引導式測驗 · 中等 等級
Amazon 的逐項推薦引擎測驗
測試歷史上的 Amazon 逐項方法、它與基於用戶的過濾有何不同、其論文聲稱的內容以及對當前系統推理的限制。
問題 1 的 8
What did the 2003 Amazon paper call its recommendation approach?
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。
- What did the 2003 Amazon paper call its recommendation approach?
- A content-only catalog keyword index
- A hand-authored list of products for each shopper
- Item-to-item collaborative filtering
- Nearest-neighbor search over customer identities only
- How does item-to-item filtering differ from a user-based approach?
- It compares items rather than first finding customers with similar histories
- It compares customer demographics and ignores item data
- It ranks only products with paid advertisements
- It selects random products without interaction evidence
- What does the 2003 paper claim about online computation for its method?
- It requires comparing every customer with every other customer for each request
- It guarantees exact recommendations for all future Amazon users
- It eliminates the need to compute item relationships
- It scales independently of the number of customers and items in the catalog
- In the item-to-item pattern described here, what can start candidate retrieval for a customer?
- The item’s advertising bid only
- Items the customer viewed, purchased, or otherwise interacted with
- A demographic profile alone
- The most expensive product in the catalog
- Why can item-to-item filtering avoid a costly per-request user-neighbor search?
- It requires the system to scan the full catalog for every customer pair
- It predicts a recommendation without item or user data
- Item relationships can be computed separately and reused at serving time
- It never compares item relationships
- Which conclusion is not established by citing Amazon’s 2003 paper?
- That Amazon published an item-to-item approach in 2003
- That the paper compared item-based and traditional collaborative filtering
- That the paper discussed scaling in its own system
- That Amazon’s current production recommender still uses exactly that system
- What would a fair comparison of item-to-item candidates with another method require?
- Skip evaluation because item similarity is intuitive
- Evaluate both methods on the same recommendation task and suitable data split
- Compare one method’s offline score with another method’s marketing claim
- Use different users and labels for each method
- Which additional pieces may a modern recommender combine with item-to-item candidates?
- User context, ranking models, content signals, and constraints
- Only the 2003 paper’s original code
- A guarantee that all candidate items will be displayed
- No ranking or business rules