Cornell researchers have developed a fairer system of recommendations — from hotels to jobs to videos — so a few big hits don’t get all the exposure.
The new ranking system still offers relevant options, but distributes users’ attention more evenly across search results. It can be applied to online marketplaces such as travel sites, recruitment platforms, and news aggregators.
Yuta Saito, doctoral student in the field of computer science and Thorsten Joachimsprofessor of computer and information science at Cornell Ann S. Bowers College of Computing and Information Science, described their new system in “Fair ranking as a fair division: individual fairness based on ranking impactpublished in the Proceedings of the Association for Computing Machinery’s 2022 Special Interest Group on Knowledge Discovery and Data Mining Conference.
“In recommender systems and search engines, whoever ranks high gets a lot of benefit,” Joachims said. “User attention is a finite resource and we need to distribute it evenly across elements.”
Conventional recommender systems attempt to rank items based solely on what users want to see, but many items receive unfairly low points in the order. Items with similar merit can end up far apart in the rankings, and for some items the chances of being discovered on a platform are worse than chance.
To correct this problem, Saito developed an improved ranking system based on ideas borrowed from economics. He applied the principles of “equitable distribution” – how to fairly distribute a limited resource, such as food, among members of a group.
Saito and Joachims demonstrated the feasibility of the ranking system using synthetic and real data. They found that it offers viable search results for the user, while fulfilling three fair distribution criteria: the benefit of each item from being ranked on the platform is better than being discovered randomly ; the impact of no element, such as income, can be easily improved; and no item would gain an advantage by changing its ranking relative to other items in a search series.
“We completely redefined ranking fairness,” Saito said. “It can be applied to any type of bilateral ranking system.”
If used on YouTube, for example, the recommendation system would present a more varied stream of videos, potentially distributing revenue more evenly to content creators. “We want to satisfy platform users, of course, but we also have to be fair to video creators, to maintain their diversity for the long term,” Saito said.
In online recruitment platforms, the fairer system would diversify search results, instead of showing the same top candidates to all employers.
Additionally, the researchers point out that this type of recommendation system could also help viewers discover new movies to watch online, allow scientists to find relevant presentations at conferences, and provide consumers with a more balanced selection of news. .
The National Science Foundation and the Funai Foundation for Information Technology funded the research.
Patricia Waldron is an editor at Cornell Ann S. Bowers College of Computing and Information Science.