Volume 7, Issue 3 March 2022 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Video and Text Summarisation using NLP

Rushali R. Shetty1, Samrudhi M.2, Sayed Ayman Bukhari3, Dr. Manjunath DR4
Undergraduate Student1,2,3, Assistant Professor 4
B.M.S. College of Engineering Bengaluru, India

Abstract:- Chunks of Information are available on the The proposed system in the project is to create an
internet, it is most important to provide a solution to get extension such that it summarises transcripts of YouTube
information most efficiently and accurately. With the videos, podcasts, etc . We are looking at two approaches, the
increased use of the internet and smartphones, there has videos for which the transcripts are available, we will be
been a steady increase in online learning, entertainment, directly summarising the content, whereas for the videos for
and other activities. But sometimes we don't have the time which the transcripts are not available we are looking to convert
to go through all the material such as videos or podcasts in theaudio file to a text file, which will finally, be summarised
case it was too long. This necessitates a mechanism that using the algorithm. This would turn out to be particularly
reduces the length of a document to one with limited words useful to save time and effort of either watching the entire video
which essentially retains the same meaning. Automatic or having to read through the entire transcript of the desired
Summarization can come in hand in this aspect. video. Along with which it also is useful to savespace if
one is looking to download and save the transcript for later.
I. INTRODUCTION
II. PROBLEM STATEMENT
Due to the exponential increase in the availability of data
in the recent years – amounting to almost a million terabytes – When the user uses the web application, they are given the
the need to summarize is imperative.With the ever-increasing choice to upload a video or link to the video. This video input
pace of life, the public no longer has the time to read long is passed through speech-to-text API to generate the subtitle for
articles or transcripts nor can they watch long videos or audio the same. Subtitles generated are passed through the algorithm
clips; a tool to develop an effective summary is much needed. of the user's choice. When passed through the algorithm a
summarised text for subtitle is generated. According to the
An enormous amount of digital data is sourced from the summarised subtitle, the video is trimmed. This trimmed video
web resources on the Internet such as websites, user reviews, is made available for the user to download.
news, blogs, social media networks, YouTube videos, Podcasts,
so on and so forth. Furthermore, another source of digital data III. LITERATURE SURVEY
could be found in the archives of various news articles, novels,
books, legal documents, biomedical documents, scientific Wealth of information and expertise that must be
papers and primarily other audio and video formats, such as any adequately summarised in order to be useful. Because of the
OTT platform videos like Youtube, Amazon Prime, Saavan increasing availability of documents, there is a need for
music, Spotify, etc. There has been a noticeable growth of extensive research in the field of natural language processing
digital content on the internet on a daily basis. As a result, users (NLP) for automatic text summarization. Automatic text
waste a lot of timetrying to find the information they desire. summarizationis the process of creating a concise and fluent
They can't even read and comprehend all of the textual content summary without the assistance of a human while maintaining
in search results, much less the visual and audio content. the sense of the original text material.
Many portions of the resulting data are repeated or unimportant.
Asa result, summarising and condensing text, visual, and audio A. Key Terms
resources becomes urgent and critical. Manual summarization  Automatic Summarization: The technique, where a
is an expensive task which requires a significant amount of time computer program shortens longer texts and generates
and effort. Manually summarisingthis massive amount of data summaries to pass the intended message
is extremely difficult for humans. The key solution to this  Speech-to-API:The Speech to Text API is a basic API that,
quandary is Automatic Summarization (AS). as the name implies, allows you to transform audio input
into written text.
The primary goal of an automatic summarization (AS)  Image analysis:Image analysis or imagery analysis is the
system is to generate a summary that includes the main ideas in extraction of meaningful information from images; mainly
the input document in as minimal space as possible while from digital images by means of digital image processing
minimising repetition.Users will benefit from the automatically techniques.
generated summaries, which will save them a significant  Natural language processing: Natural language processing
amount of time and effort. This allows them to easily access is a subfield of linguistics, computer science, and artificial
important data and saves time. intelligence concerned with computer-human interaction.

IJISRT22MAR1199 www.ijisrt.com 1583

Volume 7, Issue 3 March 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
B. Existing System E. Applications of the proposed system :
Approaches to Automatic Text Summarization (ATS) are  E-learning and class assignments: Many teachers frame
either extractive, abstractive, or hybrid. The extractive method their lectures with case studies and current events.
chooses the most important sentences from the input Summarization can help teachers update their content more
documents and concatenates them to form the summary. The quickly by producing summarised reports on their subject of
abstractive method represents the input documents in an interest. This will also make it easier for the student to learn
intermediate representation before generating the summary the important topics.
with sentences that differ from the original sentences. The  Video Scripting: Video is quickly becoming one of the
extractive and abstractive approaches are combined in the most important marketing tools. People are now sharing
hybrid approach. Despite all of the methods proposed, the videos on professional networks such as LinkedIn, in
generated summaries are still far from human-generated addition to video-focused platforms such as YouTube or
summaries. The extractive approach is the focus of the majority Vimeo. Scripting may be required in varying degrees
of studies. It is necessary to place a greater emphasis on depending on the type of video. When attempting to write a
abstractive and hybrid approaches. script that incorporates research from multiple sources,
summarization can be a valuable ally.
Automatic video summarization is implemented as a  Medical cases: With the rise of tele-health, there is an
sequential decision-making process, and a diversity- increasing need to better manage medical cases, which are
representativeness reward is used to train a Summarizer to now entirely digital. Because telemedicine networks
produce diverse and representative video summaries. The promise a more open and accessible healthcare system,
diversity reward computes the distance (expressing visual technology must make the process scalable. When it comes
resemblance) of the selected key-frames from the remaining to analysing medical cases and routing them to the
frames of the video, whereas the representativeness reward appropriate health professional, summarization can be a
measures the dissimilarity among the selected key-frames. critical component in the tele-health supply chain.
C. Shortcomings of existing system Media monitoring, newsletters, search marketing
There already exists an Automatic Text Summarization and SEO, internal document workflow, financial research,
(ATS) and on another stand alone application exists an legal contract analysis, social media marketing, question
Automatic Video Summarization. But they have not been done answering, and bots are just a few examples.
together before and that is our project’s aim.
IV. PROPOSED SOLUTION
D. NLP, LexRank and LSA in Text summarisation :
In the big data age, the amount of text data available from
various sources has exploded. This lengthy text contains a A. Requirements
LexRank is a graphical method for automatically summarising The proposed solution must satisfy the following functional
text. It's also an unsupervised method for text summarization. requirements and provide the following features:
The eigenvector centrality concept is used, and sentence
 User Interface: Users will link to the video to summarise or
importance is calculated. In this model, the adjacency matrix of
upload the file itself. Summarized video is available to
the graph representation of sentences is a connectivity matrix
download later.
based on intra-sentence cosine similarity. This sentence
 Subtitle generation: Subtitle is generated using speech-to-
extraction is primarily concerned with a set of sentences with
text api
the same intent, in which a centroid sentence is chosen to serve
as the mean for all other sentences in the document. The  Text summarisation: Extracted subtitles are summarised
sentences are then ranked based on their similarities. using the extractive text summarisation approach.
Algorithms such as Lex Rank and LSA are used.
Latent semantic analysis (LSA) is a method for extracting  Video trimmer: Video is trimmed based on the summarised
a representation of text semantics from observed words that is subtitle.
unsupervised. It was first proposed in the news domain to select
highly ranked sentences for single and multi-document B. Key Components
summarization. The LSA method begins by constructing a  Video Input
term-sentence matrix (n by m matrix), in which each row  Subtitle generator
corresponds to a word from the input (n words) and each  Summarizer
column corresponds to a sentence (m sentences).  Video trimmer.

IJISRT22MAR1199 www.ijisrt.com 1584

Volume 7, Issue 3 March 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
C. Proposed Methodology [5.] K. Zhou and Y. Qiao, “Deep Reinforcement Learning for
The goal of this project is to build an accurate video Unsupervised Video Summarization with Diversity-
summarisation webapp using its subtitles. To complete this goal Representativeness Reward,” in Proc. of the 2018 AAAI
we will be using the following steps - Conf. on Artificial Intelligence, 2018.
 Accept the video input from the user in the form of actual [6.] B. Zhao, X. Li, and X. Lu, “Property-constrained dual
video or URL. Accept the subtitle file if provided by the learning for video summarization,” IEEE Trans. on Neural
user. Networks and Learning Systems, vol. 31, no. 10, pp.
 Download the video if the URL is provided. 3989–4000, 2020.
 Generate audio files for the video if subtitle is not available. [7.] L. Lebron Casas and E. Koblents, “Video Summarization
Convert the audio file to text file using speech-to-text API. with LSTM and Deep Attention Models,” in MultiMedia
 Accept the algorithm selected by the user. Modeling, I. Kompatsiaris, B. Huet, V. Mezaris, C.
 Pass the subtitle to the selected algorithm and generate the Gurrin, W.-H. Cheng, and S. Vrochidis, Eds. Cham:
summarised subtitle. Springer International Publishing, 2019, pp. 67–79.
 According to the summarised subtitle, trim the original [8.] Niladri Chatterjee, Amol Mittal and Shubham Goyal’s
video. “Single Document Extractive Text Summarization Using
 Trimmed video is made available for the user to download. Genetic Algorithms” (2012)
[9.] Amol Tandel, Brijesh Modi, Priyasha Gupta, Shreya
V. CONCLUSION Wagle and Sujata Khedkar's "Multi-document text
summarization - A survey" 2016 International Conference
Manual summarization is a time-consuming and on Data Mining and Advanced Computing
expensive task with numerous steps. For example, to manually (SAPIENCE).
summarise a single dataset, the following steps are taken: [10.] Aditya Jain, Divij Bhatia, Manish K Thakur’s “Extractive
understanding what the document / video is about, extracting Text Summarization using Word Vector
the "most important" parts from it, and composing a summary Embedding”(2017).
that meets the following requirements: summary readability and [11.] Shi Ziyan “The Design and Implementation ofDomain-
linguistic quality, summary consistency and content coverage. specific Text Summarization System based on Co-
The absence of redundancy in the produced summary. reference Resolution Algorithm” 2010 Seventh
International Conference on Fuzzy Systems and
Because of the difficulty of manually summarising the Knowledge Discovery.
massive amount of textual content on the Internet or in various [12.] Leonhard Hennig, Winfried Umbrath, and Robert
archives, AS systems have emerged as the primary technology Wetzker. 2008. An ontology based approach to text
to address this urgent and pressing issue. summarization. In Web Intelligence and Intelligent Agent
Technology, 2008. WI-IAT’08. IEEE/WIC/ACM
Automatic summarization is a complex task with
International Conference on, Vol. 3. IEEE, 291–294.
numerous subtasks. Every subtask is capable of producing high-
[13.] Chin-Yew Lin. 2004. Rouge: A package for automatic
quality summaries. The extracted sentences are produced as a
evaluation of summaries. In Text Summarization
summarised text and converted into audio form. When
Branches Out: Proceedings of the ACL-04 Workshop. 74–
compared to a traditional approach, the proposed model
81.
improves accuracy.

REFERENCES

[1.] Z. Xian, F. Jiao, “Iterative video summarization,using

ILS-SUMM ”, vol. 405, pp. 200 – 207, 2020.
[2.] N. Gonuguntla, B. Mandal, N. Puhan et al., “Enhanced
Deep Video Summarization Network,” in 2019 British
Machine Vision Conf. (BMVC), 2019.
[3.] Song, Y., Vallmitjana, J., Stent, A., & Jaimes, A. (2015).
Tvsum: Summarizing web videos using titles. In
Proceedings of the IEEE conference on computer vision
and pattern recognition (pp. 5179- 5187)
[4.] Agyeman, R., Muhammad, R., & Choi, G. S. (2019,
March). Soccer Video Summarization Using Deep
Learning. In 2019 IEEE Conference on Multimedia
Information Processing and Retrieval (MIPR) (pp. 270-
273). IEEE.

IJISRT22MAR1199 www.ijisrt.com 1585