Sentiment analysis tool identifies sarcastic comments
Sentiment analysis has long been heralded as the way to get into the heads of consumers. What do they think about a certain product or event? How are they discussing it? To find out, sentiment trackers skip merrily into the internet ether and pick up target words and phrases, reporting back with standard positive and negative findings.
Which is all very well until the Brits come along with their wry cynicism. Or the hipsters with their irony. Or the belligerent generation Xers for which sarcasm is a way of life. Or basically any member of the population that doesn’t subscribe to clear cut ideals of language; those for whom ‘great’ is often accompanied by a sigh and an eye roll. Traditional sentiment trackers can’t interpret this, so their findings are often unreliable.
However, French company Spotter now claims to have developed an analytics tool that is able to identify such sarcastic commentary, and using a combination of linguistics, semantics and heuristics, can identify true sentiment with up to an 80% accuracy rate.
Spotter’s UK sales director Richard May told the BBC: “Nothing is fool-proof – we are talking about automated systems. But five years ago you couldn’t get this level of accuracy – we were at the 50% mark.”
Sarcasm has always been an issue in sentiment analysis, May said (noting that bad service is usually the most common catalyst for it), but he adds that there are other challenges involved in the process. “We also have to be very specific to specific industries,” he said. “The word ‘virus’ is usually negative. But if you’re talking about virus in the context of the medical industry, it might not be.”
The company said that algorithms have been developed to identify tones in 29 different languages, and reports can be verified by a human should the client wish.