5/ Examining the macro-dynamics, we used agent-based simulations integrating prior models of homophily and decoupling, showing that network composition matters: a sufficient presence of partial-overlap ties is crucial for widespread diffusion of novelty and the emergence of the unexpected.

4/ We find support for the hypothesized homophily trade-off: (a) stronger taste overlap → stronger peer influence, but also (b) stronger overlap → less exposure to novel content.

3/ To test this, we analyzed 1M Spotify users. Treating playlists as documents and artists as words, we used topic modeling to infer musical tastes, peer overlap, and exposure novelty. To distinguish selection & influence, we compared un/treated users with similar musical tastes—the key confounder.

2/ At the core is a hypothesized homophily trade-off: too much taste similarity→influence but no exposure to novelty; too little similarity→exposure to novelty but weak influence. Partial overlap strikes the balance, both exposing and influencing individuals toward content outside their repertoire.