Maximizing Influence
and Fairness
in Social Networks
at Scale

Our project updates influence maximization to find the most fair and impactful influencers.


We are replicating prior (unpublished) research from a team of researchers,
including Prof. Puya Vahabi and Prof. Alberto Todeschini at
University of California, Berkeley’s School of Information.

A large component of this project focused on incorporating our learnings from MIDS courses centered on ethics and privacy.

We utilizted two tools to accomplish this:

1. Deon Checklist
2. Consequences Scanning

It is important to note that fairness has several definitions mathematically and ethically. For this project, we utilized a definition of fairness similar to that of demographic parity.

Demographic parity might not be the best measurement in every context.

For example, we might want to consider demographic characteristics of the influencers we select (especially since being an influencer can be an extremely lucrative venture).

Another ethical consideration in this project is a limitation with respect to which demographic variables are included in the algorithm. For this project, we only looked at gender. Furthermore, we only considered two categories of gender (male or female) and ignored information spread amongst people who identify in other ways.

What if we end up with uneven information spread to individuals who identify as transgender (or another gender)? What if we manage to spread information evenly with respect to gender, but not with respect to income? Is this still fair?

Future research should expand not only to additional demographic characteristics, but should also consider intersectionality, or how different attributes might combine.

For example, we might spread information only amongst affluent females, but not females with low incomes. In the current application we would not have measured this but it would be unfair in most scenarios. We also might consider other definitions of fairness.

If you’d like to dig further into ethical considerations in this project, please visit our Github repository, found HERE.

Or email us at: team-influencers@googlegroups.com


B-Roll provided by Videezy

This 'hairball' structure is a subgraph created with GraphXR (Kineviz). You can explore a subset of the Sina Weibo social network to understand demographic makeup and visually understand how users can have varying fairness scores. Using your mouse, left click to move the graph network and right click to rotate the network.
Note: In some browsers, users encounter problems visualizing edges (the lines linking nodes). If these are not showing up for you, please try another browser (users have had luck with Chrome or Edge).