引導式測驗 · 中等 等級

如何使用 AI 測驗從會議文字稿中提取行動項目

在任務創建之前測試行動證據、承諾狀態、所有者、截止日期、成績單出處、隱私、基準限制和確認。

問題 1 的 8

A speaker says “we could send the draft Friday.” How should an extractor label it?

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. A speaker says “we could send the draft Friday.” How should an extractor label it?
    • As a possible action pending confirmation unless the group agrees
    • As a confirmed deadline for the speaker
    • As a completed task
    • As an automatic calendar event
  2. What evidence should be attached to an extracted action?
    • Only a confidence color
    • The meeting title alone
    • The transcript span, timestamp, or speaker reference supporting it
    • The model’s summary of the discussion
  3. A transcript does not specify an owner. What should the system do?
    • Assign the meeting organizer automatically
    • Mark the owner unresolved and request review
    • Choose the most senior attendee
    • Use the person who spoke last
  4. Why review inferred deadlines such as “next week”?
    • All systems interpret it as the same date
    • A due date is never needed
    • The project tracker chooses the right day
    • The phrase may lack an exact date or depend on meeting context
  5. What does QMSum provide?
    • A guarantee of perfect action-item extraction
    • A universal meeting recording law
    • A research benchmark for query-based meeting summarization
    • A tool for scheduling interview loops
  6. What should happen before extracted tasks are sent to a project tracker?
    • A reviewer confirms uncertain actions, owners, and due dates
    • The model publishes every candidate automatically
    • The transcript is deleted before checking
    • Only the meeting title is verified
  7. What should be checked before processing a meeting recording?
    • Whether the model can write a short summary
    • The meeting room’s screen size
    • Whether participants used the same calendar
    • Notice, consent, and retention rules for the organization and meeting type
  8. Which errors should evaluation track separately?
    • Only summary length
    • Missed tasks, false tasks, wrong owners, and invented deadlines
    • Only speech-to-text word count
    • The number of attendees