Skip to content

hiver-py/TextSummarization

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

18 Commits
 
 
 
 

Repository files navigation

Abstractive Text Summarization of News Using Pre-Trained Transformers

Motivation:

With a more global society today than ever, staying informed on current events is a time-consuming task. Fortunately, with advances in Natural Language Processing, tools have developed that can analyze text-based data and perform various classification and generative tasks.

These tasks has traditionally been very hard to tackle as language can be ambiguous, abstract, but is also highly context dependent in complex ways as there may exists both short and long-term contextual information.

How:

In this project the goal is to fine-tune two pre-trained networks from the HuggingFace API using news articles. These networks will then be used to generate a summary of news article, that can be compared to a highlight of the article which is written by the articles author. The two networks that are utilized for the abstractive text summarization is the T5-Small and BART.

Data:

The data set used in this project is the CNN/Daily Mail summarization dataset. It consists of 311 971 unique news articles that is written by journalists at CNN, and the Daily Mail. The data set is split into 287 113 articles in the training set, 13 368 in the validation set and 11 490 in the test set. Each observation in the data set consists of three columns: an ID, the news article, and a highlight written by the articles author which summarizes the main points of the article.

Results:

A short excerpt of the results for both models are shown, only displaying the Precision, Recall and F-measure for the Rouge-Lsum metric.

T5-Small:

Highlight Summary
A lawyer for Dr. Anthony Moschetto says the charges against him are baseless. Moschetto, 54, was arrested for selling drugs and weapons, prosecutors say. Authorities allege Moschetto hired accomplices to burn down the practice of former associate. Dr. Anthony Moschetto faces criminal solicitation, conspiracy, burglary, arson, criminal prescription sale and weapons charges. He is accused of selling heroin and oxycodone pills from Moschetto. He is accused of trying to take out a rival doctor on Long Island.
"No challenge poses more of a public threat than climate change," the President says. He credits the Clean Air Act with making Americans "a lot" healthier. Obama took part in a roundtable discussion this week on climate change. He says the issue is a public health issue that affects all of us. Obama says the Clean Air Act and its amendments have reduced early deaths. Obama says the issue is a public health issue that affects all of us.
Raul Reyes: In seeking Latino vote, Marco Rubio his own worst enemy on two key issues: immigration reform, Cuba relations. He says on health care, climate change and other issues, he breaks from Latinos' positions. Polls show they don't favor him . Marco Rubio is seeking the Republican presidential nomination. He says Rubio has embraced a typical conservative approach to immigration. Rubio has embraced a typical conservative approach to immigration. He says President Obama's executive action on immigration should be ended.
Rouge-Lsum Low Mid High
Precision 0.40 0.41 0.41
Recall 0.27 0.27 0.28
F-measure 0.32 0.32 0.32

BART:

Highlight Summary
A lawyer for Dr. Anthony Moschetto says the charges against him are baseless. Moschetto, 54, was arrested for selling drugs and weapons, prosecutors say. Authorities allege Moschetto hired accomplices to burn down the practice of former associate. "None of anything in this case has any evidentiary value," Randy Zelin says. Dr. Anthony Moschetto, 54, pleaded not guilty to all charges Wednesday. Two other men -- identified as James Chmela, 43, and James Kalamaras, 41 -- were named as accomplices.
"No challenge poses more of a public threat than climate change," the President says. He credits the Clean Air Act with making Americans "a lot" healthier. President Barack Obama took part in a roundtable discussion this week on climate change. He said the air was so bad that it prevented him from running outside. Obama credits the Clean Air Act with making Americans "a lot" healthier.
Raul Reyes: In seeking Latino vote, Marco Rubio his own worst enemy on two key issues: immigration reform, Cuba relations. He says on health care, climate change and other issues, he breaks from Latinos' positions. Polls show they don't favor him. Ruben Navarrette: Marco Rubio is running on optimistic message that he embodies the promise of the American Dream. He says Rubio has been his own worst enemy on what could have been his two signature issues: immigration reform and Cuba relations. Navarrette says Rubio's retreat on immigration means he has missed an opportunity to set himself apart from most of
Rouge-Lsum Low Mid High
Precision 0.38 0.38 0.39
Recall 0.39 0.39 0.40
F-measure 0.38 0.38 0.38

Conclusion:

It is found that news paper articles can be summarized well by using the two different pre-trained transformer networks: T5-Small and BART.

By testing the different models and fine-tune them on the CNN/daily mail data both are able to produce summaries of the news article of high quality, that captures the context of the articles, and match the highlights written by the news articles author well.

One benefit of using T5-Small is that it is significantly faster in training compared to BART, without having any noticeable difference in the quality of the generated summaries.

About

News Article Text Summarization

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published