one (possibly) negative.

Sentiment analysis that does simple statistical averaging over a whole sentence won’t work either.

Reason 3: Words are rarely simply “good” or “bad”. 

Perhaps Sentiment Analysis could look at how words relate to each other within a sentence? Some tools try this.

But they might say that “simply” was moderating the full force of the “good”. Which would be right. However, is the same always true for the effect of “rarely”? Not if you’re a dealer in antiques. Especially a simply brilliant one.

Words do have emotional associations, of course. But they’re rarely consistent.  Which is not a good thing (just to be clear).

Reason 4: Clearly, we think sentiment analysis is really, really good.

If you’ve ever been chastised by someone who didn’t understand that you were being sarcastic in the email you sent them, you’ll understand that since humans don’t get it half the time, machines won’t either.

Reason 5: Why would you turn words into numbers when you already have numbers and you’d lose 98% of the useful information anyway? 

If you’re serious about using big data to really hear what your customers are saying about your brand, that statement contains some interesting information that you’d presumably want to know more about. If you’d relied on sentiment analysis to parse it, you would never have known that information.

Instead, text analytics takes sentiment for what it actually is: The most slippery component of the whole range of things that language conveys.

Ironically, surveys and other customer feedback already come with truly robust sentiment analysis (for example, in the form of NPS scores or other quantitative, structured data you asked your customer to supply), leaving text analysis to understand what the unhappy customers are unhappy about, and what you can do to fix it.

If you’re feeling hesitant about sentiment analysis, then you can read our previous blogs on the topic for an in-depth look at out misgivings. If you want to learn more about how text analytics can give you actionable insights for marketing and customer experience, email Chris.