গাইড লিঙ্ক ক্যুইজ · মাঝারি স্তর
নামকৃত সত্তা Rec. কুইজ
এনইআর স্কিমা, টোকেন লেবেল, স্প্যান বাউন্ডারি এবং অফসেট পর্যালোচনা করুন যখন সত্তার স্বীকৃতিকে আইডেন্টিটি লিঙ্কিং এবং যাচাইকৃত তথ্য থেকে আলাদা করে রাখুন।
প্রশ্ন 1 এর 8
Before annotating a corpus, what should a team define?
প্রতিটি প্রশ্নের পূর্বরূপ দেখুন8 প্রশ্ন
আপনি কুইজ নেওয়ার সাথে সাথে উত্তর এবং ব্যাখ্যাগুলি উপস্থিত হয়।
- 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
- 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
- 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
- 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
- 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
- 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
- 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
- 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