引導式測驗 · 中等 等級
產品嵌入與項目相似性測驗
測試學習的項目向量、檢索使用、相似性測量、範數效應以及將向量鄰近度視為產品等價的限制。
問題 1 的 8
How does the guide use the term product embedding in recommendation?
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。
- How does the guide use the term product embedding in recommendation?
- A guarantee that two nearby products are interchangeable
- A learned vector representation used to relate an item to users, queries, or other items
- A fixed human-readable list of all product attributes
- A product’s inventory count stored as a single scalar
- How does cosine similarity differ from dot product in the cited Google guide?
- Cosine uses item price, while dot product uses product categories
- Dot product ignores magnitude, while cosine uses only vector length
- Both always rank every candidate in identical order
- Cosine compares vector direction, while dot product also depends on vector magnitudes
- Why might a dot-product retriever favor some frequently observed items?
- Their embedding norms may be larger, which can increase dot-product scores
- Frequent items always have better image quality
- Dot product removes popularity information
- The score ignores the learned vectors and uses item IDs only
- What can a high similarity score establish by itself?
- That the items are identical in human meaning
- That the recommendation will improve user outcomes
- That the vectors score as close under the chosen representation and metric
- That the products are safe substitutes in every use
- How can collaborative filtering learn item embeddings?
- By requiring every item to have a hand-written ontology
- By comparing product prices without interaction data
- By assigning random vectors that never change
- From user-item interaction patterns within a learned representation
- What challenge arises for an item with almost no interactions?
- It is guaranteed to rank above established items
- The model may lack collaborative evidence to learn a useful vector
- Its vector automatically becomes identical to popular items
- Cosine similarity cannot compare any vectors
- When could content features help an item recommender?
- They eliminate the need to evaluate retrieval quality
- They ensure all learned vectors remain fixed after training
- They can provide a representation for candidate retrieval when interaction history is sparse
- They prove that the item is a substitute for every neighbor
- Which check is appropriate before interpreting nearest products?
- Confirm the embedding objective and test retrieval for the intended task
- Assume the nearest item is a safe or equivalent replacement
- Treat vector coordinates as named product attributes
- Ignore the scoring method because embeddings are universal