When browsing Netflix or shopping on Amazon, you might find yourself laughing at the occasional absurdity and incorrectness of the service’s recommendations. You know how it goes: “We see you bought an ethernet cable, perhaps you might like this rice cooker.” However, researchers at the University of Fribourg in Switzerland have found a way to use particle physics to help recommendation engines suggest things that you might actually enjoy.
If you’re fed up with ridiculous recommendations, Stanislao Gualdi and two fellow researchers feel they are on track to improving these engines to a point where they make sense more often than they do currently. The researchers first focused on an issue of recommendation engines that the general public may not usually consider – not if the user would actually like the recommendation, but what happens when too many users accept the recommendation.
For example, a restaurant discovery app might recommend a good Chinese restaurant, but if it recommends that place to a large number of people, the restaurant might become too packed with patrons, turning the outing into a poor experience. To solve this problem, the team looked to an unlikely arena: That of particle physics.
Photons can infinitely occupy a given state, whereas only one electron can occupy a given state. Revisiting the over-occupied restaurant example, Gualdi and his team set out to solve that issue by comparing the available-space-to-patron ratio to the occupancy states of particles. They tested their occupancy theory using DVD renting as a model, and found that even though they were focusing on avoiding a large crowd, their method helped increase overall recommendation accuracy. This is because of consumer bias.
The team found that biases are removed when lowering the number of people that can obtain a DVD, compared to when anyone and everyone can obtain the same DVD. They also found that limiting the amount of DVDs available forces users to find other things to watch and form opinions on, which leads to a broader range of recommendations. Essentially, the system is forcing you to find something else you like, because the original thing you wanted to access isn’t available. Devious.
The biggest problem with the most popular recommendation engines (Amazon and Netflix come to mind) is that there’s virtually no reason for the two companies to artificially limit their stock. Amazon would lose money if users looking for a video game couldn’t buy that specific game, and a core philosophy of Netflix Instant – “infinite copies” of digital media available to everyone whenever they want (licensing issues aside) – stands in direct contradiction to arbitrary limits.
So, how could this particle occupancy theory help Netflix better understand that just because you liked Lost doesn’t mean you’ll like Friday Night Lights – or that those two aren’t really related? It probably won’t, unless a retailer feels like being experimental enough to artificially limit the availability of its product, and forces you to find new things you like and then tell them about it.