Paris Saclay Seminar

Efficiency vs Fairness in Link Recommendation Algorithms

Antoine Mandel (University Paris 1 Panthéon-Sorbonne)

Dec 12, 2024, 12:15

University of Evry Val d'Essonne

 

Abstract:

We study network dynamics driven by link recommendation algorithms and utility- based individual choices. Agents have preferences over links with objects (that can be their peers in the context of a social network) and have a given capacity for forming links. The network is controlled by a platform recommending to an agent a link to form at every time step. This mechanism induces a Markov process. When every new satisfactory link has a positive probability to be recommended, the process converges to an optimal or efficient network, i.e., a network in which every agent is linked to its preferred followees. We define the efficiency of an algorithm as the expected hitting time to the absorbing class of optimal networks. We the investigate the fairness of the recommendation algorithm on the basis of the stationary distribution of networks it generates. Our results provide insights on the design of fair recommendation algorithms and their regulation.
 

Location:

Room Malinvaud, CEPS University of Val d'Essonne
Boulevard François Mitterrand, 91025, Evry