Long document summarization strategies in NLP - Interactive Code Practice

Practice - 5 Tasks

1fill in blank

easy

Complete the code to load a pre-trained summarization model using Hugging Face Transformers.

NLP

from transformers import pipeline
summarizer = pipeline('summarization', model='[1]')

Drag options to blanks, or click blank then click option'

Agpt2

Broberta-base

Cfacebook/bart-large-cnn

Dbert-base-uncased

Attempts:

3 left

💡 Hint

Common Mistakes

Choosing a model not designed for summarization like 'bert-base-uncased'.

Using a language model like 'gpt2' which is not fine-tuned for summarization.

The facebook/bart-large-cnn model is a popular pre-trained model for summarization tasks.

2fill in blank

medium

Complete the code to split a long document into chunks of 500 tokens for summarization.

NLP

def chunk_text(text, chunk_size=500):
    tokens = text.split()
    return [tokens[i:i+[1]] for i in range(0, len(tokens), [1])]

Drag options to blanks, or click blank then click option'

A500

B250

C1000

D50

Attempts:

3 left

💡 Hint

Common Mistakes

Using different numbers for slicing and step size causing uneven chunks.

Choosing chunk sizes too small or too large for model input limits.

We split the text into chunks of 500 tokens to manage long documents effectively.

3fill in blank

hard

Fix the error in the code that summarizes each chunk and combines the results.

NLP

def summarize_long_text(text, summarizer):
    chunks = chunk_text(text, 500)
    summaries = []
    for chunk in chunks:
        summary = summarizer(' '.join(chunk), max_length=150, min_length=40, do_sample=False)[[1]]['summary_text']
        summaries.append(summary)
    return ' '.join(summaries)

Drag options to blanks, or click blank then click option'

A0

Btext

Csummary

D1

Attempts:

3 left

💡 Hint

Common Mistakes

Using index 1 which causes an index error.

Trying to access 'summary' or 'text' keys which do not exist.

The summarizer returns a list of dictionaries; the first element at index 0 contains the summary text.

4fill in blank

hard

Fill both blanks to create a function that uses a sliding window to summarize overlapping chunks.

NLP

def sliding_window_chunks(text, window_size=500, step_size=[1]):
    tokens = text.split()
    return [' '.join(tokens[i:i+window_size]) for i in range(0, len(tokens), [2]) if i + window_size <= len(tokens)]

Drag options to blanks, or click blank then click option'

A250

B500

C100

D50

Attempts:

3 left

💡 Hint

Common Mistakes

Using step size equal to window size, which causes no overlap.

Using step size larger than window size, which is invalid.

Using a step size of 250 creates overlapping chunks with a sliding window approach.

5fill in blank

hard

Fill all three blanks to build a pipeline that summarizes a long document by chunking, summarizing, and then summarizing the combined summary.

NLP

def hierarchical_summarization(text, summarizer):
    chunks = chunk_text(text, [1])
    summaries = [summarizer(' '.join(chunk), max_length=150, min_length=40, do_sample=False)[[2]]['summary_text'] for chunk in chunks]
    combined_summary = ' '.join(summaries)
    final_summary = summarizer(combined_summary, max_length=200, min_length=50, do_sample=False)[[3]]['summary_text']
    return final_summary

Drag options to blanks, or click blank then click option'

A500

B0

C1

D250

Attempts:

3 left

💡 Hint

Common Mistakes

Using wrong indices like 1 causing errors.

Using inconsistent chunk sizes.

Chunks of 500 tokens are summarized first. The summarizer output is accessed at index 0 for both steps.