Abstractive Summarizer for YouTube Videos

Sulochana Devi(B) , Rahul Nadar , Tejas Nichat , and Alfredprem Lucas

Department of Information Technology, Xavier Institute of Engineering, Mumbai, India

[email protected]

Abstract. The paper goal is to design a user interface where the user can get
the summary of the requested YouTube video using Natural Language Process-
ing (NLP) and Machine Learning. Enormous number of videos are uploaded to
YouTube daily. It has become difficult to find the relevant content that we are
looking for, sometimes it may take longer than expected, and our efforts be- come
futile if we are unable to extract meaningful information from it. This is where
our project becomes handy. It will summarize the video and display the summary
of it. Abstractive summarization model extracts YouTube video transcripts and
generates a summarized version. It saves our time by shortening the content to
read while maintaining the important content of the actual document. The imple-
mentation process is still ongoing. However, the entire structure and studies are
presented.

Keywords: Natural Language Processing · Machine Learning · Abstractive

summarization

1 Introduction
In this era of digitalization, the importance of technology is the most fundamental basis of
improvement in today’s society. With a massive number of users accessing the internet
throughout the day and growing at an alarming rate for a variety of rea- sons, it has
become really frustrating and time-consuming to search for relevant in- formation on
the internet. The ease with which content producers may reach a huge audience via
YouTube is excessive. As a result, there is a wide selection of content available. Around
3.7 million videos are uploaded on YouTube every day. So it has become a task to find
the perfect video with the required information contained within it. The disadvantage
of this is that it encourages the creation of endless click bait videos, which waste the
user’s time. For a given search query, there are so many results in which there are some
click bait videos and some that contain very little required or relevant information for
the user, which obviously they don’t know about and watch the whole video till the last
minute and end up with very little to zero information.
By summarizing, you can learn how to identify the key concepts in a text or video,
how to filter out unimportant material, and how to connect the main ideas in an en-
gaging way. This model will also be able to convert the summarized text’s language
from some given options made available in the extension which will be beneficial. A
short glance at the synopsis will tell the user whether the video is worth their time

© The Author(s) 2023

S. Tamane et al. (Eds.): ICAMIDA 2022, ACSR 105, pp. 431–438, 2023.
https://doi.org/10.2991/978-94-6463-136-4_36
432 S. Devi et al.

and whether it covers the themes they are looking for. It can also be used to briefly
summarize important information from the video. This can be achieved by two types of
summarization extractive and abstractive.
Extractive and abstractive are the two most common ways to summarize text. Extrac-
tive approaches depend only on extracting significant phrases from a document and later
combining those significant phrases into a coherent summary. Abstractive method anal-
yses the document of text, which generates a coherent summary with new words and
phrases just like how humans create a syntactic and well-formed representation of the
documents in their brain. This paper focuses on video summarization using abstractive
summarization method while some papers are based on extractive summarization.

2 Literature Survey

From [1], to create a summary using the provided YouTube video, the author suggest-
ed a model. If the transcript of the video is not available then it will convert the audio of
the video into transcript and then apply an Abstractive method for text summarization
in the transcript and obtain the summary of the given YouTube link video. By giving
the user only the knowledge, they need to solve their problems, their paper helps them
avoid having to watch those lengthy videos and makes better use of their time.
In [2], the project had been created using an embedding layer that converted words
into some vector representation so that the model could generalize the words to do any
kind of prediction or summary generation. The encoder records the context of the input
sequence as a hidden state vector. Finding important terms in the corpus is done using
the TFIDF technique
An Automatic NLP based LSA summarization algorithm has been performed on the
subtitle of the videos to generate the summary. In [3] research paper they have used the
LSA Natural Language Processing algorithm, which requires less processing power and
no training data required to train the algorithm.
They have used Latent Semantic Analysis (LSA) technique to extract the features of
the sentences that cannot be directly mentioned. This summarization technique works
in such a way that the top most ranked subtitles are taken into consideration for the final
video. They found the average duration of each subtitle by dividing the Total duration
of the video with the Number of subtitles. This summarization technique works in
such a way that the top most ranked subtitles are taken into consideration for the final
summarization.
In this paper [4], authors surveyed on Abstractive Transcript Summarization of
YouTube Videos. The difference between the Extractive and Abstractive summariza-
tion methods are stated they researched on many Abstractive Summarization methods
and models. To make the summary effective in the least amount of time by choosing
the best summarization methods and models. They also mentioned the pros and cons of
each method.
In [5], the paper “YouTube Transcript Summarizer” in which they have used Hugging
Face Transformer to perform abstractive summarization on the transcripts and have used
RESTAPI in the backend. The YouTube transcript is used as an input parameter for the
model, which outputs a condensed summary.
 Abstractive Summarizer for YouTube Videos 433

From [6], It is safe to say that video summarization and skimming have evolved into
essential tools for any real world video management system. This article offers a guide
to the current abstraction work for typical videos. The article also discusses the authors’
current work on movie skimming, which makes use of rhythmic analysis of audio and
visual content as well as some cinematic conventions.
They examine a number of topics that the AMIS project has looked into in this
paper. By summarizing the original video, a system for comprehending a foreign video
has been created. [7] They wanted to extract the key idea from a video and translate it
into English. Several subsystems have been implemented to create a larger system, with
each of them posing a significant scientific challenge.
From [8] they have provided a technical background for document summarizing in
this study. This essay has also covered a number of difficulties with the current summa-
rizing techniques. They have learned from these conversations that many methodologies
face numerous difficulties. For instance, the clustering-based method needs accurate
information about the number of clusters it is building, and the MMR uses various
strategies for the coverage and non-redundancy elements of the summary. Pre-processing
and textual unit assessment are the fundamental processing, components, and resources
needed to complete these steps.
The extractive strategy is the main emphasis of this study work. As a result, extrac-
tive summarization’s literature is far superior to abstractive summarizations. This study
looked at several strategies and techniques. [9] They come to the conclusion that com-
bining two ways is more likely to achieve ideal outcomes and improve the quality of the
summarization than utilizing only one approach.
In this paper [10] the author suggests a transcript summarizer that uses NLP methods
to extract and summarize material from video files. The video transcripts are split into
two sections: It will first split the video into many frame-based audio chunks, then the
audio chunks will be further split into tokens, and finally each token will be extracted
to text. The summarizing model is then provided with the resulting text. Extractive text
summarization is a method for summarizing where it extracts summary from topranking
sentences that make sense. Videos of various sizes can be used to test the effectiveness
of summarization.
The paper goal is to design a chrome extension for YouTube Transcript Summarizer
which has the following features over other researches:

• You can use our research to block out background noise or keep track of your to-do
list to obtain a clear picture of all you need to finish.
• Because it has access to its own set of APIs, our extension is able to per- form tasks
that a website cannot.
• Our extension provides many extra capabilities, such the ability for users to share via
email, obtain a summary transcript, translate, or use the speak feature.

3 Proposed Work
The proposed work of YouTube Transcript Summarizer. In this model a HTTP request to
the back end is created from YouTube transcript and perform transcript summarization
434 S. Devi et al.

Fig. 1. Project Stages

and return it as HTTP response then the summarized transcript is shown on the Extension.
Below is a detailed explanation of the features and method- ology of the summary utilized
for this purpose (Fig. 1).
The processes taken to complete the transcript information are as follows: (Fig. 2).

1. Open a YouTube video and click on summarize in Chrome Extension to create a

HTTP request to the back end.
2. Request Transcript for the given YouTube ID.
3. Perform transcript Summarization and return it as HTTP response.
4. The summarized transcript is shown on the extension.

3.1 Methodology
Our transcript summarization method is divided into following parts:

3.1.1 Text Cleaning

We’ll make use of the Spacy library. Spacy is made to be built for systems that ex-
tract information and understand natural language. The Spacy library has the ability
 Abstractive Summarizer for YouTube Videos 435

Fig. 2. Process of Summarization

to segment text into words, punctuation, and assign word roots. Additionally, Spacy is
capable of serialization and text classification. Using spacy any text may be turned into
a processed Doc object using this technique, and properties can be inferred.

3.1.2 Sentence Tokenization

The sent tokenize function makes use of an instance of the Punkt Sentence Tokenizer
from the nltk tokenize module, which has already been trained and is quite knowledge-
able about where to mark a sentence’s beginning and conclusion with respect to charac-
ters and punctuation. The advantages of word tokenization using NLTK include White
Space Tokenization, Dictionary Based Tokenization, Rule-Based Tokenization, Regu-
lar Expression Tokenization, Penn Treebank Tokenization, Spacy Tokenization, Moses
Tokenization, and Subword Tokenization. The text normalization procedure includes all
kinds of word tokenization. The accuracy of the language understanding algorithms is
increased by stemming and lemmatizing the text to normalize it.

3.1.3 Word Tokenization

A sequence of strings is tokenized when it is divided up into numerous components,
such as words, phrases, symbols, and other so-called tokens. A wrapper function called
WordTokenize() calls tokenize() on an instance of the Treebank. Class of words in Word
Tokenizer Table. Splitting a big sample of text into words is called word tokenization.
This is necessary for jobs involving natural language processing, where each word must
be recorded and submitted to additional analysis, such as clas- sification and counting
for a specific sentiment, etc.

3.1.4 Summarization
We will be calculating the frequency of each word in our text data and store the fre-
quency together with the text data in a dictionary. Then we tokenize the text data. We
will include the sentences in our final summary data that contain more high frequency
sentences which will be calculated using each word frequency.
436 S. Devi et al.

Fig. 3. Flow Chart of the system

3.1.5 Checking Grammar

Grammar and spelling checks are performed using Language Tool, an open-source pro-
gramme also used as OpenOffice’s spellchecker. This package enables programmers to
find grammatical and spelling errors using a Command-line interface or a Python code
snippet (CLI).

3.2 Flowchart

As shown in Fig. 3

• Insert the YouTube URL into the summary extension.

• A transcript is generated if the video includes subtitles; otherwise, audio transcription
is used to create the transcript.
• The transcript will be summarized using abstractive-based summarization when it is
generated.
• Finally, the extension will display the summery.
 Abstractive Summarizer for YouTube Videos 437

3.3 Features

The user is given four options after the content has been summarized (buttons).

• Translation: The user has the option of translating their text summary into another
language, like Hindi or Marathi
• Speak: The user can hear the condensed transcript read aloud by pressing the speak
button.
• Download: The output for downloading the condensed text into several file formats
allows the user to obtain their condensed text in a variety of file formats.
• Send to mail: The user can utilize this if they wish to send the transcript file to their
email.

4 Conclusion and Future Scope

In this project, the summary of the transcript is done by Abstractive summarization

method. It is a technique which does not make use of sentences from the original content
to make the summary rather it uses paraphrasing of the original text. Existing video
summarization systems require a good hold of technical knowledge. Summarizing videos
based on its subtitle is the fastest way of generating summary, because dealing with text
is much easier and faster compared to training various videos using machine learning
models.
This project may benefit hearing-impaired people who have trouble understanding
videos without subtitles or transcripts because it will be accessible to them. For such
videos they can understand it with the help of a generated summary. In future work
we can manage to change the language of the summarized text to different available
languages from the extension. The idea of Transcript Summarizer can be implemented
in other streaming services as well in future.

References
1. Lavish Mangal, Dhruv Aggarwal, Gulshan Kumar, Jai Kumar, Meenakshi Ag- garwal,
“YouTube Transcript Summarizer.”, 2022.
2. Sanjana R, Sai Gagana V, Vedhavathi K R, Kiran K N, “Video Summarization using NLP.”,
2021.
3. Shraddha Yadav, Arun Kumar Behra, Chandra Shekhar Sahu, Nilmani Chan- drakar,
“Summary and Keyword Extraction from YouTube Video Tran- script.”, 2021.
4. S. Tharun, R. Kranthi Kumar, P. Sai Sravanth, G. Srujan Reddy, B. Akshay, “Survey on
Abstractive Transcript Summarization of YouTube Videos.”, 2022.
5. Gousiya Begum, N. Musrat Sultana, Dharma Ashritha, “YouTube Transcript Summarizer.”,
2022.
6. Li, Y., Lee, S. H., Yeh, C. H., Kuo, C. C, “Techniques for movie content anal- ysis and
skimming.”, 2006.
7. Smaïli, K., Fohr, D., González-Gallardo, C. E., Grega, M., Janowski, L., Jouvet, D., Garcia-
Zapirain, B, “A first summarization system of a video in a target language.” 2018.
438 S. Devi et al.

8. Verma, P., Verma, A, “A review on text summarization techniques.”, 2020.

9. S. Alhojely., J. Kalita, “Recent Progress on Text Summarization.”, 2020.
10. Porwal, K., Srivastava, H., Gupta, R., Pratap Mall, S., Gupta, N, “Video Tran- scription and
Summarization using NLP.”, 2022.

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-
NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/),
which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any
medium or format, as long as you give appropriate credit to the original author(s) and the source,
provide a link to the Creative Commons license and indicate if changes were made.
The images or other third party material in this chapter are included in the chapter’s Creative
Commons license, unless indicated otherwise in a credit line to the material. If material is not
included in the chapter’s Creative Commons license and your intended use is not permitted by
statutory regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder.