Questionário de classificação de texto

As respostas e explicações aparecem conforme você responde ao teste.

1. How does a text classifier structure the output for an input document?
   • By necessarily translating it into another language
   • By assigning one or more predefined category labels
   • By generating only a new unrelated paragraph
   • By changing its compressed file format
2. Qual cenário é um exemplo de classificação multirótulo?
   • Marcando um e-mail como spam ou não spam
   • Encaminhando um ticket para exatamente um departamento
   • Escolhendo uma classificação com uma única estrela
   • Marcar um artigo com 'política' e 'economia' ao mesmo tempo
3. On a text dataset dominated by one class, what can a high overall accuracy conceal?
   • An inability to calculate any metric
   • Proof that recall equals one for every category
   • Failure to identify cases of a rare class
   • A guarantee that every minority example is correct
4. O que a recordação mede em uma tarefa de classificação?
   • Quantos dos casos reais o sistema detectou com sucesso
   • Quantos itens sinalizados estavam realmente corretos
   • Quão rápido o modelo é executado
   • Qual é o tamanho do modelo
5. In task-specific zero-shot text classification, what is absent for the new classification task?
   • All model pretraining of any kind
   • Every description of the candidate categories
   • The ability to use any learned representation
   • Labeled training examples for that task
6. How can learned text embeddings contribute to a classification pipeline?
   • They guarantee perfect predictions on every new domain
   • They provide numerical features used to compute label scores
   • They delete every grammatical distinction by definition
   • They replace all label definitions with filenames
7. A stated text-classification evaluation has six true positives and two false positives. What is precision for the designated positive class?
   • 75 percent
   • 25 percent
   • 100 por cento
   • Six percent
8. Which output-score design permits several text labels to score highly at once rather than requiring scores to sum to one?
   • A single softmax over mutually exclusive classes
   • A rule requiring exactly one positive label
   • A separate sigmoid output for each label
   • A filename-only representation without label scores