In the make-believe world of movies and TV (and just because I love a quotable one-liner), sometimes ‘there’s just no room for sentiment’.
In the very real world of organisational research of course, sentiment is absolutely everything. After all, how can companies and cultures really change if leaders don’t understand how the working environments they create actually make people feel? In the past, such empathetic consideration might have been looked-down on as HR folk demonstrating yet more touchy-feely nonsense, but not only is sentiment back – not least with the emerging mental wellness debate – but business leaders are at least recognising that feeling matters, with real bottom-line impact.
Is Artificial Intelligence the Solution?
So far so good and in its wake, the quest for understanding how people feel has spawned a whole new IT industry – sentiment analysis software – you know, the stuff that uses state-of-the-art artificial intelligence. In theory it sounds fantastic. Complex algorithms are able to ‘read’ and understand exactly how people feel by reading how strings of words are put together. The very best, it is claimed, can even identify hidden irony, sarcasm and darn-right belligerence. Not surprisingly, the customer satisfaction sector was first to leap on this, as a way of processing email complaints faster, siphoning them off to be handled – sometimes automatically, sometimes by real people. But, unsurprisingly again, it’s now starting to infiltrate the HR world too.
In many ways, this was only to be expected. With ever-more regular staff pulse surveys and attempts to understand how pockets of people feel in the business more often, sentiment analysis appears to take all the leg-work out of getting to the heart of what’s on people’s minds. But, if you really stop to think about it, is this really the case?
The Flaws of Today’s Natural Language Processing Software…
Natural language analysis software claims to be able to rate sentiment in terms of polarity (i.e. how negative or positive a particular test extract is). But, in looking at what’s out there ourselves, I can categorically report that some are a lot better than others. Put simply, any sentiment analysis tool has to have its parameters defined at some point. As such, it should be no surprise that these tools are only really as good as how they are calibrated. And it’s here that problems can arise. Take a simple word – like ‘thin’. When used on a hotel comparison website [‘the walls are thin’] – this word is usually negative. When used to rate a smartphone or tablet however, being ‘thin’ is normally very good. In other words, a word with polar opposite meanings. You can see where I’m going with this. Any artificial intelligence is only as good as knowing the context with which its own definitions are set – and this is a huge, and very complicated task to get to grips with, and different organisations will need to set their systems up differently.
Why Not Just Ask People How They Feel?
Which brings me to a conclusion I often come to when thinking about sentiment analysis. If you really want to know how staff truly feel, why try and deduce it with software trying to understand hidden meanings, when you can simply ask people instead?
It sounds obvious – perhaps overly so – but those with a background in psychology and psychometrics will know there is a lot of science that backs this up. Principally, it’s that when people are asked to self-report how they feel right now (rather than what they aspire to, or want in the future), research has consistently shown this data is much more accurate.
To me, sentiment analysis needs common sense. Artificial intelligence might well be improving and starting to understand the minutiae of complex human language, but it can still only make conclusions along the lines it’s been programmed. It might sound old-fashioned, but sometimes, you can’t really beat asking someone exactly how they feel. When you create the right culture, nine times out of ten they’ll tell you what they feel when they’re asked, and they’ll tell you directly so you won’t be left wondering what they really mean.
- In Blog