How Particle Physics Can Improve Your Netflix Recommendations

Unlimited products are like bosons and restaurant reservations are like fermions.

Standard Model

Fermilab

From OpenTable to Amazon to your Netflix queue, algorithms sift through what we seem to like and offer future suggestions tailored to fit those trends. But the problem is they do this for everybody. So if everyone gets the same recommendations on OpenTable, everyone will try to reserve a table, and there won't be any seats left. What's more, if everyone gets a movie recommendation and everyone decides to watch it, the movie gets more popular--creating biases in the system. To improve matters, some researchers in Switzerland took a cue from the master rules of physics.

In particle physics, a given boson or lepton tends to occupy the most favorable energy state. If it's a force-carrying boson--like a photon, a W boson or a Higgs boson--there's no limit to how many particles can share real estate in that state. But if it's a fermion--like a quark, or an electron or proton--then only a certain number can be in the same place at the same time.

Algorithms should take this approach and function according to the rules of fermions rather than bosons, according to Stanislao Gualdi of the University of Fribourg and colleagues. After all, an object's utility declines with an increase in the number of people using it, they argue. It's like everyone buys the same guidebook and goes to the same quiet beach, meaning the beach is no longer quiet.

To study this concept, Gualdi and colleagues looked at DVD rentals. Using this model, a service like Netflix could limit the number of people who can have a single DVD at a time, forcing other DVDs to be recommended and chosen as secondary options. This limits biases that can happen when you give everyone unfettered access to the same thing, and this is good because it gives the whole recommendation engine some new fodder.

As Tech Review's arXiv blog points out, this is not necessarily a way to increase profits, so it's hard to see any recommendation service implementing the idea anytime soon. But it's an interesting concept.

"Crowd-avoidance can be applied to find a good compromise between satisfying the preferences of users and distributing them among objects evenly," the authors write. Their paper is posted to the arXiv preprint server.