DeepSecure: A Real-Time Deep Learning-Based System for Enhancing Cybersecurity in Social Media through DeepFake Detection using LSTM and ResNext CNN

© 2023, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 226
DeepSecure: A Real-Time Deep Learning-Based System for Enhancing
Cybersecurity in Social Media through DeepFake Detection using LSTM
and ResNext CNN
Nikhil Dhiman1, Nitesh Sharma2, Vikalp Vashisth3
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - With the exponential rise of DeepFake content
circulating on social media platforms like Twitter, the need
for robust and real-time detection systems has become
paramount to safeguard digital trust and authenticity. In
response to this pressing concern, "DeepSecure," an
innovative deep learning-based solution tailored for
efficient DeepFake detection. By harnessing the power of
Long Short-Term Memory (LSTM) networks [1] and
Residual Next (ResNext) Convolutional Neural Networks
(CNNs), DeepSecure adeptly analyzes multimedia content
on social media feeds. The proposed system empowers users
and platform administrators to combat the escalating
threat of deceptive content, thereby fortifying the
cybersecurity landscape in social media ecosystems.
Through rigorous experimentation and real-world
implementation, this research endeavors to offer a reliable
and timely defense against the proliferation of DeepFake
content on popular social media platforms.
Key Words: DeepSecure, Real-Time Deep Learning,
Cybersecurity, Social Media, DeepFake Detection, LSTM,
ResNext CNN, Enhancing, Detection System, Digital
Security
1. INTRODUCTION
With the unprecedented growth of social media and the
widespread sharing of multimedia content, the rise of
DeepFake technology has emerged as a significant
cybersecurity threat, jeopardizing the authenticity and
trustworthiness of information circulated online.
DeepFake techniques employ advanced machine learning
algorithms, including Long Short-Term Memory (LSTM)
[1] networks and Residual Next (ResNext) Convolutional
Neural Networks (CNNs) [2], to produce highly realistic
and deceptive fake videos and images. As the prevalence of
DeepFake content continues to escalate on social media
platforms, there is a pressing need for real-time, robust,
and efficient solutions to detect and mitigate the
dissemination of false and misleading information.
In this research paper, we present "DeepSecure," a
pioneering real-time deep learning-based system that uses
the power of LSTM networks [1], Python, and ResNext
CNNs [2] to effectively combat DeepFake threats in the
realm of social media. DeepSecure is meticulously
engineered to intelligently analyze multimedia content
shared on social media platforms, enabling rapid and
accurate identification of DeepFake content. Leveraging
the strength of LSTM [1] and ResNext CNN [2]
architectures, our system enhances cybersecurity
measures by providing a reliable and scalable solution to
counteract the growing sophistication of DeepFake
technology.
Through rigorous experimentation and performance
evaluation, we aim to demonstrate the effectiveness and
practicality of DeepSecure in safeguarding the integrity
and trustworthiness of multimedia content on social
media. The integration of deep learning techniques and
Python programming enables DeepSecure to operate in
real-time, allowing for swift detection and response to
potential DeepFake threats. By presenting a
comprehensive analysis of DeepSecure's capabilities, we
endeavor to contribute valuable insights towards the
ongoing efforts to mitigate the adverse impact of
DeepFake content on social media platforms, fostering a
safer and more secure digital landscape for all users.
2. LITERATURE SURVEY
These days, a number of new threads are emerging as the
usage of AI technology grows. A media or video can be
edited using deep learning to create a false version and
raise security concerns on social media sites. The changed
media may be utilized for journalism, entertainment, and
politics. Some excellent materials, such as an IEEE
(Spectrum) publication [3], help to improve the quality of
Deepfake development and leads to more fake content
over the social media.
Numerous research endeavors have been dedicated to the
detection of deepfakes; however, achieving real-time
detection remains a challenging pursuit. This research
paper seeks to address this crucial gap by focusing on the
development of a real-time deep learning-based system
for robust deepfake detection. By exploring the fusion of
LSTM [1] and ResNext CNN [2] models, our study aims to
contribute to the advancement of cybersecurity in social
media, enabling swift and efficient identification of
manipulated content in a dynamically evolving digital
landscape.
In conclusion, the realm of deepfake detection has
witnessed extensive research efforts, yet the challenge of
real-time detection persists. This research paper takes a
International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 10 Issue: 08 | Aug 2023 www.irjet.net p-ISSN: 2395-0072

significant stride towards addressing this limitation by
presenting "DeepSecure," a real-time deep learning-based
system designed to enhance cybersecurity in social media
through effective deepfake detection. By leveraging the
synergistic potential of LSTM [1] and ResNext CNN [2]
models, our study endeavors to bridge the gap and
provide a robust solution for the timely identification of
manipulated content. This contribution is poised to bolster
the resilience of digital platforms against the pervasive
threat of deepfakes, fostering a more secure and
trustworthy online environment.
3. TECHNOLOGIES USED
3.1 LSTM (LONG SHORT - TERM MEMORY):
A RNN called LSTM [1] is used as it can identify a segment
of signals that recurs frequently in the whole temporal
signal sequence and LRD in sequential data. It utilizes a
memory cell with gating mechanisms, such as the forget
gate, input gate, and output gate, to control the flow of
information. These gates enable LSTM to selectively retain
important information over time, making it suitable for
tasks involving sequential data. LSTM has proven to be
effective in various applications where context and
temporal dependencies are crucial, making it a prominent
choice in deep learning models.
Follow Chart - 1: LSTM Structure
3.2 RESNEXT CNN (RESIDUAL NEXT CONVOLUTIONAL
NEURAL NETWORK):
is an advanced variant of the Convolutional Neural
Network architecture, featuring a cardinality parameter
that enhances its representation learning capabilities. By
partitioning input data into multiple groups, ResNext CNN
[2] enables the network to learn diverse feature mappings,
leading to improved performance in computer vision
tasks. Its flexible and scalable design has made it a popular
choice in state-of-the-art deep learning models for image
recognition and object detection. In the context of
DeepSecure, ResNext CNN's spatial understanding
complements LSTM's temporal analysis, contributing to
robust DeepFake detection and bolstering cybersecurity
measures in social media platforms. A thorough
exploration of ResNext CNN's applications and
advancements in computer vision tasks will provide
valuable insights for its role in deep learning-based
DeepFake detection.
3.3 Django:
Django, a prominent high-level open-source Python web
framework, occupies a significant position in modern web
development due to its resilience, scalability, and rapid
development capabilities. Built on the principles of Don't
Repeat Yourself (DRY) and rapid development, Django
empowers developers to create dynamic web applications
with exceptional efficiency, embracing the Model-View-
Template (MVT) architectural pattern. This framework
revolutionizes the development process, facilitating the
separation of concerns, enhancing maintainability, and
fostering seamless collaboration between components.
Follow Chart - 2: Django MVT Architecture
The framework's comprehensive array of integrated
functionalities, encompassing intricate aspects such as
authentication, database administration, and URL routing,
expedites development workflows and enables a
concentrated emphasis on the nucleus of application logic.
Its unwavering commitment to industry best practices,
encompassing impregnable authentication protocols and
fortification against prevalent web vulnerabilities, further
accentuates its stature as an indispensable tool for crafting
secure and multifaceted web solutions.
3.4 TWEEPY:
With Tweepy [6], developers can interact with Twitter
data, such as fetching tweets, posting tweets, accessing
user information, and performing searches. It's simple and
intuitive interface allows for seamless integration with
Python applications, making it a popular choice among
developers for accessing and analyzing Twitter data.
By leveraging Tweepy's functionalities, researchers and
data analysts can gather real-time data from Twitter,
enabling them to perform sentiment analysis, track
trending topics, and conduct social media monitoring.
Tweepy's comprehensive documentation [6] and active
community support make it an invaluable tool for
accessing and utilizing Twitter data in various
applications, from market research to social media
analytics. Its ability to handle authentication and rate
limiting further simplifies the process of accessing Twitter

data, making it an essential library for anyone working
with Twitter data in Python.
Follow Chart - 3: Tweepy Working
4. DEEPSECURE ARCHITECTURE
4.1 MODEL BUILDING
4.1.1 DATA SETS USED
As a real-time deepfake detection system, DeepSecure
needed an extremely effective model to distinguish
between authentic and fraudulent content on social media
postings. We used a range of data sets from several
sources, including Deep fake Detection Challenge data [4],
and FaceForensic++ [5], in order to make the system
reliable. We attempted to get an equal quantity of false and
true data. 163 media was gathered.
4.1.2 PRE PROCESSING
In the pre-processing stage, the media go through a variety
of steps to filter out background noise and extract the
crucial information, namely the face characteristics. A
media is split up into frames in the first step of deepSecure
preprocessing such that each frame isolates the face and
produces a new media that only contains the frames that
include faces.
4.1.3 SEPARATING TEST AND TRAIN DATA
We chose to design a distribution of 70% train data and
30% test data in order to make DeepSecure very efficient.
The DeepSecure model will be trained using train data,
whereas the model may be tested using test data. Equal
amounts of false and genuine data are present.
4.1.4 MODEL BUILDING
Convolutional neural networks (CNN) and recurrent
neural networks are both integrated into our model's
design. The preprocessed frames are now sent to extract
features using Trained ResNext CNN (2). After receiving
the characteristics from ResNext CNN, we send them to a
long short-term memory network [1], which classifies the
media as either true or false.
4.2 FETCHING DATA FROM SOCIAL MEDIAL
Data collection from social media networks, in this case
Twitter, is the first step towards putting DeepSecure into
practice. Twitter offers developers access to its APIs so it
can retrieve tweet data. Using Tweepy [6], a Python
module that allows for the retrieval of tweets based on
keywords, we are able to receive the stream of Twitter
data into our system.
Once we fetch the stream of data from Twitter, we filter
the data stream to get only the tweets which include
media, this can be done by filtering the data based on the
media_url key in the data received from the Twitter
stream. After getting all the posts which contain media we
save the details like post_id, posted_at and the media_url
in a centralized database and also download the media
from the URL present in the media_url key. This locally
downloaded media will be checked for deepfake.
4.3 RUNNING MODEL ON REAL-TIME SOCIAL
MEDIA DATA
The model that was built in point 4.1 is now given the data
that we received in point 4.2. The authenticity of locally
downloaded material is verified before it is remapped to
the matching post_id stored in the central database and
the final outcome is shown to the users in real time.
Follow Chart - 4: DeepSecure Architecture
5. RESULTS AND PERFORMANCE
The DeepSecure framework demonstrated remarkable
proficiency in the seamless retrieval and subsequent
processing of real-time Twitter data, thereby enabling a
meticulous differentiation between authentic and spurious
media artifacts. This intricate analysis culminated in the

attainment of an outstanding accuracy rate of 95.83%,
accompanied by an impressively low loss value of 0.177.
These compelling performance indicators serve as a
testament to DeepSecure's exceptional prowess as a
cutting-edge real-time deepfake detection system. The
substantiality of these findings underscores the system's
robustness and underscores its potential significance in
bolstering cybersecurity efforts in the realm of social
media.
Screenshot - 1: DeepSecure Home Page
Screenshot - 2: DeepSecure Prediction Page
6. CONCLUSIONS
In conclusion, "DeepSecure" is a cutting-edge solution for
bolstering cybersecurity in social media by detecting and
mitigating DeepFake content. DeepSecure leverages the
power of both LSTM [1] and RestNext CNN [2] to make a
efficient and robust real-time deepfakes detection system
With its balanced dataset and integration of CNN and RNN
components, DeepSecure offers real-time protection
against deceptive content, ensuring a safer and more
secure digital landscape for social media users. This
research paper contributes valuable insights into the
domain of DeepFake detection and highlights the potential
of DeepSecure as a reliable tool for fortifying cybersecurity
in social media platforms.
REFERENCES
[1] Ralf C. Staudemeyer, “Understanding LSTM – a tutorial
into Long Short-Term Memory Recurrent Neural
Networks”, arXiv:1909.09586v1 [cs.NE] 12 Sep 2019
[2] Jeremy Jordan, “Common architectures in
convolutional neural networks.”, 19 Apr 2018
[3] SallyAdee, “What Are Deepfakes and How Are They
Created?” IEEE Spectrum
[4] Kaggle, “Deep Fake Detection Challenge”
[5] Kaggle, “FaceForensics++”
[6] Joshua Roesslein, “tweepy Documentation Release
3.6.0”, Mar 02, 2018

DeepSecure: A Real-Time Deep Learning-Based System for Enhancing Cybersecurity in Social Media through DeepFake Detection using LSTM and ResNext CNN

More Related Content

Similar to DeepSecure: A Real-Time Deep Learning-Based System for Enhancing Cybersecurity in Social Media through DeepFake Detection using LSTM and ResNext CNN (20)

More from IRJET Journal (20)

Recently uploaded (20)

DeepSecure: A Real-Time Deep Learning-Based System for Enhancing Cybersecurity in Social Media through DeepFake Detection using LSTM and ResNext CNN