Embedded Files
Recommender systems, moderation, search, sort, filter, and decision-support systems have largely been developed in parallel, despite aiming to solve similar computational problems over shared item universes. Together, these tools comprise visibility allocation systems (VASs) that decide which (processed) data to present a human user with. I investigate:
how to unify the language for these different tools through formal definitions of the computational problems they solve,
common frameworks and methodologies for evaluating their impact, and
using these frameworks, how VASs impact our society across applicaiton domains.
Poster_RecSys_VAS.pdfFormally introducing VASs
Formally introducing VASs
We first presented the work that formally defines VASs and their tools at RecSys'26 as part of the USRW. And this is our poster.
A visibility allocation system (VAS) is a (semi-)automated system that decides which (processed) data to present to a human user:
We mainly think about recommender systems.
But, in fact, a VAS can contain many other tools. A social media platform, for instance, also has moderation, search, filter, and ads shown via an auction.
It is often not one specific tool, but the VAS as a whole that creates a specific effect.
This is why our work introduces the term VAS and then presents a method for investigating its effects as a whole. We use VASs in school choice as an example. But this framework is based on our approach for researching several other applications:
VASs in the creator economy can lead to income inequalities. [1, 2, 3, 4]
VASs in online dating can affect the level of racial homogamy. [5]
VASs in refugee assignment can exacerbate employment inequalities. [6]
In these works, we connected VAS designs and societal effects. And this helped us find tool-specific interventions.
Page updated
Google Sites
Report abuse