Chapter 9: Text Summarization
9.1 Extractive Summarization
In today's era of information overload, the ability to automatically summarize lengthy text has become an essential tool. The field of Natural Language Processing (NLP) has developed a subfield specifically focused on text summarization, which involves creating a concise summary while retaining the most important points of the original text. Summaries can significantly enhance comprehension by reducing the amount of time needed to absorb information, making them useful in a variety of applications, such as news article summaries, customer review summaries, and summaries of scientific research papers.
There are two primary approaches to text summarization: extractive and abstractive. Extractive summarization involves selecting and extracting sentences that are deemed the most important in the original text, while abstractive summarization involves generating new sentences to summarize the content. In this chapter, we will delve into the details of both approaches, discussing their respective strengths and weaknesses, and providing examples of implementation for each.
Let's begin by exploring extractive summarization.
Extractive summarization is a technique that involves identifying the most important sections of a text and generating a summary by copying them verbatim. The key objective of this approach is to identify sentences or passages that provide a good representation of the overall content of the text.
This is a simpler approach as it does not require the model to generate new text, but simply to identify the most significant portions of the existing text. To achieve this goal, a variety of techniques can be used such as natural language processing, machine learning, and deep learning.
These techniques can help in the identification of relevant sentences and passages that can be used to generate a summary of the text. In addition, extractive summarization can be used in a variety of applications such as document summarization, news article summarization, and automatic text summarization. Extractive summarization is a useful technique for generating summaries of long texts in an efficient and accurate manner.
9.1.1 Principle of Extractive Summarization
The extractive summarization technique is a powerful tool used to summarize a source document. It works on the principle of selection and ranking, where the algorithm selects the most important sentences from the document and concatenates them to form a summary.
The ranking of the sentences can be determined by various features such as the frequency of words, sentence length, the presence of named entities, or the location of a sentence in the document. The higher the rank of a sentence, the more important it is in the summary.
This technique is particularly useful when dealing with large volumes of text, as it can quickly generate a summary that captures the key ideas of the source document. Extractive summarization can be used to highlight the most important points of a document, making it easier for readers to identify the key takeaways.
The extractive summarization technique is an effective way to condense large amounts of information into a concise and easily digestible format.
9.1.2 Techniques for Extractive Summarization
Several methods can be used for extractive summarization. Here are a few common ones:
Frequency-based Approach
This is the simplest approach to summarize a document. It involves calculating the frequency of each word in the document, excluding stop words. Stop words are commonly used words such as "the", "a", "an", etc. After the frequency of each word is calculated, the sentences in the document are ranked based on the sum of the frequencies of the words in each sentence.
This approach is simple and easy to implement, but may not always provide the most accurate summary of the document.
There are other approaches to summarization that take into account the context and meaning of words, which may result in a more precise summary. However, the frequency-based approach is still widely used due to its simplicity and efficiency.
Graph-based Approach
The graph-based approach is a popular method for text summarization. In this method, a graph is constructed where each sentence in the document is a node. The edges between the nodes are weighted based on the similarity between the sentences. By using a ranking algorithm, such as PageRank, to score the sentences, the most important sentences can be identified and used to create a summary of the document.
This approach is particularly useful for dealing with large amounts of text, as it can quickly identify the most important sentences without requiring manual intervention. Furthermore, it is highly flexible and can be adapted to different types of documents and summarization goals. Overall, the graph-based approach is a powerful tool for summarizing text and extracting the most important information from a document.
Feature-based Approach
The Feature-based Approach to summarization involves the selection of a determined set of features that represent the importance of a sentence. These features may include sentence length, term frequency, named entities, among others. The purpose of this method is to train a machine learning model which will then be used to rank the sentences based on the aforementioned features.
By doing this, the most important sentences are selected and used to create a summary of the text. This method is particularly useful for large texts where manual summarization would be time-consuming and inefficient. It is important to note, however, that the selection of the features is crucial to the success of this approach.
Therefore, extensive research and analysis must be conducted beforehand to ensure that the most relevant features are selected and utilized in the training of the machine learning model.
Let's now look at a simple code example of extractive summarization using the Python library gensim.
Example:
from gensim.summarization import summarize
# Assume text contains the original document text
summary = summarize(text)
print(summary)In this example, we use the summarize function from gensim, which implements an extractive summarization method based on the TextRank algorithm. TextRank is a graph-based ranking model for text processing which can be used for extractive summarization. It ranks the sentences in the document based on their similarity to other sentences. The top-ranked sentences are then included in the summary.
Remember that while extractive summarization methods are effective and relatively simple to implement, they have some limitations. For instance, they can't handle nuances and may miss important information that is spread across multiple sentences.
Also, they can't generate summaries in a different wording or style than the source text, as they merely copy selected sentences from the original text. These limitations lead us to the other main approach to text summarization, called abstractive summarization, which we'll cover in the next section.
9.1.3 Practical Example
To further enrich the understanding of extractive summarization, let's take a deeper dive into a practical Python example that uses the networkx library and sklearn to implement an extractive summarization technique.
This example will create a matrix of sentence similarities, then use this matrix to create a graph with networkx. The PageRank algorithm, which is a part of the networkx library, will then be used to rank the sentences based on their scores.
For this example, we'll use the NLTK library to download the stop words and the punkt package, which is a sentence tokenizer.
import nltk
nltk.download('punkt') # for sentence tokenization
nltk.download('stopwords') # for removing stop wordsNext, we'll install and import all the necessary libraries:
!pip install numpy networkx sklearn
import numpy as np
import networkx as nx
from nltk.corpus import stopwords
from nltk.tokenize import sent_tokenize
from sklearn.metrics.pairwise import cosine_similarity
from sklearn.feature_extraction.text import CountVectorizerNow, let's define a function that calculates the similarity matrix:
def build_similarity_matrix(sentences, stopwords=None):
# Create a CountVectorizer object for transforming the text data to a matrix of token counts
count_vectorizer = CountVectorizer(stop_words=stopwords)
# Transform the sentences to a matrix of token counts
count_matrix = count_vectorizer.fit_transform(sentences)
# Compute cosine similarity between the sentences
sentence_similarity = cosine_similarity(count_matrix, count_matrix)
return sentence_similarityWith this function, we can create a graph and calculate the scores of the sentences:
def generate_summary(text, top_n=5):
stop_words = stopwords.words('english')
summarize_text = []
# Step 1 - Read text and split it
sentences = sent_tokenize(text)
# Step 2 - Generate similarity matrix across sentences
sentence_similarity_matrix = build_similarity_matrix(sentences, stop_words)
# Step 3 - Rank sentences in the similarity matrix
sentence_similarity_graph = nx.from_numpy_array(sentence_similarity_matrix)
scores = nx.pagerank(sentence_similarity_graph)
# Step 4 - Sort the rank and pick the top sentences
ranked_sentence = sorted(((scores[i], s) for i, s in enumerate(sentences)), reverse=True)
for i in range(top_n):
summarize_text.append(" ".join(ranked_sentence[i][1]))
# Step 5 - Output the summarized text
return ". ".join(summarize_text)
# Assume we have some long text data in the variable 'text'
print("Summary: \n", generate_summary(text, 2))This code will output the top 2 sentences from the text as the summary. This is a simple yet effective way to perform extractive summarization by ranking sentences according to their similarity with other sentences in the document.
Keep in mind that while this method works well for many tasks, it can sometimes oversimplify or miss important nuances in the text. This is due to the fact that it is extracting sentences without considering how they might be rephrased or summarized more succinctly. For more complex tasks, it might be necessary to use more sophisticated methods like abstractive summarization, which we'll explore in the next topic.
9.1.4 Evaluation Metrics for Extractive Summarization
ROUGE Score
ROUGE is an acronym for Recall-Oriented Understudy for Gisting Evaluation. It is a set of metrics that is used to evaluate the quality of automatic summarization and machine translation. The primary way that this is done is by comparing an automatically generated summary or translation with a set of reference summaries, which are typically created by humans.
ROUGE is an important tool for researchers who are working on improving the accuracy and effectiveness of automatic summarization and machine translation systems. By using ROUGE, researchers can get a better understanding of how well their systems are performing and what areas need improvement.
In addition to being a useful tool for researchers, ROUGE is also used in industry. Many companies that produce automatic summarization and machine translation software use ROUGE to evaluate their products and make sure that they are meeting the needs of their customers.
Overall, ROUGE is an essential part of the field of natural language processing and is likely to continue playing an important role in the development of automatic summarization and machine translation technology in the years to come.
ROUGE-N, ROUGE-L, and ROUGE-S are the most commonly used ROUGE metrics:
- ROUGE-N: Measures the number of matching n-grams. This is to say, ROUGE-1 refers to the overlap of unigrams between the system and reference summaries. ROUGE-2 refers to the overlap of bigrams.
- ROUGE-L: Measures the longest matching sequence of words using LCS (Longest Common Subsequence). It does not require consecutive matches but in-sequence matches.
- ROUGE-S: Measures the skip-bigram co-occurrence statistics. Skip-bigram is any pair of words in their sentence order.
Example:
!pip install rouge
from rouge import Rouge
def evaluate_summary(summary, reference):
rouge = Rouge()
scores = rouge.get_scores(summary, reference)
return scores
# Assume we have a generated summary 'gen_summary' and a reference summary 'ref_summary'
print(evaluate_summary(gen_summary, ref_summary))This code calculates and prints out the ROUGE-1, ROUGE-2, and ROUGE-L scores for the generated summary compared to the reference.
BLEU Score
The Bilingual Evaluation Understudy (BLEU) score is a metric that measures the quality of a summary. Although it is commonly used in machine translation, it can also be used for text summarization tasks. BLEU analyzes the overlap between the predicted and actual words and phrases, up to a specified length (such as n-grams).
It is worth noting, however, that the BLEU score function in NLTK is more effective for tasks where multiple reference summaries are possible, such as machine translation. In tasks like text summarization, where there is usually only one reference summary, BLEU may not be the best choice. Despite this, BLEU remains a widely used metric for evaluating the quality of text summaries and is an important tool for researchers and practitioners alike.
These evaluation metrics provide a quantitative way to measure the quality of a text summary. However, they still have limitations and do not always align perfectly with human judgement, as they mostly rely on word matching and can overlook semantic understanding.
9.1.5 Limitations and Challenges of Extractive Summarization
Extractive summarization, while useful, does have certain limitations:
Lack of coherence
One of the potential issues with using extractive techniques for summarization is that sentences are chosen based solely on their individual scores. While this approach can be effective at capturing important information, it may not fully consider the overall flow or structure of the original text. This can result in summaries that lack coherence and fail to convey a clear narrative.
It is important to note that extractive techniques often ignore the broader context of the text. As a result, the summary may miss key themes or ideas that are only apparent when looking at the text as a whole. Additionally, these techniques may not be able to capture the nuance and complexity of certain passages, leading to summaries that oversimplify or distort the original meaning.
To address these issues, some researchers have proposed using more advanced techniques for summarization, such as abstractive techniques that generate summaries by synthesizing new text rather than simply selecting and reordering existing sentences. While these techniques are still in the early stages of development, they offer the potential for more coherent and comprehensive summaries that capture the full scope of the original text.
Redundancy
One potential issue with the extractive approach is that it may select and include the same information multiple times in the summary, particularly if the source text includes repetitive information.
This can result in a summary that is not only shorter than the original text, but also lacks clarity and conciseness. In order to avoid this issue, it may be necessary to implement techniques such as sentence clustering or topic modeling to identify and eliminate redundant information from the summary. A human editor may need to review the summary to ensure that it accurately captures the key ideas of the source text in a comprehensive and cohesive manner.
Lack of abstraction
Extractive summarization does not generate new sentences or paraphrase the information, which could be a disadvantage for certain applications where a more human-like summary is desired.
One potential disadvantage of extractive summarization is its lack of abstraction. Unlike other summarization techniques, extractive summarization does not have the ability to generate new sentences or paraphrase the information it is summarizing.
This can be a problem in certain applications where a more human-like summary is desired, as the extractive summary may not capture the nuance and complexity of the original text. Therefore, it may be necessary to use alternative summarization techniques in these cases, or to supplement the extractive summary with additional analysis and interpretation.
Missing important information
One issue that can arise when using extractive summarization is that important information may be spread across different parts of the text and not encapsulated in a single sentence. This can lead to a less comprehensive summary that misses key details.
For example, if a long article discusses a complex topic, a summary that only includes the most frequently occurring words or phrases may not capture the nuances and complexities of the original text. As a result, it is important to carefully review summaries generated through extractive techniques to ensure that all pertinent information is included.
Understanding these limitations is essential as it helps in choosing the right summarization technique for a given task and provides directions for future research and improvements in extractive summarization techniques.
9.1 Extractive Summarization
In today's era of information overload, the ability to automatically summarize lengthy text has become an essential tool. The field of Natural Language Processing (NLP) has developed a subfield specifically focused on text summarization, which involves creating a concise summary while retaining the most important points of the original text. Summaries can significantly enhance comprehension by reducing the amount of time needed to absorb information, making them useful in a variety of applications, such as news article summaries, customer review summaries, and summaries of scientific research papers.
There are two primary approaches to text summarization: extractive and abstractive. Extractive summarization involves selecting and extracting sentences that are deemed the most important in the original text, while abstractive summarization involves generating new sentences to summarize the content. In this chapter, we will delve into the details of both approaches, discussing their respective strengths and weaknesses, and providing examples of implementation for each.
Let's begin by exploring extractive summarization.
Extractive summarization is a technique that involves identifying the most important sections of a text and generating a summary by copying them verbatim. The key objective of this approach is to identify sentences or passages that provide a good representation of the overall content of the text.
This is a simpler approach as it does not require the model to generate new text, but simply to identify the most significant portions of the existing text. To achieve this goal, a variety of techniques can be used such as natural language processing, machine learning, and deep learning.
These techniques can help in the identification of relevant sentences and passages that can be used to generate a summary of the text. In addition, extractive summarization can be used in a variety of applications such as document summarization, news article summarization, and automatic text summarization. Extractive summarization is a useful technique for generating summaries of long texts in an efficient and accurate manner.
9.1.1 Principle of Extractive Summarization
The extractive summarization technique is a powerful tool used to summarize a source document. It works on the principle of selection and ranking, where the algorithm selects the most important sentences from the document and concatenates them to form a summary.
The ranking of the sentences can be determined by various features such as the frequency of words, sentence length, the presence of named entities, or the location of a sentence in the document. The higher the rank of a sentence, the more important it is in the summary.
This technique is particularly useful when dealing with large volumes of text, as it can quickly generate a summary that captures the key ideas of the source document. Extractive summarization can be used to highlight the most important points of a document, making it easier for readers to identify the key takeaways.
The extractive summarization technique is an effective way to condense large amounts of information into a concise and easily digestible format.
9.1.2 Techniques for Extractive Summarization
Several methods can be used for extractive summarization. Here are a few common ones:
Frequency-based Approach
This is the simplest approach to summarize a document. It involves calculating the frequency of each word in the document, excluding stop words. Stop words are commonly used words such as "the", "a", "an", etc. After the frequency of each word is calculated, the sentences in the document are ranked based on the sum of the frequencies of the words in each sentence.
This approach is simple and easy to implement, but may not always provide the most accurate summary of the document.
There are other approaches to summarization that take into account the context and meaning of words, which may result in a more precise summary. However, the frequency-based approach is still widely used due to its simplicity and efficiency.
Graph-based Approach
The graph-based approach is a popular method for text summarization. In this method, a graph is constructed where each sentence in the document is a node. The edges between the nodes are weighted based on the similarity between the sentences. By using a ranking algorithm, such as PageRank, to score the sentences, the most important sentences can be identified and used to create a summary of the document.
This approach is particularly useful for dealing with large amounts of text, as it can quickly identify the most important sentences without requiring manual intervention. Furthermore, it is highly flexible and can be adapted to different types of documents and summarization goals. Overall, the graph-based approach is a powerful tool for summarizing text and extracting the most important information from a document.
Feature-based Approach
The Feature-based Approach to summarization involves the selection of a determined set of features that represent the importance of a sentence. These features may include sentence length, term frequency, named entities, among others. The purpose of this method is to train a machine learning model which will then be used to rank the sentences based on the aforementioned features.
By doing this, the most important sentences are selected and used to create a summary of the text. This method is particularly useful for large texts where manual summarization would be time-consuming and inefficient. It is important to note, however, that the selection of the features is crucial to the success of this approach.
Therefore, extensive research and analysis must be conducted beforehand to ensure that the most relevant features are selected and utilized in the training of the machine learning model.
Let's now look at a simple code example of extractive summarization using the Python library gensim.
Example:
from gensim.summarization import summarize
# Assume text contains the original document text
summary = summarize(text)
print(summary)In this example, we use the summarize function from gensim, which implements an extractive summarization method based on the TextRank algorithm. TextRank is a graph-based ranking model for text processing which can be used for extractive summarization. It ranks the sentences in the document based on their similarity to other sentences. The top-ranked sentences are then included in the summary.
Remember that while extractive summarization methods are effective and relatively simple to implement, they have some limitations. For instance, they can't handle nuances and may miss important information that is spread across multiple sentences.
Also, they can't generate summaries in a different wording or style than the source text, as they merely copy selected sentences from the original text. These limitations lead us to the other main approach to text summarization, called abstractive summarization, which we'll cover in the next section.
9.1.3 Practical Example
To further enrich the understanding of extractive summarization, let's take a deeper dive into a practical Python example that uses the networkx library and sklearn to implement an extractive summarization technique.
This example will create a matrix of sentence similarities, then use this matrix to create a graph with networkx. The PageRank algorithm, which is a part of the networkx library, will then be used to rank the sentences based on their scores.
For this example, we'll use the NLTK library to download the stop words and the punkt package, which is a sentence tokenizer.
import nltk
nltk.download('punkt') # for sentence tokenization
nltk.download('stopwords') # for removing stop wordsNext, we'll install and import all the necessary libraries:
!pip install numpy networkx sklearn
import numpy as np
import networkx as nx
from nltk.corpus import stopwords
from nltk.tokenize import sent_tokenize
from sklearn.metrics.pairwise import cosine_similarity
from sklearn.feature_extraction.text import CountVectorizerNow, let's define a function that calculates the similarity matrix:
def build_similarity_matrix(sentences, stopwords=None):
# Create a CountVectorizer object for transforming the text data to a matrix of token counts
count_vectorizer = CountVectorizer(stop_words=stopwords)
# Transform the sentences to a matrix of token counts
count_matrix = count_vectorizer.fit_transform(sentences)
# Compute cosine similarity between the sentences
sentence_similarity = cosine_similarity(count_matrix, count_matrix)
return sentence_similarityWith this function, we can create a graph and calculate the scores of the sentences:
def generate_summary(text, top_n=5):
stop_words = stopwords.words('english')
summarize_text = []
# Step 1 - Read text and split it
sentences = sent_tokenize(text)
# Step 2 - Generate similarity matrix across sentences
sentence_similarity_matrix = build_similarity_matrix(sentences, stop_words)
# Step 3 - Rank sentences in the similarity matrix
sentence_similarity_graph = nx.from_numpy_array(sentence_similarity_matrix)
scores = nx.pagerank(sentence_similarity_graph)
# Step 4 - Sort the rank and pick the top sentences
ranked_sentence = sorted(((scores[i], s) for i, s in enumerate(sentences)), reverse=True)
for i in range(top_n):
summarize_text.append(" ".join(ranked_sentence[i][1]))
# Step 5 - Output the summarized text
return ". ".join(summarize_text)
# Assume we have some long text data in the variable 'text'
print("Summary: \n", generate_summary(text, 2))This code will output the top 2 sentences from the text as the summary. This is a simple yet effective way to perform extractive summarization by ranking sentences according to their similarity with other sentences in the document.
Keep in mind that while this method works well for many tasks, it can sometimes oversimplify or miss important nuances in the text. This is due to the fact that it is extracting sentences without considering how they might be rephrased or summarized more succinctly. For more complex tasks, it might be necessary to use more sophisticated methods like abstractive summarization, which we'll explore in the next topic.
9.1.4 Evaluation Metrics for Extractive Summarization
ROUGE Score
ROUGE is an acronym for Recall-Oriented Understudy for Gisting Evaluation. It is a set of metrics that is used to evaluate the quality of automatic summarization and machine translation. The primary way that this is done is by comparing an automatically generated summary or translation with a set of reference summaries, which are typically created by humans.
ROUGE is an important tool for researchers who are working on improving the accuracy and effectiveness of automatic summarization and machine translation systems. By using ROUGE, researchers can get a better understanding of how well their systems are performing and what areas need improvement.
In addition to being a useful tool for researchers, ROUGE is also used in industry. Many companies that produce automatic summarization and machine translation software use ROUGE to evaluate their products and make sure that they are meeting the needs of their customers.
Overall, ROUGE is an essential part of the field of natural language processing and is likely to continue playing an important role in the development of automatic summarization and machine translation technology in the years to come.
ROUGE-N, ROUGE-L, and ROUGE-S are the most commonly used ROUGE metrics:
- ROUGE-N: Measures the number of matching n-grams. This is to say, ROUGE-1 refers to the overlap of unigrams between the system and reference summaries. ROUGE-2 refers to the overlap of bigrams.
- ROUGE-L: Measures the longest matching sequence of words using LCS (Longest Common Subsequence). It does not require consecutive matches but in-sequence matches.
- ROUGE-S: Measures the skip-bigram co-occurrence statistics. Skip-bigram is any pair of words in their sentence order.
Example:
!pip install rouge
from rouge import Rouge
def evaluate_summary(summary, reference):
rouge = Rouge()
scores = rouge.get_scores(summary, reference)
return scores
# Assume we have a generated summary 'gen_summary' and a reference summary 'ref_summary'
print(evaluate_summary(gen_summary, ref_summary))This code calculates and prints out the ROUGE-1, ROUGE-2, and ROUGE-L scores for the generated summary compared to the reference.
BLEU Score
The Bilingual Evaluation Understudy (BLEU) score is a metric that measures the quality of a summary. Although it is commonly used in machine translation, it can also be used for text summarization tasks. BLEU analyzes the overlap between the predicted and actual words and phrases, up to a specified length (such as n-grams).
It is worth noting, however, that the BLEU score function in NLTK is more effective for tasks where multiple reference summaries are possible, such as machine translation. In tasks like text summarization, where there is usually only one reference summary, BLEU may not be the best choice. Despite this, BLEU remains a widely used metric for evaluating the quality of text summaries and is an important tool for researchers and practitioners alike.
These evaluation metrics provide a quantitative way to measure the quality of a text summary. However, they still have limitations and do not always align perfectly with human judgement, as they mostly rely on word matching and can overlook semantic understanding.
9.1.5 Limitations and Challenges of Extractive Summarization
Extractive summarization, while useful, does have certain limitations:
Lack of coherence
One of the potential issues with using extractive techniques for summarization is that sentences are chosen based solely on their individual scores. While this approach can be effective at capturing important information, it may not fully consider the overall flow or structure of the original text. This can result in summaries that lack coherence and fail to convey a clear narrative.
It is important to note that extractive techniques often ignore the broader context of the text. As a result, the summary may miss key themes or ideas that are only apparent when looking at the text as a whole. Additionally, these techniques may not be able to capture the nuance and complexity of certain passages, leading to summaries that oversimplify or distort the original meaning.
To address these issues, some researchers have proposed using more advanced techniques for summarization, such as abstractive techniques that generate summaries by synthesizing new text rather than simply selecting and reordering existing sentences. While these techniques are still in the early stages of development, they offer the potential for more coherent and comprehensive summaries that capture the full scope of the original text.
Redundancy
One potential issue with the extractive approach is that it may select and include the same information multiple times in the summary, particularly if the source text includes repetitive information.
This can result in a summary that is not only shorter than the original text, but also lacks clarity and conciseness. In order to avoid this issue, it may be necessary to implement techniques such as sentence clustering or topic modeling to identify and eliminate redundant information from the summary. A human editor may need to review the summary to ensure that it accurately captures the key ideas of the source text in a comprehensive and cohesive manner.
Lack of abstraction
Extractive summarization does not generate new sentences or paraphrase the information, which could be a disadvantage for certain applications where a more human-like summary is desired.
One potential disadvantage of extractive summarization is its lack of abstraction. Unlike other summarization techniques, extractive summarization does not have the ability to generate new sentences or paraphrase the information it is summarizing.
This can be a problem in certain applications where a more human-like summary is desired, as the extractive summary may not capture the nuance and complexity of the original text. Therefore, it may be necessary to use alternative summarization techniques in these cases, or to supplement the extractive summary with additional analysis and interpretation.
Missing important information
One issue that can arise when using extractive summarization is that important information may be spread across different parts of the text and not encapsulated in a single sentence. This can lead to a less comprehensive summary that misses key details.
For example, if a long article discusses a complex topic, a summary that only includes the most frequently occurring words or phrases may not capture the nuances and complexities of the original text. As a result, it is important to carefully review summaries generated through extractive techniques to ensure that all pertinent information is included.
Understanding these limitations is essential as it helps in choosing the right summarization technique for a given task and provides directions for future research and improvements in extractive summarization techniques.
9.1 Extractive Summarization
In today's era of information overload, the ability to automatically summarize lengthy text has become an essential tool. The field of Natural Language Processing (NLP) has developed a subfield specifically focused on text summarization, which involves creating a concise summary while retaining the most important points of the original text. Summaries can significantly enhance comprehension by reducing the amount of time needed to absorb information, making them useful in a variety of applications, such as news article summaries, customer review summaries, and summaries of scientific research papers.
There are two primary approaches to text summarization: extractive and abstractive. Extractive summarization involves selecting and extracting sentences that are deemed the most important in the original text, while abstractive summarization involves generating new sentences to summarize the content. In this chapter, we will delve into the details of both approaches, discussing their respective strengths and weaknesses, and providing examples of implementation for each.
Let's begin by exploring extractive summarization.
Extractive summarization is a technique that involves identifying the most important sections of a text and generating a summary by copying them verbatim. The key objective of this approach is to identify sentences or passages that provide a good representation of the overall content of the text.
This is a simpler approach as it does not require the model to generate new text, but simply to identify the most significant portions of the existing text. To achieve this goal, a variety of techniques can be used such as natural language processing, machine learning, and deep learning.
These techniques can help in the identification of relevant sentences and passages that can be used to generate a summary of the text. In addition, extractive summarization can be used in a variety of applications such as document summarization, news article summarization, and automatic text summarization. Extractive summarization is a useful technique for generating summaries of long texts in an efficient and accurate manner.
9.1.1 Principle of Extractive Summarization
The extractive summarization technique is a powerful tool used to summarize a source document. It works on the principle of selection and ranking, where the algorithm selects the most important sentences from the document and concatenates them to form a summary.
The ranking of the sentences can be determined by various features such as the frequency of words, sentence length, the presence of named entities, or the location of a sentence in the document. The higher the rank of a sentence, the more important it is in the summary.
This technique is particularly useful when dealing with large volumes of text, as it can quickly generate a summary that captures the key ideas of the source document. Extractive summarization can be used to highlight the most important points of a document, making it easier for readers to identify the key takeaways.
The extractive summarization technique is an effective way to condense large amounts of information into a concise and easily digestible format.
9.1.2 Techniques for Extractive Summarization
Several methods can be used for extractive summarization. Here are a few common ones:
Frequency-based Approach
This is the simplest approach to summarize a document. It involves calculating the frequency of each word in the document, excluding stop words. Stop words are commonly used words such as "the", "a", "an", etc. After the frequency of each word is calculated, the sentences in the document are ranked based on the sum of the frequencies of the words in each sentence.
This approach is simple and easy to implement, but may not always provide the most accurate summary of the document.
There are other approaches to summarization that take into account the context and meaning of words, which may result in a more precise summary. However, the frequency-based approach is still widely used due to its simplicity and efficiency.
Graph-based Approach
The graph-based approach is a popular method for text summarization. In this method, a graph is constructed where each sentence in the document is a node. The edges between the nodes are weighted based on the similarity between the sentences. By using a ranking algorithm, such as PageRank, to score the sentences, the most important sentences can be identified and used to create a summary of the document.
This approach is particularly useful for dealing with large amounts of text, as it can quickly identify the most important sentences without requiring manual intervention. Furthermore, it is highly flexible and can be adapted to different types of documents and summarization goals. Overall, the graph-based approach is a powerful tool for summarizing text and extracting the most important information from a document.
Feature-based Approach
The Feature-based Approach to summarization involves the selection of a determined set of features that represent the importance of a sentence. These features may include sentence length, term frequency, named entities, among others. The purpose of this method is to train a machine learning model which will then be used to rank the sentences based on the aforementioned features.
By doing this, the most important sentences are selected and used to create a summary of the text. This method is particularly useful for large texts where manual summarization would be time-consuming and inefficient. It is important to note, however, that the selection of the features is crucial to the success of this approach.
Therefore, extensive research and analysis must be conducted beforehand to ensure that the most relevant features are selected and utilized in the training of the machine learning model.
Let's now look at a simple code example of extractive summarization using the Python library gensim.
Example:
from gensim.summarization import summarize
# Assume text contains the original document text
summary = summarize(text)
print(summary)In this example, we use the summarize function from gensim, which implements an extractive summarization method based on the TextRank algorithm. TextRank is a graph-based ranking model for text processing which can be used for extractive summarization. It ranks the sentences in the document based on their similarity to other sentences. The top-ranked sentences are then included in the summary.
Remember that while extractive summarization methods are effective and relatively simple to implement, they have some limitations. For instance, they can't handle nuances and may miss important information that is spread across multiple sentences.
Also, they can't generate summaries in a different wording or style than the source text, as they merely copy selected sentences from the original text. These limitations lead us to the other main approach to text summarization, called abstractive summarization, which we'll cover in the next section.
9.1.3 Practical Example
To further enrich the understanding of extractive summarization, let's take a deeper dive into a practical Python example that uses the networkx library and sklearn to implement an extractive summarization technique.
This example will create a matrix of sentence similarities, then use this matrix to create a graph with networkx. The PageRank algorithm, which is a part of the networkx library, will then be used to rank the sentences based on their scores.
For this example, we'll use the NLTK library to download the stop words and the punkt package, which is a sentence tokenizer.
import nltk
nltk.download('punkt') # for sentence tokenization
nltk.download('stopwords') # for removing stop wordsNext, we'll install and import all the necessary libraries:
!pip install numpy networkx sklearn
import numpy as np
import networkx as nx
from nltk.corpus import stopwords
from nltk.tokenize import sent_tokenize
from sklearn.metrics.pairwise import cosine_similarity
from sklearn.feature_extraction.text import CountVectorizerNow, let's define a function that calculates the similarity matrix:
def build_similarity_matrix(sentences, stopwords=None):
# Create a CountVectorizer object for transforming the text data to a matrix of token counts
count_vectorizer = CountVectorizer(stop_words=stopwords)
# Transform the sentences to a matrix of token counts
count_matrix = count_vectorizer.fit_transform(sentences)
# Compute cosine similarity between the sentences
sentence_similarity = cosine_similarity(count_matrix, count_matrix)
return sentence_similarityWith this function, we can create a graph and calculate the scores of the sentences:
def generate_summary(text, top_n=5):
stop_words = stopwords.words('english')
summarize_text = []
# Step 1 - Read text and split it
sentences = sent_tokenize(text)
# Step 2 - Generate similarity matrix across sentences
sentence_similarity_matrix = build_similarity_matrix(sentences, stop_words)
# Step 3 - Rank sentences in the similarity matrix
sentence_similarity_graph = nx.from_numpy_array(sentence_similarity_matrix)
scores = nx.pagerank(sentence_similarity_graph)
# Step 4 - Sort the rank and pick the top sentences
ranked_sentence = sorted(((scores[i], s) for i, s in enumerate(sentences)), reverse=True)
for i in range(top_n):
summarize_text.append(" ".join(ranked_sentence[i][1]))
# Step 5 - Output the summarized text
return ". ".join(summarize_text)
# Assume we have some long text data in the variable 'text'
print("Summary: \n", generate_summary(text, 2))This code will output the top 2 sentences from the text as the summary. This is a simple yet effective way to perform extractive summarization by ranking sentences according to their similarity with other sentences in the document.
Keep in mind that while this method works well for many tasks, it can sometimes oversimplify or miss important nuances in the text. This is due to the fact that it is extracting sentences without considering how they might be rephrased or summarized more succinctly. For more complex tasks, it might be necessary to use more sophisticated methods like abstractive summarization, which we'll explore in the next topic.
9.1.4 Evaluation Metrics for Extractive Summarization
ROUGE Score
ROUGE is an acronym for Recall-Oriented Understudy for Gisting Evaluation. It is a set of metrics that is used to evaluate the quality of automatic summarization and machine translation. The primary way that this is done is by comparing an automatically generated summary or translation with a set of reference summaries, which are typically created by humans.
ROUGE is an important tool for researchers who are working on improving the accuracy and effectiveness of automatic summarization and machine translation systems. By using ROUGE, researchers can get a better understanding of how well their systems are performing and what areas need improvement.
In addition to being a useful tool for researchers, ROUGE is also used in industry. Many companies that produce automatic summarization and machine translation software use ROUGE to evaluate their products and make sure that they are meeting the needs of their customers.
Overall, ROUGE is an essential part of the field of natural language processing and is likely to continue playing an important role in the development of automatic summarization and machine translation technology in the years to come.
ROUGE-N, ROUGE-L, and ROUGE-S are the most commonly used ROUGE metrics:
- ROUGE-N: Measures the number of matching n-grams. This is to say, ROUGE-1 refers to the overlap of unigrams between the system and reference summaries. ROUGE-2 refers to the overlap of bigrams.
- ROUGE-L: Measures the longest matching sequence of words using LCS (Longest Common Subsequence). It does not require consecutive matches but in-sequence matches.
- ROUGE-S: Measures the skip-bigram co-occurrence statistics. Skip-bigram is any pair of words in their sentence order.
Example:
!pip install rouge
from rouge import Rouge
def evaluate_summary(summary, reference):
rouge = Rouge()
scores = rouge.get_scores(summary, reference)
return scores
# Assume we have a generated summary 'gen_summary' and a reference summary 'ref_summary'
print(evaluate_summary(gen_summary, ref_summary))This code calculates and prints out the ROUGE-1, ROUGE-2, and ROUGE-L scores for the generated summary compared to the reference.
BLEU Score
The Bilingual Evaluation Understudy (BLEU) score is a metric that measures the quality of a summary. Although it is commonly used in machine translation, it can also be used for text summarization tasks. BLEU analyzes the overlap between the predicted and actual words and phrases, up to a specified length (such as n-grams).
It is worth noting, however, that the BLEU score function in NLTK is more effective for tasks where multiple reference summaries are possible, such as machine translation. In tasks like text summarization, where there is usually only one reference summary, BLEU may not be the best choice. Despite this, BLEU remains a widely used metric for evaluating the quality of text summaries and is an important tool for researchers and practitioners alike.
These evaluation metrics provide a quantitative way to measure the quality of a text summary. However, they still have limitations and do not always align perfectly with human judgement, as they mostly rely on word matching and can overlook semantic understanding.
9.1.5 Limitations and Challenges of Extractive Summarization
Extractive summarization, while useful, does have certain limitations:
Lack of coherence
One of the potential issues with using extractive techniques for summarization is that sentences are chosen based solely on their individual scores. While this approach can be effective at capturing important information, it may not fully consider the overall flow or structure of the original text. This can result in summaries that lack coherence and fail to convey a clear narrative.
It is important to note that extractive techniques often ignore the broader context of the text. As a result, the summary may miss key themes or ideas that are only apparent when looking at the text as a whole. Additionally, these techniques may not be able to capture the nuance and complexity of certain passages, leading to summaries that oversimplify or distort the original meaning.
To address these issues, some researchers have proposed using more advanced techniques for summarization, such as abstractive techniques that generate summaries by synthesizing new text rather than simply selecting and reordering existing sentences. While these techniques are still in the early stages of development, they offer the potential for more coherent and comprehensive summaries that capture the full scope of the original text.
Redundancy
One potential issue with the extractive approach is that it may select and include the same information multiple times in the summary, particularly if the source text includes repetitive information.
This can result in a summary that is not only shorter than the original text, but also lacks clarity and conciseness. In order to avoid this issue, it may be necessary to implement techniques such as sentence clustering or topic modeling to identify and eliminate redundant information from the summary. A human editor may need to review the summary to ensure that it accurately captures the key ideas of the source text in a comprehensive and cohesive manner.
Lack of abstraction
Extractive summarization does not generate new sentences or paraphrase the information, which could be a disadvantage for certain applications where a more human-like summary is desired.
One potential disadvantage of extractive summarization is its lack of abstraction. Unlike other summarization techniques, extractive summarization does not have the ability to generate new sentences or paraphrase the information it is summarizing.
This can be a problem in certain applications where a more human-like summary is desired, as the extractive summary may not capture the nuance and complexity of the original text. Therefore, it may be necessary to use alternative summarization techniques in these cases, or to supplement the extractive summary with additional analysis and interpretation.
Missing important information
One issue that can arise when using extractive summarization is that important information may be spread across different parts of the text and not encapsulated in a single sentence. This can lead to a less comprehensive summary that misses key details.
For example, if a long article discusses a complex topic, a summary that only includes the most frequently occurring words or phrases may not capture the nuances and complexities of the original text. As a result, it is important to carefully review summaries generated through extractive techniques to ensure that all pertinent information is included.
Understanding these limitations is essential as it helps in choosing the right summarization technique for a given task and provides directions for future research and improvements in extractive summarization techniques.
9.1 Extractive Summarization
In today's era of information overload, the ability to automatically summarize lengthy text has become an essential tool. The field of Natural Language Processing (NLP) has developed a subfield specifically focused on text summarization, which involves creating a concise summary while retaining the most important points of the original text. Summaries can significantly enhance comprehension by reducing the amount of time needed to absorb information, making them useful in a variety of applications, such as news article summaries, customer review summaries, and summaries of scientific research papers.
There are two primary approaches to text summarization: extractive and abstractive. Extractive summarization involves selecting and extracting sentences that are deemed the most important in the original text, while abstractive summarization involves generating new sentences to summarize the content. In this chapter, we will delve into the details of both approaches, discussing their respective strengths and weaknesses, and providing examples of implementation for each.
Let's begin by exploring extractive summarization.
Extractive summarization is a technique that involves identifying the most important sections of a text and generating a summary by copying them verbatim. The key objective of this approach is to identify sentences or passages that provide a good representation of the overall content of the text.
This is a simpler approach as it does not require the model to generate new text, but simply to identify the most significant portions of the existing text. To achieve this goal, a variety of techniques can be used such as natural language processing, machine learning, and deep learning.
These techniques can help in the identification of relevant sentences and passages that can be used to generate a summary of the text. In addition, extractive summarization can be used in a variety of applications such as document summarization, news article summarization, and automatic text summarization. Extractive summarization is a useful technique for generating summaries of long texts in an efficient and accurate manner.
9.1.1 Principle of Extractive Summarization
The extractive summarization technique is a powerful tool used to summarize a source document. It works on the principle of selection and ranking, where the algorithm selects the most important sentences from the document and concatenates them to form a summary.
The ranking of the sentences can be determined by various features such as the frequency of words, sentence length, the presence of named entities, or the location of a sentence in the document. The higher the rank of a sentence, the more important it is in the summary.
This technique is particularly useful when dealing with large volumes of text, as it can quickly generate a summary that captures the key ideas of the source document. Extractive summarization can be used to highlight the most important points of a document, making it easier for readers to identify the key takeaways.
The extractive summarization technique is an effective way to condense large amounts of information into a concise and easily digestible format.
9.1.2 Techniques for Extractive Summarization
Several methods can be used for extractive summarization. Here are a few common ones:
Frequency-based Approach
This is the simplest approach to summarize a document. It involves calculating the frequency of each word in the document, excluding stop words. Stop words are commonly used words such as "the", "a", "an", etc. After the frequency of each word is calculated, the sentences in the document are ranked based on the sum of the frequencies of the words in each sentence.
This approach is simple and easy to implement, but may not always provide the most accurate summary of the document.
There are other approaches to summarization that take into account the context and meaning of words, which may result in a more precise summary. However, the frequency-based approach is still widely used due to its simplicity and efficiency.
Graph-based Approach
The graph-based approach is a popular method for text summarization. In this method, a graph is constructed where each sentence in the document is a node. The edges between the nodes are weighted based on the similarity between the sentences. By using a ranking algorithm, such as PageRank, to score the sentences, the most important sentences can be identified and used to create a summary of the document.
This approach is particularly useful for dealing with large amounts of text, as it can quickly identify the most important sentences without requiring manual intervention. Furthermore, it is highly flexible and can be adapted to different types of documents and summarization goals. Overall, the graph-based approach is a powerful tool for summarizing text and extracting the most important information from a document.
Feature-based Approach
The Feature-based Approach to summarization involves the selection of a determined set of features that represent the importance of a sentence. These features may include sentence length, term frequency, named entities, among others. The purpose of this method is to train a machine learning model which will then be used to rank the sentences based on the aforementioned features.
By doing this, the most important sentences are selected and used to create a summary of the text. This method is particularly useful for large texts where manual summarization would be time-consuming and inefficient. It is important to note, however, that the selection of the features is crucial to the success of this approach.
Therefore, extensive research and analysis must be conducted beforehand to ensure that the most relevant features are selected and utilized in the training of the machine learning model.
Let's now look at a simple code example of extractive summarization using the Python library gensim.
Example:
from gensim.summarization import summarize
# Assume text contains the original document text
summary = summarize(text)
print(summary)In this example, we use the summarize function from gensim, which implements an extractive summarization method based on the TextRank algorithm. TextRank is a graph-based ranking model for text processing which can be used for extractive summarization. It ranks the sentences in the document based on their similarity to other sentences. The top-ranked sentences are then included in the summary.
Remember that while extractive summarization methods are effective and relatively simple to implement, they have some limitations. For instance, they can't handle nuances and may miss important information that is spread across multiple sentences.
Also, they can't generate summaries in a different wording or style than the source text, as they merely copy selected sentences from the original text. These limitations lead us to the other main approach to text summarization, called abstractive summarization, which we'll cover in the next section.
9.1.3 Practical Example
To further enrich the understanding of extractive summarization, let's take a deeper dive into a practical Python example that uses the networkx library and sklearn to implement an extractive summarization technique.
This example will create a matrix of sentence similarities, then use this matrix to create a graph with networkx. The PageRank algorithm, which is a part of the networkx library, will then be used to rank the sentences based on their scores.
For this example, we'll use the NLTK library to download the stop words and the punkt package, which is a sentence tokenizer.
import nltk
nltk.download('punkt') # for sentence tokenization
nltk.download('stopwords') # for removing stop wordsNext, we'll install and import all the necessary libraries:
!pip install numpy networkx sklearn
import numpy as np
import networkx as nx
from nltk.corpus import stopwords
from nltk.tokenize import sent_tokenize
from sklearn.metrics.pairwise import cosine_similarity
from sklearn.feature_extraction.text import CountVectorizerNow, let's define a function that calculates the similarity matrix:
def build_similarity_matrix(sentences, stopwords=None):
# Create a CountVectorizer object for transforming the text data to a matrix of token counts
count_vectorizer = CountVectorizer(stop_words=stopwords)
# Transform the sentences to a matrix of token counts
count_matrix = count_vectorizer.fit_transform(sentences)
# Compute cosine similarity between the sentences
sentence_similarity = cosine_similarity(count_matrix, count_matrix)
return sentence_similarityWith this function, we can create a graph and calculate the scores of the sentences:
def generate_summary(text, top_n=5):
stop_words = stopwords.words('english')
summarize_text = []
# Step 1 - Read text and split it
sentences = sent_tokenize(text)
# Step 2 - Generate similarity matrix across sentences
sentence_similarity_matrix = build_similarity_matrix(sentences, stop_words)
# Step 3 - Rank sentences in the similarity matrix
sentence_similarity_graph = nx.from_numpy_array(sentence_similarity_matrix)
scores = nx.pagerank(sentence_similarity_graph)
# Step 4 - Sort the rank and pick the top sentences
ranked_sentence = sorted(((scores[i], s) for i, s in enumerate(sentences)), reverse=True)
for i in range(top_n):
summarize_text.append(" ".join(ranked_sentence[i][1]))
# Step 5 - Output the summarized text
return ". ".join(summarize_text)
# Assume we have some long text data in the variable 'text'
print("Summary: \n", generate_summary(text, 2))This code will output the top 2 sentences from the text as the summary. This is a simple yet effective way to perform extractive summarization by ranking sentences according to their similarity with other sentences in the document.
Keep in mind that while this method works well for many tasks, it can sometimes oversimplify or miss important nuances in the text. This is due to the fact that it is extracting sentences without considering how they might be rephrased or summarized more succinctly. For more complex tasks, it might be necessary to use more sophisticated methods like abstractive summarization, which we'll explore in the next topic.
9.1.4 Evaluation Metrics for Extractive Summarization
ROUGE Score
ROUGE is an acronym for Recall-Oriented Understudy for Gisting Evaluation. It is a set of metrics that is used to evaluate the quality of automatic summarization and machine translation. The primary way that this is done is by comparing an automatically generated summary or translation with a set of reference summaries, which are typically created by humans.
ROUGE is an important tool for researchers who are working on improving the accuracy and effectiveness of automatic summarization and machine translation systems. By using ROUGE, researchers can get a better understanding of how well their systems are performing and what areas need improvement.
In addition to being a useful tool for researchers, ROUGE is also used in industry. Many companies that produce automatic summarization and machine translation software use ROUGE to evaluate their products and make sure that they are meeting the needs of their customers.
Overall, ROUGE is an essential part of the field of natural language processing and is likely to continue playing an important role in the development of automatic summarization and machine translation technology in the years to come.
ROUGE-N, ROUGE-L, and ROUGE-S are the most commonly used ROUGE metrics:
- ROUGE-N: Measures the number of matching n-grams. This is to say, ROUGE-1 refers to the overlap of unigrams between the system and reference summaries. ROUGE-2 refers to the overlap of bigrams.
- ROUGE-L: Measures the longest matching sequence of words using LCS (Longest Common Subsequence). It does not require consecutive matches but in-sequence matches.
- ROUGE-S: Measures the skip-bigram co-occurrence statistics. Skip-bigram is any pair of words in their sentence order.
Example:
!pip install rouge
from rouge import Rouge
def evaluate_summary(summary, reference):
rouge = Rouge()
scores = rouge.get_scores(summary, reference)
return scores
# Assume we have a generated summary 'gen_summary' and a reference summary 'ref_summary'
print(evaluate_summary(gen_summary, ref_summary))This code calculates and prints out the ROUGE-1, ROUGE-2, and ROUGE-L scores for the generated summary compared to the reference.
BLEU Score
The Bilingual Evaluation Understudy (BLEU) score is a metric that measures the quality of a summary. Although it is commonly used in machine translation, it can also be used for text summarization tasks. BLEU analyzes the overlap between the predicted and actual words and phrases, up to a specified length (such as n-grams).
It is worth noting, however, that the BLEU score function in NLTK is more effective for tasks where multiple reference summaries are possible, such as machine translation. In tasks like text summarization, where there is usually only one reference summary, BLEU may not be the best choice. Despite this, BLEU remains a widely used metric for evaluating the quality of text summaries and is an important tool for researchers and practitioners alike.
These evaluation metrics provide a quantitative way to measure the quality of a text summary. However, they still have limitations and do not always align perfectly with human judgement, as they mostly rely on word matching and can overlook semantic understanding.
9.1.5 Limitations and Challenges of Extractive Summarization
Extractive summarization, while useful, does have certain limitations:
Lack of coherence
One of the potential issues with using extractive techniques for summarization is that sentences are chosen based solely on their individual scores. While this approach can be effective at capturing important information, it may not fully consider the overall flow or structure of the original text. This can result in summaries that lack coherence and fail to convey a clear narrative.
It is important to note that extractive techniques often ignore the broader context of the text. As a result, the summary may miss key themes or ideas that are only apparent when looking at the text as a whole. Additionally, these techniques may not be able to capture the nuance and complexity of certain passages, leading to summaries that oversimplify or distort the original meaning.
To address these issues, some researchers have proposed using more advanced techniques for summarization, such as abstractive techniques that generate summaries by synthesizing new text rather than simply selecting and reordering existing sentences. While these techniques are still in the early stages of development, they offer the potential for more coherent and comprehensive summaries that capture the full scope of the original text.
Redundancy
One potential issue with the extractive approach is that it may select and include the same information multiple times in the summary, particularly if the source text includes repetitive information.
This can result in a summary that is not only shorter than the original text, but also lacks clarity and conciseness. In order to avoid this issue, it may be necessary to implement techniques such as sentence clustering or topic modeling to identify and eliminate redundant information from the summary. A human editor may need to review the summary to ensure that it accurately captures the key ideas of the source text in a comprehensive and cohesive manner.
Lack of abstraction
Extractive summarization does not generate new sentences or paraphrase the information, which could be a disadvantage for certain applications where a more human-like summary is desired.
One potential disadvantage of extractive summarization is its lack of abstraction. Unlike other summarization techniques, extractive summarization does not have the ability to generate new sentences or paraphrase the information it is summarizing.
This can be a problem in certain applications where a more human-like summary is desired, as the extractive summary may not capture the nuance and complexity of the original text. Therefore, it may be necessary to use alternative summarization techniques in these cases, or to supplement the extractive summary with additional analysis and interpretation.
Missing important information
One issue that can arise when using extractive summarization is that important information may be spread across different parts of the text and not encapsulated in a single sentence. This can lead to a less comprehensive summary that misses key details.
For example, if a long article discusses a complex topic, a summary that only includes the most frequently occurring words or phrases may not capture the nuances and complexities of the original text. As a result, it is important to carefully review summaries generated through extractive techniques to ensure that all pertinent information is included.
Understanding these limitations is essential as it helps in choosing the right summarization technique for a given task and provides directions for future research and improvements in extractive summarization techniques.