引導式測驗 · 中等 等級
YouTube 推薦演算法如何運作測驗
測試公共 YouTube 訊號、Home 與 Up Next 上下文、歷史候選者和排名研究以及推斷當前公式的限制。
問題 1 的 8
According to current YouTube Help, which signal is especially important for the Home page?
預覽每個問題8 問題
當您參加測驗時,就會出現答案和解釋。
- According to current YouTube Help, which signal is especially important for the Home page?
- The creator’s upload schedule only
- A universal list of channel subscribers
- Watch history
- Video tags alone
- Which signal does YouTube say can inform future recommendations after a viewer responds?
- A “Not interested” selection
- A video’s file name
- A creator’s advertising budget
- The order of search results on a different service
- What architecture did the 2016 YouTube recommendations paper describe?
- One global popularity list with no personalization
- A single text-search system for all surfaces
- A fixed sequence of user-created playlists only
- Candidate generation followed by a separate ranking model
- What does the candidate-generation stage do in the cited architecture?
- Rewrites viewer watch history
- Retrieves a smaller set of videos for later scoring or ranking
- Makes the final display order without another model
- Decides which videos are legally permitted
- How should the 2016 paper be used when discussing YouTube today?
- As a guarantee that the same exact model remains in production
- As proof that user controls no longer affect recommendations
- As the only source for all current YouTube features
- As a historical public system description, not a complete current internal specification
- Why can Up Next recommendations differ from Home recommendations?
- Home is always chronological while Up Next is always random
- Up Next ignores the video being watched
- Public YouTube guidance says the currently watched video is a key Up Next signal, while Home primarily uses watch history
- Both surfaces are stated to use only upload tags
- What does the general three-stage recommender pattern add beyond the paper’s two stages?
- A guarantee that all items are shown
- A re-ranking step that can apply additional constraints after scoring
- A mandatory human vote on every candidate
- A separate training-only stage before data collection
- What should a creator avoid inferring from one metric or public paper?
- A complete, permanent formula for how every current recommendation is ranked
- That Analytics can show traffic sources
- That viewer feedback may influence recommendations
- That recommendation behavior can vary by surface