If a potential customer enjoys your advert, they are more likely to buy your product. It’s a simple enough concept, but it is extremely difficult to know how well your advert is being received in the real world. Now a new system could help advertisers know exactly how their latest offering is going down with viewers, just by watching their face.
The system, developed by Daniel McDuff and colleagues at the Massachusetts Institute of Technology Media Lab, looks at how muscles in the face move in response to watching a video. Software can then classify what counts as positive facial responses and smiles during the video and from that predict which adverts the viewer most enjoys.
The team collected more than 3200 videos of people, whose faces were filmed by their own computer’s webcam as they watched three adverts online during the Super Bowl in 2011. After each commercial – one for Doritos, one for Google and one for Volkswagen – the viewers were asked if they liked the video and whether they would want to watch it again. They had three choices of response: “Heck ya! I loved it!”, “Meh! It was ok” and “Na… not my thing”.
An algorithm was trained on an area around the viewers’ mouths to gauge how much they were smiling throughout each commercial. The smile intensity was tracked during the video and then used to predict whether the viewers would respond positively or negatively (the “Meh” response was ignored). In tests, the system made correct predictions more than 75 per cent of the time.
The system would be a boon for advertisers trying to grab the attention of potential customers on the internet, says Abhinav Dhall, who studies similar face recognition technology at the Australian National University in Canberra. “It is a showcase of mature emotion-recognition technologies that work well out of the lab environment,” he says.
In future, the system could be used to personalise adverts for viewers as they watch programmes online, depending on their reaction, or as a more effective way of testing how good a new advert is at appealing to customers. The work was presented at the IEEE’s Automatic Face and Gesture Recognition conference in Shanghai, China, last month.