Californian researchers have been working on an algorithm capable of identifying the sub-culture to which a person belongs from photos posted on social networks, creating potential for more effective brand and message targeting.

A Person’s ‘Urban Tribe’ Will Soon Be Identifiable on Social Networks

In addition to providing instinctive tools for instant communication, social networks have been built around enabling people to highlight themselves. They use photos, comments and focuses of interest which have been carefully designed to provide the most individual and personal image of the user on the Internet. This personalisation trend has proved to be extremely useful both for sociological analysis and as regards marketing techniques and the drive to develop products which meet customer needs more precisely. But are we really as unique as we would like to believe? As emerged from the sociological analyses carried out in the 1960s and 70s, this apparent uniqueness promoted on the social networks is often only partial. In actual fact, individual people tend to adhere to a specific group or ‘sub-culture’. Researchers at the University of California have now developed an algorithm which enables them to identify a range of social groupings – which they call ‘urban tribes’ – on the basis of various pieces of information published on the social networks.

Uniformity in difference

In fact most of us are already aware of the existence of urban tribes, whether we are talking about the Surfer type, the Hipster or what the researchers call Formal. Such affiliations are often easily recognisable by other people.  However, up to now computers have not been able to recognise types of this kind with any degree of accuracy.  Although certain signs and signals can be highly charged with cultural significance, existing algorithms have so far not been able to derive a system from the values of these signals. This is in part what the researchers at the University of California have been doing. Using a method known as a ‘parts and attributes’ approach, this algorithm is able to analyse step by step the photos fed to it and highlight distinctive signs.  For instance, jewellery, tattoos, haircut and hair colour all enable the algorithm to accurately identify the associated sub-culture in 48% of cases. While this might not seem a very high success rate, if you compare it with other methods (for example chance produces the right answer only 9% of the time) it no longer looks so low. “This is a first step,” said Serge Belongie, a computer science professor at the Jacobs School of Engineering at the University of California, San Diego, who co-authored the report, adding: “We’re scratching the surface to figure out what the signals are.”  In an attempt to increase accuracy the research team have tried to integrate ‘accessory’ analysis, an analysis of facial features and other attributes within the system, extending the field of attributes to cover such items as makeup.

Group (p)re-targeting

While the technological accomplishment in itself is quite impressive, it is the potential spin-offs that are looking particularly valuable. The analysis started out from images posted on the social networks, the primary objective being to improve recommendations and sharing tools that put people who demonstrate similar social characteristics into the same informal grouping. By extension, the same principle can also be used to target advertising.  Re-targeting works on the basis of identifying website pages the user has visited and following up with relevant offers and messages. The new algorithm could now add a second level, targeting potential buyers in advance – pre-targeting – by identifying the consumption mode and/or products likely to be of interest to the ‘tribe’. The Californian researchers are now looking to further refine their research and identification criteria, but they also intend to extend their algorithm from working just on the basis of photos to working with video. This capability could then perhaps be used with surveillance systems, as a means of detecting risks and preventing incidents or crime.

By Quentin Capelle