Data-Mine Other People’s Flickr Photos to Generate Your Travel Itinerary
Say you have three days to spend in New York City, the . The Yahoo software starts by separating tourist photos taken in New York from photos posted by city-dwellers, using geolocation data to ensure a user's string of NYC photos covered a short span of time before moving on elsewhere. It then uses the frequency of different landmarks to determine what attractions the crowd finds most popular.
It then begins looking at the timestamps across that universe of NYC tourist photos. How long did most people spend at each attraction? How long did it take people, on average, to get from the Statue of Liberty downtown to the Metropolitan Museum uptown? From Rockefeller Center over to the nearby MoMA?
By crunching all this visual and geolocation data, the program can turn out a detailed itinerary to help travelers make the most of their limited time at a destination. Moreover, the crowdsourced itineraries scored rather well against professionally prepared itineraries among human travelers that were asked to compare the two, with 70% of human testers finding the computer generated itineraries superior to the travel agents'.
The problem with these crowd-based itineraries is that they favor the absolute middle. They average people's interest, and therefore come up with itineraries that are geared toward very general preferences. The next step is greater personalization -- allowing different travelers to input their personal interests and figure a way to have the program bias that itinerary toward Flickr users with similar interests.
In tests the program was limited to London, Paris, Barcelona, New York, and San Francisco, but with some further work and refinement could be extended to smaller and less popular destinations as well.
[]