WHAT do you fancy doing tonight? Just ask a bunch of strangers.
Online firms like Netflix and Amazon use algorithms to try to second-guess our desires. Now a team of researchers is bringing people back into the equation, using crowds of online workers to find your fancy.
Netflix-like algorithms work well when they have large amounts of data to learn from, but they fall down when asked to divine human preferences about sets of objects that are either very niche, personal or in flux.
Advertisement
That鈥檚 not a problem for humans, so Peter Organisciak at the University of Illinois at Urbana-Champaign and his team wondered if crowdsourcing could work out what we like with very little data to work with. To do so, they hired crowdsourcing workers on Mechanical Turk to make personal recommendations.
To test it, they took 100 different salt and pepper shakers from Amazon鈥檚 online store 鈥� some sleek and silver, some modelled after gnomes 鈥� and 100 photos of different types of meals from popular restaurants in Boston and San Francisco. The workers were presented with some of the shakers and meals and asked to give each one a suitability rating out of five for a target person. The only information these human recommendation engines, which Organisciak calls 鈥渢aste grokkers鈥�, had to go on was a small sample of the individual鈥檚 actual taste in shakers and food. They did well. The average rating from the top three recommenders matched the target person鈥檚 own ratings to within half a star. The results will be presented at the Conference on Human Computation & Crowdsourcing in Pittsburgh in November.
Organisciak says that using crowds rather than algorithms to determine preference is useful in personal data sets for which training an algorithm is impossible 鈥� identifying the best photographs from a large personal collection, for instance.
鈥淲hen you come back from vacation with 2000 photos it鈥檚 fun looking through them, but the whole task of culling it down to 50 for Facebook or 200 to show your family can be tiresome,鈥� he says. Paying a few dollars to crowdsource human opinion could remove that pain.
聯Crowdsourcing opinions on your 2000 holiday snaps could help you cull them to 50 for Facebook聰
Anand Kulkarni, CEO of crowdsourcing firm LeadGenius, says the technique is a great way to give people their own 鈥減ersonal shopper鈥� on the internet.
This article appeared in print under the headline 鈥淭he strangers who do your choosing for you鈥�
