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。

  1. 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
  2. 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
  3. 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
  4. 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
  5. 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
  6. 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
  7. 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
  8. 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