Extractive summarization in NLP - Interactive Code Practice

Practice - 5 Tasks

1fill in blank

easy

Complete the code to import the library used for extractive summarization.

NLP

from sklearn.feature_extraction.text import [1]

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

ADictVectorizer

BCountVectorizer

CHashingVectorizer

DTfidfVectorizer

Attempts:

3 left

💡 Hint

Common Mistakes

Choosing CountVectorizer which only counts word occurrences without weighting.

Using HashingVectorizer which is less interpretable for summarization.

The TfidfVectorizer converts text to a matrix of TF-IDF features, which is commonly used in extractive summarization.

2fill in blank

medium

Complete the code to split the text into sentences for summarization.

NLP

import nltk
nltk.download('punkt')
sentences = nltk.tokenize.[1](text)

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

Aword_tokenize

Bsent_tokenize

Cregexp_tokenize

Dtweet_tokenize

Attempts:

3 left

💡 Hint

Common Mistakes

Using word_tokenize which splits text into words, not sentences.

Forgetting to download the 'punkt' package.

sent_tokenize splits text into sentences, which is essential for extractive summarization.

3fill in blank

hard

Fix the error in the code to compute cosine similarity matrix for sentence vectors.

NLP

from sklearn.metrics.pairwise import [1]
similarity_matrix = cosine_similarity(sentence_vectors)

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

Acosine_similarity

Beuclidean_distances

Cmanhattan_distances

Dpairwise_distances

Attempts:

3 left

💡 Hint

Common Mistakes

Using distance functions which measure dissimilarity, not similarity.

Importing a function but calling a different one.

cosine_similarity computes similarity between vectors, which is used to find sentence similarity in extractive summarization.

4fill in blank

hard

Fill both blanks to rank sentences using PageRank algorithm.

NLP

import networkx as nx
sentence_graph = nx.[1](similarity_matrix)
scores = nx.[2](sentence_graph)

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

Afrom_numpy_array

Bpagerank

Cdegree_centrality

Dto_numpy_array

Attempts:

3 left

💡 Hint

Common Mistakes

Using degree_centrality which is simpler but less effective for ranking.

Confusing graph creation and ranking functions.

We create a graph from the similarity matrix using from_numpy_array and rank sentences with pagerank.

5fill in blank

hard

Fill all three blanks to select top sentences and join them as summary.

NLP

top_sentences = sorted(((scores[i], s) for i, s in enumerate(sentences)), reverse=True)[:[1]]
summary = ' '.join([[2] for _, [3] in top_sentences])

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

A3

Bsentence

Cs

Dscore

Attempts:

3 left

💡 Hint

Common Mistakes

Joining scores instead of sentences.

Selecting too many or too few sentences.

We select top 3 sentences, then join the sentence strings (variable s) to form the summary.