Тест по анализу настроений

Ответы и пояснения появляются по мере прохождения теста.

1. Given “The screen is sharp, but setup failed,” why could one document label mislead?
   • It would compress praise and criticism about different aspects
   • It would automatically identify the writer
   • It would separate every word into a token
   • It would show the writer’s internal emotion
2. What helps two annotators label a complaint consistently?
   • A longer model name
   • Documented label and span rules
   • An unstated intuition
   • A larger font
3. Why evaluate a sentiment model on messages from the actual channel?
   • The channel changes the meaning of all labels
   • A model cannot process short text
   • Language and context differ across channels and domains
   • Evaluation removes the need for review
4. A message contains sarcasm, negation or a polite complaint. What is the key review?
   • Count only positive adjectives
   • Treat all politeness as praise
   • Ignore the sentence after “but”
   • Inspect context instead of relying on isolated positive words
5. A dataset is imbalanced across sentiment labels. Which reporting is useful?
   • Only the number of messages
   • Per-class precision and recall as well as overall accuracy
   • Only overall accuracy
   • The average word length
6. An average sentiment score comes from a small vocal group. What limitation should be stated?
   • It measures every customer equally
   • It establishes why users feel that way
   • It may not represent all users because participation and topics are selective
   • It proves the product is objectively good
7. Why keep representative messages alongside an aggregate trend?
   • They let reviewers inspect the pattern behind the summary
   • They prove all customers agree
   • They remove the need for consent rules
   • They convert estimates into facts
8. A positive sentiment label directly tells you that a person
   • Feels happy internally
   • Intends to purchase the product
   • Agrees with every aspect mentioned
   • Used positive language under a labeling scheme, not their private emotional state