Crowdly

Додати до Chrome

Університети
vns.itstep.edu.ua
Natural Language Processing

Natural Language Processing

Шукаєте відповіді та рішення тестів для Natural Language Processing? Перегляньте нашу велику колекцію перевірених відповідей для Natural Language Processing в vns.itstep.edu.ua.

Отримайте миттєвий доступ до точних відповідей та детальних пояснень для питань вашого курсу. Наша платформа, створена спільнотою, допомагає студентам досягати успіху!

What is the key difference between abstractive and extractive summarization?

Extractive systems always produce higher-quality summaries than abstractive ones

Abstractive systems write new sentences that may not appear word-for-word in the source

Extractive systems copy sentences directly from the source document

Abstractive systems can introduce hallucinations — content not in the original text

Переглянути це питання

What is the main practical difference between BERT and GPT in terms of what tasks they are best at?

BERT is best for understanding tasks: classification, NER, and question answering

GPT is best for generation tasks: writing text, dialogue, and completion

BERT can generate fluent open-ended text just as well as GPT

GPT cannot be used for any classification task at all

Переглянути це питання

PEGASUS uses Gap-Sentence Generation (GSG) as its pre-training task. Which statements are correct?

GSG masks individual words randomly, just like BERT

This pre-training task closely resembles summarization, so the model needs less fine-tuning data

Entire sentences are removed from the document and the model must generate them

The sentences chosen for removal are the most important ones, measured by overlap with the rest of the document

Переглянути це питання

What are the weaknesses of representing a document as the average of its word vectors?

Word order is completely lost

Short documents always produce a zero vector

You cannot compute similarity between two documents this way

All words contribute equally, whether they are important or not

Переглянути це питання

Which types of sentences does a basic BiLSTM sentiment classifier tend to get wrong?

Very short, clearly positive or negative sentences

Sentences with negation like "not bad" or "not unpleasant"

Sarcastic sentences that require world knowledge

Sentences where the key sentiment word appears far from both ends

Переглянути це питання

Why is a very small learning rate like 2e-5 used when fine-tuning BERT?

BERT always needs small learning rates regardless of the task

Large learning rates would erase the pre-trained representations too quickly

To let the model adapt to the new task without destroying what it learned during pre-training

To avoid overwriting the useful knowledge already stored in the pre-trained weights

Переглянути це питання

Which statements about ROUGE and NLP fairness are correct?

Hard debiasing of word embeddings completely removes all gender bias from the model

In the COMPAS case, Black defendants had nearly twice the false positive rate of white defendants — a violation of equalized odds

ROUGE measures how much of the reference summary appears in the model's output

A summary can score high on ROUGE while still containing invented facts

Переглянути це питання

Setting max_df=0.7 in a TF-IDF vectorizer removes which words?

Words that appear in fewer than 70% of documents

Only stop words

Words that appear in more than 70% of documents

Words with a count below 70

Переглянути це питання

What does the transition matrix in a CRF layer do?

It is used by the Viterbi algorithm to find the best tag sequence at inference time

It helps the model predict globally consistent tag sequences for the whole sentence

It replaces the scores produced by the BiLSTM

It learns a score for every pair of consecutive tags (e.g., can I-PER follow B-PER?)

Переглянути це питання

Why can the BIOES tag scheme (Begin, Inside, Other, End, Single) work better than BIO for NER?

BIOES has fewer tags than BIO, making it simpler

The richer tag set gives the CRF more precise transition constraints to learn

It marks the end of entities explicitly, giving the model stronger signals at boundaries

Single-word entities get their own S- tag, avoiding confusion with multi-word entities

Переглянути це питання

Попередня
1
2
3
Наступна

Хочете миттєвий доступ до всіх перевірених відповідей на vns.itstep.edu.ua?

Отримайте необмежений доступ до відповідей на екзаменаційні питання - встановіть розширення Crowdly зараз!

Додати до Chrome

Telegram Instagram TikTok Question Bank

Умови використання Зв'яжіться з нами

Додати до Chrome

Додати до Chrome