Recommendation equation wins U team global honor

A group of U researchers was recognized for an algorithm that suggests products.

by Keaton Schmitt

A University team has garnered recognition for creating a formula responsible for curating the perfect Netflix marathon. 
The Seoul Test of Time Award — previously only claimed by the founders of Google — is designed to honor scientific papers with worldwide impacts. Three University of Minnesota professors and an alumnus will accept the award at the World Wide Web Conference Friday in Montreal, Canada. 
The group of recipients earned the award with its 2001 paper on recommender systems, used by companies like Netflix to suggest products to users based on what they’ve viewed before.
“The idea is to filter out a lot of information that exists out there to identify the few pieces of information that are most relevant to you,” said recipient George Karypis, a University computer science professor.
Before the paper, systems recommended products based on what other users consumed, Karypis said. The group’s method instead categorized items often viewed together, narrowing down the process. 
“[The paper] kind of flipped the equation. … Instead of basing recommendation on what similar users had done, it based recommendations on similar items,” he said.
This method gives better suggestions and lets companies find recommendations offline before users search, Karypis said. It works best for large companies like Netflix and in cases where there are more users than items.
Today, the method is still influential and a good deal of research in the last 15 years has used it as a starting point, said Mats Heimdahl, head of the University’s computer science and engineering department. 
“This algorithm is something that has spent quite a bit of time as the major commercial way people built [recommender systems]. … This was really the first big technique that was easily scalable to business scale,” said Joseph Konstan, recipient and University computer science professor.
Several companies, like, have used the method and even filed patents on it, Konstan said.
“I don’t think anybody back then saw the huge impact that was going to have 15 years later,” Heimdahl said. 
The field is still expanding, said Badrul Sarwar, award recipient and University alumnus who now works at LinkedIn. Netflix, for instance, awarded $1 million to anyone who could design a system better than its algorithm.
As recommender systems become more prevalent, several of the recipients are still working on them, Karypis said, including his team, which is specifically looking to broaden their uses.
Sarwar said the group is looking forward to reunite at the conference but will not be joined by John Reidl, who died from cancer in 2013 and will receive the award posthumously. 
He said Reidl greatly influenced his career, adding he feels fortunate he had Reidl as his Ph.D. adviser.
“One of the things [about] all the congratulatory emails that go back and forth, they made me realize it’s too bad John’s not around,” Karypis said. “He was a great guy, and he is missed in the department.”