Online social networks often mirror inequality in real-world networks, from historical prejudice, economic or social factors. Such disparities are often picked up and amplified by algorithms that leverage social data for the purpose of providing recommendations, diffusing information, or forming groups. In this talk, we’ll discuss possible explanations for algorithmic bias in social networks, specifically in (i) recommendation algorithms and (ii) the influence maximization problem. Using the preferential attachment model with unequal communities, we’ll characterize the relationship between homophily, network centrality, and bias through the power-law degree distributions of the nodes, and study the conditions in which diversity interventions can actually yield more efficient and equitable outcomes. In addition, we’ll see that recommendations which use the neighborhood of individuals may hinder the incoming connections of minority groups, while algorithms that use centrality-based measures in diffusing information may leave minorities out of the loop. In addition, we’ll discuss a set of heuristics that leverage the network structure to maximize the diffusion of a message while not creating disparate impact among participants based on sensitive demographics like gender or race. To wrap up, I’ll discuss research outputs from the Mechanism Design for Social Good initiative, in which I have been serving as a co-organizer for the last two years.
[…]