Summarization in Prompt Engineering / GenAI - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

What is the main difference between extractive and abstractive summarization?

Summarization methods can be broadly categorized into extractive and abstractive. Which statement correctly describes their main difference?

A Extractive summarization selects key sentences from the original text, while abstractive summarization generates new sentences that capture the meaning.

B Extractive summarization generates new sentences, while abstractive summarization copies sentences directly from the text.

C Both extractive and abstractive summarization only select sentences without generating new text.

D Abstractive summarization only shortens text by removing stopwords, extractive summarization rewrites the text.

💡 Hint

Think about whether the summary uses original sentences or creates new ones.

Extractive summarization picks important sentences from the original text unchanged. Abstractive summarization rewrites or generates new sentences to convey the main ideas.

❓ Predict Output

intermediate

What is the output of this simple extractive summarization code?

Given the following Python code that extracts the first sentence as a summary, what will be printed?

Prompt Engineering / GenAI

text = "Machine learning helps computers learn from data. It is widely used in AI. Summarization is one application."
summary = text.split('.')[0] + '.'
print(summary)

A It is widely used in AI.

B Machine learning helps computers learn from data. It is widely used in AI.

C Summarization is one application.

D Machine learning helps computers learn from data.

💡 Hint

Look at how the text is split and which part is selected.

The code splits the text at periods and selects the first part, then adds a period. So it prints the first sentence only.

❓ Model Choice

advanced

Which model architecture is best suited for abstractive summarization?

You want to build an abstractive summarization system that generates new sentences. Which model architecture is most appropriate?

A K-means clustering algorithm

B Simple feedforward neural network

C Sequence-to-sequence model with attention mechanism

D Decision tree classifier

💡 Hint

Consider models that can generate sequences from input sequences.

Sequence-to-sequence models with attention can generate new text based on input sequences, making them ideal for abstractive summarization.

❓ Metrics

advanced

Which metric is commonly used to evaluate summarization quality?

When evaluating how good a summary is compared to a reference summary, which metric is most commonly used?

A Accuracy

B ROUGE score

C Mean Squared Error

D F1 score for classification

💡 Hint

Think about metrics that compare overlap of words or phrases.

ROUGE measures overlap of n-grams between generated and reference summaries, making it standard for summarization evaluation.

🔧 Debug

expert

Why does this abstractive summarization model output repetitive phrases?

Consider a trained abstractive summarization model that often repeats the same phrase multiple times in its output. What is the most likely cause?

A The model's decoding method lacks a mechanism to prevent repetition, such as coverage or beam search diversity.

B The training data contains only repetitive phrases, so the model learned to repeat.

C The input text is too short, causing the model to repeat phrases.

D The model uses a feedforward network instead of a recurrent network.

💡 Hint

Think about how models generate sequences and avoid repeating themselves.

Repetitive output often happens when the decoding strategy does not include mechanisms like coverage or diversity to avoid repeating the same phrases.