Extent of amplification effects on out-of-network recommendations

Determine the extent to which algorithmic amplification of political groups and media sources observed in in-network tweets on Twitter/X affects out-of-network tweet recommendations in Twitter/X’s “For You” timelines, quantifying the magnitude and direction of exposure bias for accounts that users do not follow.

Background

The paper notes that approximately half of tweets in Twitter/X users’ timelines are out-of-network recommendations. Prior research has shown amplification of certain political groups and media sources within in-network content, but the authors highlight that how this amplification carries over to out-of-network recommendations is not yet characterized.

Because out-of-network tweets do not have a reverse-chronological baseline, measuring exposure bias is challenging. The authors motivate their sock-puppet audit as a way to paper algorithmic behavior and gauge whether similar amplification dynamics apply to recommended content outside users’ follow networks.