Public evidence on retention effects of item-level survey-based ranking

Establish whether publicly available empirical evidence demonstrates that incorporating predicted item-level survey responses into content ranking algorithms increases long-term user retention metrics (e.g., views, sessions) and clarify how this effect depends on survey wording and the specific retention definition over different time scales.

Background

In Section 5.4, the authors discuss internal observations from multiple platforms suggesting that predictions of item-level survey responses can correlate with long-term retention, while noting variability based on survey wording and retention definitions. However, they explicitly state a lack of clear public evidence verifying these effects.

This creates a concrete gap between internal platform findings and publicly documented, reproducible evidence, motivating the need to establish and publish empirical results that quantify the retention impact of using item-level survey predictions in ranking.