The personalized feed curation process operates on a multi-stage scoring pipeline. When an uploader publishes a video, the engine routes it to a baseline test cluster of 300 to 500 active users. Depending on how that sample cohort interacts with the clip, the distribution either expands to broader interest clusters or flatlines entirely.
FYP ranking factors are not created equal. To visualize how independent research labs and developer audits break down the underlying signal hierarchy, the relative influence of these inputs can be mapped directly against audience behavioral patterns.
| Ranking Signal | Estimated Algorithmic Weight | Primary Trigger | Behavioral Impact |
|---|---|---|---|
| Full Completion & Looping | Extreme (Primary Multiplier) | Rewatching or finishing 100% of the runtime | Pushes video to wider test tiers immediately |
| Direct Share / Off-Platform Send | Very High | Copying link, sending via DM or messaging apps | Flags content as virally shareable and socially sticky |
| Comment Section Dwell Time | High | Reading or writing replies while audio loops | Inflates average session duration metrics |
| Explicit Likes & Favorites | Moderate | Tapping the heart or bookmark icon | Confirms topical affinity; weak indicator of retention |
| Metadata & Caption Tags | Moderate to Low | Hashtags, keywords, and description text | Assists initial indexing; overridden by behavioral data |
| Device & Account Settings | Baseline Filter | Language preference, country IP, OS type | Sets geographic boundaries; least influential for niche feeds |
External shares rank right behind video loops. When a user sends a video to an external messaging app, they become an unpaid acquisition channel. The platform rewards that distribution by surfacing the creator to comparable lookalike profiles across the network.
