Bài kiểm tra có liên kết hướng dẫn · Trung bình Cấp độ

Bài kiểm tra đề xuất dựa trên phiên

Đề xuất dựa trên phiên kiểm tra: tương tác theo thứ tự, ranh giới phiên, mô hình trình tự, đánh giá mục tiếp theo, đường cơ sở, quyền riêng tư và ngắn hạn

Đường dẫn hướng dẫn liên quanĐề xuất dựa trên phiên
Câu hỏi 1 của 8

A visitor has no account history but has viewed several products in one visit. Which information can a session recommender use?

Xem trước mọi câu hỏi8 câu hỏi

Câu trả lời và giải thích xuất hiện khi bạn làm bài kiểm tra.

  1. A visitor has no account history but has viewed several products in one visit. Which information can a session recommender use?
    • A complete long-term profile that does not exist
    • The ordered interactions from the current visit
    • Only the product catalog’s average price
    • The visitor’s identity inferred from device ownership
  2. Why should a short sequence not automatically become a durable preference profile?
    • Session data always reveal stable taste
    • A session contains no item information
    • Long-term profiles cannot be corrected
    • It may reflect a temporary task, accidental clicks, or incomplete evidence
  3. How does a recurrent session model use item order?
    • It updates a hidden representation as events arrive in sequence
    • It treats every interaction as an unordered count
    • It uses only the first item viewed
    • It predicts without receiving any events
  4. Why test on later sessions instead of randomly splitting interaction rows?
    • Random splitting guarantees future conditions
    • Later sessions contain no user behavior
    • A chronological or session-level split reduces leakage between training and evaluation
    • Chronology matters only for image models
  5. What does Recall@K measure in a next-item task?
    • Whether a user liked every item in the catalog
    • Whether the observed next item appears within the top K recommendations
    • Thời gian cần thiết để huấn luyện mô hình
    • Whether the session boundary was legally correct
  6. Why can treating every unclicked item as disliked distort training data?
    • The user may never have noticed or been shown the item
    • Clicks are always accidental
    • Unclicked items are identical to purchases
    • A model cannot use negative examples
  7. What additional behavior should an online evaluation monitor beyond clicks?
    • Only model parameter count
    • Only page-load color
    • The number of candidate items before ranking
    • Returns, cancellations, coverage, and user controls
  8. RepeatNet-style models distinguish repeat behavior from exploration in which way?
    • Both always recommend the most popular item
    • Repeat behavior means repeating the same search query
    • The system models both returning to an earlier item and suggesting a different one
    • Exploration prevents a user from seeing prior items