Hong Kong researchers build tool to predict personal details from photos on social media

Team from University of Science and Technology says system can predict consumer behaviour

PUBLISHED : Wednesday, 29 November, 2017, 8:03pm
UPDATED : Wednesday, 29 November, 2017, 10:55pm

Researchers at a Hong Kong university have developed technology that can predict details about social media users by analysing the pictures they share online, without accessing their personal or sensitive information.

The tool, they say, will help to better predict consumer behaviour and tailor advertisements to specific users.

The researchers, from the University of Science and Technology, claim their algorithm is up to 60 per cent more accurate than similar prediction tools.

The technology, already employed in food and lifestyle mobile apps, could be further developed to make pre-emptive recommendations based on health and consumer patterns, the researchers said.

The team dismissed privacy concerns for the tool, insisting all data would be collected in the public domain or with the users’ consent.

Hong Kong chief executive calls for measures to attract best talent

The implications of using “big data”, referring to the vast amounts of information posted online each day, has become a buzzword amid the government’s push to develop Hong Kong into a smart city.

Chief Executive Carrie Lam Cheng Yuet-ngor announced the establishment of a big data analytics platform in her maiden policy address in October, while the Innovation and Technology Bureau has been tasked to formulate related policies.

The current research by the University of Science and Technology began five years ago.

With the help of the school’s Social Media Lab, lead researcher Cheung Pak-ming has collected 11 million photos shared by users from 150 countries to build a multimedia database.

Rather than just identifying objects in each photo, the system picks up “social signals” – visual features that disclose information about the users – for example, if the user frequently shares images of racing cars, then the user is most likely a male in his 40s.

Cheung said the technology now has an 80 per cent success rate in predicting a user’s gender by analysing their shared images.

“Generally speaking, we would be able to make an accurate prediction if we can gather 100 photos from a specific user,” he said.

Facebook changes algorithm on news feed to put the clamps on spammers

He said the team is now working with at least two lifestyle service firms and an outdoor advertising agency in Hong Kong.

“For example, apps can use our technology to provide restaurant suggestions to users … our tool has boosted the response rate of these suggestions by 60 per cent,” Cheung said.

He rejected privacy concerns.

“It all depends how the apps gather data from users. If you did not authorise the apps to access your private photos, then they would not be gathered and added to the database,” said James She, director of the Social Media Lab, who oversaw the research.

Funded by the university and Cyberport, the technology is now being patented, a process which can take five years.