引導式測驗 · 中等 等級
推薦系統測驗中的冷啟動問題
測試推薦器冷啟動:新用戶與新項目、輔助資訊、入門、探索、防洩漏評估、多樣性和隱私。
問題 1 的 8
A new user has not interacted with any items. Which problem is present?
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。
- A new user has not interacted with any items. Which problem is present?
- Item cold start because every item is new
- A data-labeling error
- A ranking-calibration problem only
- User cold start because the system lacks that person’s interaction history
- A newly listed book has no ratings or clicks. Which condition does this illustrate?
- User cold start
- Item cold start
- Feedback leakage
- A popularity prior
- How can item descriptions help a recommender before users interact with a new item?
- Content features can provide an initial representation for matching with user interests
- They guarantee the item will be popular
- They replace all user feedback permanently
- They reveal the author’s sales history
- Why can a popularity-only fallback disadvantage new creators?
- Popularity scores are always private
- New items have more ratings than established items
- Existing popular items may keep receiving exposure and interactions while new items remain unseen
- A fallback removes every catalog item
- Which evaluation split best tests a genuinely new item?
- Randomly hold out interaction rows for items still seen in training
- Train on future interactions and test on earlier ones
- Evaluate only on popular items with many ratings
- Hold out entire items from training and evaluate their first recommendations
- What tradeoff does exploration introduce?
- It guarantees immediate user satisfaction
- It eliminates the need for content data
- It can gather useful information but may show less relevant items in the short term
- It prevents the system from learning
- Why should onboarding preference questions be optional and limited?
- More personal data always improves recommendations
- They reduce friction and avoid collecting more personal information than needed
- Questions eliminate uncertainty
- Optional questions prevent evaluation
- Which metric complements ranking relevance when evaluating cold-start recommendations?
- Catalog coverage and exposure for new items
- Model parameter count
- Training time only
- Number of onboarding screens