Викторина, связанная с руководством · Средний Уровень

Именованный объект Rec. Викторина

Просматривайте схемы NER, метки токенов, границы диапазонов и смещения, сохраняя при этом распознавание объектов отличным от привязки идентификаторов и проверенных фактов.

Связанные направляющие путиРаспознавание именованного объекта
Вопрос 1 из 8

Before annotating a corpus, what should a team define?

Предварительный просмотр каждого вопроса8 вопросы

Ответы и пояснения появляются по мере прохождения теста.

  1. Before annotating a corpus, what should a team define?
    • Entity categories and span-boundary rules
    • The model vendor’s pricing plan
    • The number of pages in each file
    • A list of every person’s biography
  2. Why might “Labs” be included in “Northstar Research Labs” for one dataset?
    • All suffixes are always excluded
    • The annotation rules define whether organization suffixes are part of the span
    • NER decides company ownership
    • Tokenization makes boundaries irrelevant
  3. When token labels are aligned to subword tokens, what requires care?
    • Translating every token into a new language
    • Replacing offsets with a summary
    • Mapping labels back to the original text spans
    • Treating all subwords as separate people
  4. Why preserve character offsets for an extracted entity?
    • The system can infer a person’s identity
    • Offsets supply missing context automatically
    • They turn NER into sentiment analysis
    • An application can highlight the exact source span
  5. Which evaluation check matters for NER?
    • Check only whether a name appears somewhere
    • Assess both entity boundaries and entity types
    • Measure average sentence length only
    • Count model parameters
  6. A recognizer tags “Jordan” as a person. What would be needed to link it to a particular real-world record?
    • A longer NER label
    • Another punctuation mark
    • Separate entity linking with additional evidence
    • The same classifier run twice
  7. The system returns “Northstar” but the annotated organization span is “Northstar Research Labs.” What kind of error may this be?
    • A span-boundary error even if the type is right
    • Proof that entity linking succeeded
    • A sentiment classification error
    • A change in the source document
  8. Why is high average NER performance not enough for redaction?
    • Redaction does not use extracted spans
    • Every NER label hides the whole document
    • Offsets prevent misses
    • A missed sensitive name can remain visible despite a strong average score