Skip to Content

How Particle Physics Is Improving Recommendation Engines

The same physics that governs the behaviour of photons and electrons may also improve online shopping recommendations, say researchers

Most online shoppers will be familiar with phrases of the type “You liked X, so you might like Y” that are generated by the current crop of recommendation engines.  These play an increasingly important role for online retailers since they can increase sales by significant amounts.

And yet the actual recommendations can seem somewhat haphazard so there is no shortage of groups studying ways to improve them.

Today, Stanislao Gualdi at the University of Fribourg in Switzerland and a couple of pals say they’ve found a surprising new twist in this black art that increases the accuracy of these systems.

These guys begin by considering items that are adversely affected when too many people use them. For example, recommending a beach or a picnic spot because it is quiet can end up destroying the peace that gives it value. Similarly, restaurant recommendations can lead to overcrowding or difficulty getting a table which again makes the dining experience unpleasant. 

So Gualdi and co ask how to deal with recommendations for objects whose value diminishes with the number of people who use it. And their conclusions are surprising and counterintuitive. They say that this approach not only works well in these unusual circumstances but also improves conventional recommendations for objects or resources that are not in limited supply.

The approach these guys take is based on thinking used in particle physics, where particles tend to occupy the most energetically favourable states. If the particles are bosons, such as photons, there is no limit to the number that can occupy a given state.

But if they are fermions, like electrons, their physical properties dictate that no two can occupy the same state. Clearly the resulting distribution of these different types of particles is entirely different.The analogy here is with goods that any number of people can share or that only one person can have.

Gualdi and co’s use this physics-based approach to explore the space between these extremes where a single object/state can be shared with a relatively small number of users/particles.  In a sense, their model simulates a way to avoid crowds in restaurants, for example, while allowing small groups to dine.

These guys test the model on empirical data from DVD renting. They note that DVDs are not usually oversubscribed but say their approach produces surprising results. “We show that including crowd avoidance in the recommendation process can increase the accuracy of the recommendation,” they say.

The reason is subtle. They argue that by limiting the number of people who can have a single DVD, they remove biases that can emerge when an unlimited number of people have access.

This approach has other benefits too.  Preventing oversubscription ensures that the population of users sample a wider range of DVDs, which in turn provides a broader range of recommendations. This keeps the entire recommendation ecosystem healthy, say Gualdi and co.

That’s an interesting approach but one that is likely to need some significant real world testing and fine tuning. Retailers are not just interested in renting DVDs or selling books or whatever. They want to maximise profits.

So the idea of limiting the number of people who can rent the latest Hollywood blockbuster or buy the next Harry Potter on the day they are released may not be easy to swallow. Unless, of course, Gualdi and co can come up with good evidence that playing this game is more profitable in the long run.

But for and adventurous retailer, it might just be worth trying.

Ref: arxiv.org/abs/1301.1887: Crowd Avoidance and Diversity in Socio-Economic Systems and Recommendation

Keep Reading

Most Popular

Large language models can do jaw-dropping things. But nobody knows exactly why.

And that's a problem. Figuring it out is one of the biggest scientific puzzles of our time and a crucial step towards controlling more powerful future models.

OpenAI teases an amazing new generative video model called Sora

The firm is sharing Sora with a small group of safety testers but the rest of us will have to wait to learn more.

Google’s Gemini is now in everything. Here’s how you can try it out.

Gmail, Docs, and more will now come with Gemini baked in. But Europeans will have to wait before they can download the app.

This baby with a head camera helped teach an AI how kids learn language

A neural network trained on the experiences of a single young child managed to learn one of the core components of language: how to match words to the objects they represent.

Stay connected

Illustration by Rose Wong

Get the latest updates from
MIT Technology Review

Discover special offers, top stories, upcoming events, and more.

Thank you for submitting your email!

Explore more newsletters

It looks like something went wrong.

We’re having trouble saving your preferences. Try refreshing this page and updating them one more time. If you continue to get this message, reach out to us at customer-service@technologyreview.com with a list of newsletters you’d like to receive.