If you’re a newspaper, you want to know what your readers think of you. But if you ask them online and millions reply, how can you be sure you’re listening to the right thing?
If you’re an online businesses, your competitors are only a click away, so it’s vital to know how your customers feel about you. And if you’re a newspaper with a sizable online readership, then you’d think that life is made easier by an abundance of readers’ forthright comments. But save those woo-hoos, dear Editor.
We’ve noticed 3 distinct problems. Text analytics can solve some of them, but there’s still a big issue over sentiment analysis (a phrase that we hear a lot, to be honest).
1. The good news.
Text analytics will allow you to sift through a hundred or a hundred thousand comments and find statistical and reliably robust results. As we say, good text analytics software is rules-based and one of its rules is “It doesn’t go out drinking the night before it’s doing analysis”.
We do an increasing amount of text analytics work for marketing and customer insight teams of multi-nationals. In one case, where we were working with a supermarket, our text analytics work showed that one of the biggest drivers of customers ranking their store visit as unhappy was staff rudeness. A typical comment was, “The girl was rude”.
2. The kinda good, kinda fine news
At last count, Microsoft Word never wrote a good novel, Excel never told you what to do next – and similarly, even the best text analytics for marketing software can only give you the (reliable) figures. We believe you’ll still need to apply a linguistic layer across the findings. If you were in charge of Customer Experience at the supermarket, what would you do first to deal with the issue of “The girl is rude”?
When you’re a linguist, you’ll know the interesting stuff in that comment was at the front, not the back, of the sentence. Since when has it been socially acceptable to call an experienced, valuable employee “the girl”?
If you’re not sure of the social connotations of “the girl”, we can point you in the direction of some of our work in corpus linguistics. Or lend you our copy of “The Girl with the Pearl Earring”.
The way this word is being used tells us that this isn’t about rudeness at all – it’s about the customer’s perception of the abilities and responsibility of a staff member.
3. The ‘Oh shit, what now?’ news
But what happens when you’re looking at a newspaper’s readers’ comments? Think of this: Even if you can reliably measure the sentiment in a piece of text, you can’t be sure what it refers to.
The sentiment in a reader’s comment on an article may be unrelated to the article’s content or style. No one’s finishing their comments on Syria with smiley faces at the moment, however brilliantly insightful the newspaper’s commentary is.
But there’s more– people are negative online. This negativity is a well-documented phenomenon in social media. But it extends to any public discussion form. If you’ve ever attended a real town hall meeting, people rarely turn up to say how much they like the efficient design of the new tax collection system, or that overall, things are fine.
There will always be negative “background noise”.
Let’s think positively
Only when we go home in the evenings do we claim to know everything. During daylight hours, we have to make some best guesses.
Rather than rely on sentiment alone, we think newspapers and in fact all brands, can take advantage of the clues provided by metadata – such as the historical activity data for individual user accounts.
Even more immediately, there are clues about how people feel if you just look at what they’re talking about. Text analytics software can index topics of discussion in any dataset, regardless of size. You can quickly work out what’s important – and how different topics relate to each other.
Recently, a retail client gave us a dataset in which “price” was one of the most common topics. Not exactly surprising. What was surprising was that in 95% of the instances when customers discussed “price”, they mentioned “fair” too.
“Fair” tells us a lot about how the customer feels about price – more than “slightly positive” ever could. “Fair” is the grudging admission that you can’t ask for anything better, given the circumstances. So we could tell our client the price was right – but treading a very fine line.
But there’s really only one way to beat problems like this – collaboration. If you’re working in this area (possible) and you know something we don’t (probable)- we’re all ears. Let us know your thoughts and we’ll be happy to share ours.
If you’ve got any ideas (or you want to tell us this article’s rubbish), please email Chris to discuss them. Be sure to visit us at verbalidentity.com for regular updates on our work in text analytics for marketing.
Verbal Identity are brand strategy and language consultants. They use text analytics and linguistics to bring actionable customer insights to marketing. Clients include global brewers, automotive makers, a multi-national online clothing retailer, a global hotel chain and other brands which are frankly interesting.