top of page
  • Writer's pictureAlex

Can a machine understand emotions?

Updated: Mar 28, 2019

As far back as 2011, a US study showed that consumers received five times as much information every day compared to 1986. From family WhatsApp groups to keeping up with twitter memes, it’s easy to be overwhelmed. Anger over scandals on hosting content that is disturbing, fake or extremist has further fuelled a backlash as advertisers pull content on from news sites and platforms.

The emotional consequence of using social media have been well documented, but for all of the messages encouraging people to delete Facebook or manage their screen time, the numbers of users and interactions are still rising.

Most traditional print or location based advertising provide a poor return, with a few notable demographic exceptions – generally men are more receptive to location based advertising. One example: I ran a small campaign for a school for just £2,000. Traditionally, a school might pay twice that for a half page ad in the local paper to reach 11,000 people, mainly over the age of 50. By contrast, the digital ads were A/B tested, targeted at local parents, reached 20,000 and drove 1,000 new site visitors. School applications are up, which given each pupil can be worth £30,000 in funding, pays for itself. The most effective adverts were those that talked not about the success of the school (measured by Ofsted / results) but those that discussed the potential impact on the child (i.e. we will help your child get to university).

Emotions matter

This comes back to your core brand proposition. There are many questions people might have about your product - who you work with, what you do, where you operate. The most effective messages start with why. Simon Sinek describes this approach as the ‘golden circle’ The brain’s emotional core - the limbic system - drives instinctive behaviour and is responsible for promoting trust and loyalty. Successful marketing messages start with "why" so you attract customers who share your beliefs. This ‘Remarkable Lives’ poem from schools group AET is a nice example.

It helps to understanding your audience’s present state of mind. Research from Yahoo found that when consumers are feeling upbeat, they are 30% more likely to engage with native video content than in other emotional states. How on earth do you begin to predict when people will be in an upbeat state of mind? After all, you can’t share all your news on a Friday afternoon.

An important side-note here: being trusted is not the same as being liked, as Steve van Riel explains, showing up some of the limitations in existing surveys such as the Edelman Trust barometer, which tend to have very similar results as surveys on how much you like a company. For instance, you might not like what a company does (drilling for oil, making plastic), but you may trust them to do their job competently. That can still affect the bottom line: one experiment suggests consumers are willing to pay an 8% premium for the same product from a company they trust. 

How to define success? The industry has spent huge amounts of time considering readership, CPC, AVE and so on, but sentiment analysis still remains in its infancy as a way to measure the impact of a crisis or longer term reputation analysis.  Most algorithms, like ones used by media monitoring companies, simply assign a negative or positive emotion to key terms, which falls apart in a meme-hungry, sarcastic medium like Twitter.

Can a machine understand sarcasm?

The late Marvin Minsky, who is acknowledged as the father of modern artificial intelligence, wrote in his book ‘The Emotion Machine’ that emotions, rather than being a ‘different’ way of thinking are an integral part of rational thought, and that AI would only really succeed in replicating human tasks when it could decode the complexities of human emotions.

According to emerging research from MIT, deep learning can provide some answers: “instead of explicitly telling the machine how to recognise emotions, we ask the machine to learn from many examples of actual text. In particular, deep learning models can learn very subtle representations, and figure out for themselves what makes something happy or sad, serious or sarcastic.”

Their 'DeepMoji' model uses a dataset of 55 billion tweets to learn about emotional concepts in text like sarcasm and irony. The answer was brilliantly simple: “many tweets already use something like a labelling system for emotional content: emoji. Having taken advantage of this to help the system read tweets for emotion in general, the researchers were then able to teach it to recognise sarcasm.”

Of course, sarcasm is hugely context specific, but that means humans can misunderstand it too. But just like the chess computer, the machine beat the humans: in a trial, DeepMoji was 82 percent accurate at identifying sarcasm correctly, compared with an average score of 76 percent for the human volunteers.

This is a potentially huge development: not just in how to measure corporate reputation, but to tackle online bullying or hate speech. You can try the tool itself here.

Be different

Buzz Carter from Bulldog digital adds that running a campaign can be “a daunting task, but one key bit of advice to follow is to do something different to everyone else in the space, whether that’s creating a unique piece of content people can’t resist sharing or taking a different approach to your digital mind-set all together…so when the industry zigs, if you zag, you’ll stand out.” This documentary video by Perspective Pictures filmed along Bolivia’s ‘death road’ for insurance provider Compare the Market, upended their normal style of content to create a sense of drama and fear.

There are clear generational differences. Older consumers and newer internet users are less likely to recognise content as an advert, while younger audiences are increasingly switching off when they feel they're seeing an ad. Brands marketing to younger audiences need to produce meaningful, narrative-driven, content. Younger audiences are more likely to trust recommendations from influencers on YouTube and Instagram – almost as much as from their own friends, which to be honest slightly scares me.

None of this means that books, or newspapers or printed materials are ‘dead’, just that they will play a different role. Polly Allen, a marketer and journalist who works for Voscur, says that getting emotional buy-in from partners is critical “to share and cross-promote”, and it’s worth remembering that many service users won’t be digital natives: “we often send printable materials via email that groups can print off and display in community centres and so on.” Another former colleague swears by sending printed materials to generate new business for her writing workshops as they are much harder to ignore than emails.

The bottom line is: the most successful brands get inside the limbic system of their audiences, creating content that will draw not just views but shares and likes too.

This post originally appeared on Linkedin.



bottom of page