Text analytics isn’t straightforward – but we’ve got it sussed.
Consider the phrase “barely audible”. For many of our automotive clients, this phrase is categorised as a negative:
“I plugged my phone into the bluetooth system and it was barely audible”. Ie, the Bluetooth system wasn’t enabling the customer to hear their phone = a negative comment.
However, for another client, whose business is electronic white goods, this phrase needs to be categorised as a positive:
“I switched on my new dishwasher and it was barely audible”. Ie, the dishwasher ran very quietly = a positive comment.
Another example we often come across is “Best Buy”. Additional context around this phrase is needed to make sense of it. For example, Feedback Ferret contextual coding would distinguish between these three examples by use of existing Lexicon contextual phrases that would assign these to different Categories:-
“I am going to Best Buy for a new laptop” = Categorized as ‘Best Buy (Retailer)’
“This car has been my best buy ever” = Categorized as ‘Purchase Experience – Good’
“This was hardly a best buy in anyone’s language” = Categorized as ‘Purchase Experience – Poor’
So how does the Feedback Ferret text analysis machine, that has been taught automated rules, understand whether phrases such as “barely audible” and “best buy” are positive or negative?
It’s all thanks to the “machine + human” approach. Through rigorous quality control and ongoing updates to the coding of phrases, humans intervene in the automated process to develop and evolve the machine to correctly code ambiguous phrases.
Our text analytics software is built around a Lexicon of millions of human words and phrases, coded into our system by hand over the last 20 years. We estimate that more than 100 man years of Lexicon coding have been invested to insure that we capture all the ways humans have of describing their experience in a written format. The in-house Lexicon team continually reviews and enhances the Lexicon content every day. This includes bad grammar, colloquialisms, incomplete sentences, misspellings, typos, slang terms, sarcasm, irony and, of course, ambiguous phrases.
With accuracy rates over 90%, our clients are often astounded by how the Ferret understands complex phrases that appear in their customer feedback.
By Piers Alington, Managing Director