I’ve been involved with Sentiment Analysis and AI processing of text for years (going on 15 now). It was a huge push back in the heyday of CRM with companies like Kana creating Automated Response Tools for email service and support. They were supposed to be magic – just have your customers free-form email your company and Kana would figure out what they wanted and respond with a Knowledge Base article. It was the service/support holy grail.
Anyone heard of Kana lately?
Maybe you think I’m being an old guy… and I just don’t get it. Maybe you think the algorithms and increases in processing power have brought deriving sentiment and intent from text into the realm of the completely banal – it is easy, and anyone can do it. Maybe, but I’m a sceptic. And here is why.
How many times a week do you have to have a conversation with a family member, friend or colleague in which you attempt to explain what you really meant in that post, tweet, or email?
How often do you end up having 30 minute conversations because you were defining a key term differently?
How many times do you search on Google and get stuff you weren’t looking for?
Here is the thing, text is a HORRIBLE CONVEYOR OF SENTIMENT AND INTENT. The reality is humans have a really tough time deriving sentiment and intent from text – never-mind algorithms. Only the very best writers are able to consistently convey exactly what they meant to convey – and I’ll assure you it is because they used far more than 140 characters.
So when you see sentiment analysis being bounced around the Social Media echo chamber – before you buy in and start judging your company, product or brand based on some number spit out by a website, remember the last time you had to tell someone – “I know what I wrote, but that’s not what I MEANT” and proceed with caution.