It's always interesting to explain sentiment from a text or speech analytics standpoint to someone who has never heard of it before.
Sometimes "emotion detection" is used, but everything has emotion, right? Right!? RIGHT! There are subtle differences in how we write and how we speak that make sentiment analysis particularly tricky.
First off, I'll admit, it's not very accurate from a numbers perspective. Today the technology just isn't there to have 100% accuracy. But what does that even mean, 100% accuracy? Are humans 100% accurate?
Sentiment is a subjective exercise at best and depending on cultural and geographic differences it can be a total guess at worst.
What sentiment analysis IS is consistent. Machines don't get tired and they aren't biased by the latest football score or election result. You can set up your software and have confidence that the content you want examined is analysed in a consistent manner.
With consistency comes the ability to track changes over time. This sense of direction–if things are going up or going down–adds a whole other dimension to analysis. A single snapshot in time doesn't always tell the whole story of the effect a marketing campaign or some other event has on your sample audience.
Sentiment analysis is also good at capturing the outliers in your set, people with "elevated" emotion. Was there someone who had a particularly bad experience, or someone who just couldn't stop going on about how great their service was?
One thing that surprises people is the large amount of content that falls in the "neutral" category. For many sample sets, the majority of the crowd is just surfing along in the middle, not overly impressed but not terribly dissatisfied, so they go along like most conversations in the office lift.
"How are you?"
"Good. How are you?"
<Awkward silence. Is that my floor yet?>
What would you do if someone ran into the elevator and started jumping up and down yelling, "This is the best day of my life! I just closed a huge deal! I'm going to Disneyland!" You just smiled, didn't you? The elevated emotion from that person drew an emotional response out of you.
The real stories
While the tails of our data often represent a small percentage of the overall set, they tell the stories that are most interesting and that indicate where things are definitely going well, and where things definitely need to be improved. And they make for more entertaining presentations.
When you find a story hidden in thousands of comments that illustrates a clear success or a clear failure, that is where you gain trust and credibility for the technology. Pulling a needle out of the haystack of your clients own data is the best way to show the value of sentiment analysis. One good story from their stuff is worth far more than your polished power point deck or peer-reviewed research paper.
Most people don't care about the theory behind sentiment analysis or how smart you were to figure it all out. They want to know, "Will it help me do better in my business? Will it save time, money, and resources? Can I fix small problems before they become big?"
Bottom line, sentiment isn't a silver bullet, but it is still a very versatile and powerful tool that when used correctly can help find the real story inside your mountain of data. Happy digging!
- Craig Golightly is passionate about solving problems with the right technology and loves how text and speech analytics can help companies listen and customers be heard. Craig is a contributing author in the book "Achieve – Conversations with Top Achievers".