Urbangems uses crowdsourcing to convert people's perceptions of neighbourhoods into quantities that capture fuzzy qualities such as calm, beauty, and happiness.


A user glances at two street views side-by-side, then votes on which one is more beautiful (or quiet or happy). The user has also to guess the fraction of individuals who would share the same view. The more the user guesses correctly, the higher his/her score. As each image is compared, a ranking of beautiful (or calm or happy) pictures emerges.


Initially, we are focusing on London: users are shown places that are considered beautiful/quiet. Users also receive personalised recommendation of places they might like based on the ratings of like-minded individuals.


Out of these rankings, we could answer questions like: Are certain areas seen as more beautiful? And, if so, why? What are the most common visual cues among pictures considered beautiful?
There has been extensive research on the relationship between urban perception and social deprivation. For example, in 1960, Kevin Lynch published "The Image of the City" and established how people perceive the cities they inhabit and what impression neighbourhoods left on them. In 1982, Wilson and Kelling put forward their theory of "broken windows" - cues of disorder in public are highly visible and constitute a salient marker of urban spaces, and "broken windows" (appearance) might lead to future crime (reality). More recently, Sampson has shown that perceptions of the same neighbourhoods differ among residents and are shaped by one's position in society (especially one's race).

So What (Criticism)

One problem with this study is that "what is perceived" is not necessarily "what is there". Users' votes might be influenced by: picture quality; position in society and race; and shared priors (e.g., reputation of a neighbourhood built over the years). A second problem is that the study cannot establish any casual mechanism - stimulate beautiful neighbourhoods might be less beneficial than reducing actual crime.