Cambridge, MA, Sept. 19, 2024 (GLOBE NEWSWIRE) — Online marketplace platforms often prioritize top-selling or well-known products in their item displays. Such favoritism, however, can limit content diversity and raise concerns about fair competition. From the user’s perspective, platforms aimed at maximizing revenue may not align with their preferences. This focus can lead to an emphasis on highly lucrative products or social influencers with large followings, rather than catering to many users’ genuine interests — particularly those that fall outside the mainstream — which may reduce platform loyalty.
Researchers at MIT Sloan School of Management have created two new algorithms to help solve these issues.
In one research paper, “Fair Assortment Planning,” Negin Golrezaei, MIT Sloan associate professor of operations management, MIT Sloan PhD candidate Qinyi Chen, and MIT PhD Fransisca Susan focused on item display fairness. Algorithms typically display items based on a score that calculates how likely an item is to be relevant to a particular user; however, only a small selection of available items typically pops up.
“Most assortment planning algorithms prioritize platform revenue at the expense of fairness, leading to a ‘rich get richer’ phenomenon where popular items continue to dominate,” said Golrezaei. By using the researchers’ algorithm, the platform can randomize over a small number of assortments, ensuring the platform revenue remains high while items are treated more fairly.
In a another research paper, “Interpolating Item and User Fairness in Recommendation Systems,” Golrezaei and Chen along with MIT PhD Jason Cheuk Nam Liang and Dajallel Bouneffouf with IBM Research, expanded on their research with a focus on algorithmic recommendations for bolstering user engagement and driving revenue. Recommendations, the authors noted, can impact multiple stakeholders simultaneously — the platform, sellers and customers — each with their unique objectives, making it difficult to find the right middle ground that accommodates all stakeholders. To address this, they have proposed a novel fair recommendation framework that aims to sustain platform revenue while aligning with user preferences, thereby ensuring fairness for both items and users.
“Existing methods often fail to balance fairness of display with learning user preferences and frequently overlook the platform’s objectives. Our algorithm is the first to address all three factors simultaneously,” said Golrezaei.
Navigating the complex landscape of fairness in recommendation systems requires a comprehensive approach that addresses the interests of all stakeholders and finds an appropriate middle ground, said Golrezaei. With food delivery apps, for instance, drives can also be taken into consideration as well as the restaurant, user, and the app itself.
“Our framework, which is freely available to platform companies, is general enough to be extended to include other stakeholders of interest, such as service providers,” said Gonrezaei. The platform can choose any ‘fairness notion’ that best suits each group of stakeholder’s needs and circumstances.”
Notably, some companies have just begun to address these issues. For example, LinkedIn’s initial recommendation algorithms favored users with more connections/activities, overlooking newer or less active members with relevant skills. In response, LinkedIn recently rolled out fairness toolkits to ‘provide equal opportunities to equally qualified members.’
“Putting legislative issues aside, being fair to users and items while keeping short-term costs low by considering the platform’s objectives is an investment in long-term growth,” said Golrezaei. She stresses that this approach attracts more users with potentially niche tastes and encourages more items to join the platform. In short, it provides an inclusive and welcoming environment for all the parties involved.
“Assuming there is no regulation in place, platforms can continue to ignore fairness,” Golrezaei said. “But in the long run they will suffer from this neglect. Our research shows that the price of fairness can be small and benefits for the entire system can be huge.”
- A New and Novel Fair Recommendation Framework