Smart Contract Vulnerability Detection

KARAN RANA(RA2111003030446), SUHAIL SAIFI(RA2111003030439), MOHD. ZUFAR HASAN

ALVI(RA2111003030435), SOHAIL(RA2111003030436)
CSE, SRMIST GHAZIABAD
[email protected] [email protected]
[email protected]
[email protected]

ABSTRACT

The proliferation of social media platforms has led to a significant rise in cyberbullying incidents, which poses
serious challenges for online safety and mental well-being. This paper presents a comprehensive study on leveraging
tweet data for automated cyberbullying detection through advanced machine learning techniques. We propose a novel
framework that employs natural language processing (NLP) and machine learning algorithms to identify and classify
cyberbullying content within Twitter data. The framework integrates various feature extraction methods and
classification models to enhance detection accuracy. Our experimental results demonstrate the effectiveness of the
proposed approach, achieving high precision and recall rates in distinguishing between abusive and non-abusive
tweets. This research contributes to the development of automated tools for monitoring and mitigating cyberbullying
on social media platforms, offering insights into the potential for improved online safety through technological
interventions.

I. INTRODUCTION significant implications for mental health and social

harmony.
The pervasive influence of social media platforms has Cyberbullying, defined as the use of electronic
revolutionized communication, yet it has also given rise communication to bully a person by sending threatening,
to significant challenges, among which cyberbullying is a intimidating, or malicious messages, represents a
prominent concern. Cyberbullying, characterized using growing concern in the digital era. Unlike traditional
digital platforms to inflict psychological harm on bullying, cyberbullying operates in a virtual environment
individuals, represents a growing threat that impacts where the perpetrators can remain anonymous, and the
users' mental health and well-being. The anonymous and victims can experience harassment at any time and from
often unregulated nature of online interactions any location. The psychological impact of cyberbullying
exacerbates the difficulties in identifying and mitigating can be profound, leading to issues such as anxiety,
such harmful behaviours. depression, and social withdrawal. The anonymity and
scale of social media platforms like Twitter exacerbate
In the digital age, social media platforms have become these issues, making it challenging to identify and address
integral to personal and professional communication, cyberbullying effectively.
providing unprecedented opportunities for individuals to The sheer volume of content generated on Twitter—
connect, share, and express their thoughts. Twitter, a with over 500 million tweets posted daily—presents a
widely used microblogging service, exemplifies this shift significant challenge for manual detection of
with its real-time, concise, and dynamic nature. While cyberbullying. Traditional methods of identifying
Twitter facilitates positive interactions and community harmful content, such as human moderation and manual
building, it also serves as a venue for detrimental reporting, are insufficient for managing the vast amount
behaviors, including cyberbullying—a phenomenon with of data and the speed at which content is produced. This
limitation highlights the need for automated solutions that

1
can efficiently process and analyse large datasets to detect system robustness, such as incorporating contextual
instances of cyberbullying in real-time. information and leveraging ensemble methods.
Recent advancements in machine learning (ML) and By providing a detailed analysis of these
natural language processing (NLP) offer promising methodologies and challenges, this research aims to
approaches for addressing this challenge. Machine contribute to the development of effective and scalable
learning, a subset of artificial intelligence, involves the solutions for combating cyberbullying on social media
development of algorithms that can learn from and make platforms. The goal is to enhance the ability of automated
predictions or decisions based on data. When applied to systems to identify and mitigate harmful behaviour,
textual data, ML algorithms can identify patterns and thereby fostering safer and more supportive online
anomalies that may indicate cyberbullying. Natural environments. Through this study, we seek to advance the
language processing, which enables computers to field of cyberbullying detection and offer practical
understand and interpret human language, further insights for improving the well-being of social media
enhances these algorithms by providing the ability to users.
analyse the context, sentiment, and intent behind textual
content. As social media continues to evolve, so too must the
The integration of ML and NLP techniques into the strategies and technologies designed to address its
automated detection of cyberbullying involves several associated risks. This research underscores the
critical components. Data preprocessing is a foundational importance of advancing automated systems for
step, involving the cleaning and normalization of tweet cyberbullying detection, emphasizing the need for
data to prepare it for analysis. Feature extraction, which ongoing innovation and refinement in machine learning
includes techniques such as word embeddings and and natural language processing techniques. The findings
sentiment analysis, plays a crucial role in transforming of this study are anticipated to provide valuable insights
raw text into meaningful inputs for machine learning not only into the effectiveness of current methodologies
models. The choice of algorithms, such as support vector but also into potential areas for future research and
machines, random forests, or deep learning models, development. By enhancing the capabilities of automated
impacts the accuracy and efficiency of detection systems. detection systems, we aim to contribute to a safer digital
Evaluating these models requires robust metrics and environment where individuals can engage in online
validation methods to ensure that they can generalize well interactions free from the fear of harassment and abuse.
to new and diverse datasets. Ultimately, the research aspires to support broader efforts
This research paper aims to explore the potential of to foster a more respectful and empathetic online
leveraging tweet data for automated cyberbullying community, benefiting both individuals and society at
detection through a comprehensive machine learning large.
approach. The study will begin with an overview of the
theoretical framework underpinning ML and NLP
techniques, followed by a detailed examination of the
methods employed in preprocessing and feature
extraction. The core of the research will involve
developing and evaluating various machine learning
models to assess their effectiveness in detecting
cyberbullying. The evaluation will consider factors such
as precision, recall, and F1-score, as well as the ability of
the models to handle the inherent variability and
complexity of natural language.
In addition to the technical aspects, the paper will
address the challenges associated with automated
detection systems. These include dealing with the LITERATURE REVIEW
evolving nature of language, the risk of false positives and
negatives, and the ethical considerations surrounding Cyberbullying has emerged as a significant
privacy and data security. The research will also explore concern in the digital age, characterized by the
potential strategies for improving detection accuracy and use of electronic communication to intimidate,
threaten, or demean individuals. Unlike

2
traditional bullying, cyberbullying occurs in the 2018). More recent advances include deep
virtual space, where anonymity and the potential learning models, such as Convolutional Neural
for widespread dissemination of harmful content Networks (CNNs) and Recurrent Neural
exacerbate the impact on victims. Research has Networks (RNNs), which have demonstrated
demonstrated that cyberbullying can lead to superior performance in capturing contextual
severe psychological effects, including anxiety, and semantic nuances in text (Kim, 2014; Tang
depression, and social withdrawal, particularly et al., 2015).
among adolescents (Kowalski et al., 2014;
Slonje & Smith, 2008). The pervasive nature of III. Natural Language Processing (NLP)
digital communication means that victims can Techniques
experience harassment at any time and from any
location, making it a persistent and challenging Natural Language Processing (NLP) plays a
issue to address. crucial role in transforming raw text data
into meaningful features for machine
I. The Role of Social Media in Cyberbullying learning models. Key NLP techniques
include tokenization, stemming, and
Twitter, a widely used social media lemmatization, which prepare the text for
platform, plays a dual role in both analysis by reducing it to its base
facilitating and combating components (Bird et al., 2009). Feature
cyberbullying. The platform's design— extraction methods, such as Term
characterized by short, real-time posts— Frequency-Inverse Document Frequency
creates an environment where harmful (TF-IDF) and word embeddings (e.g.,
interactions can quickly escalate and Word2Vec and GloVe), enable the
reach a broad audience (Kumar et al., conversion of text into numerical
2018). Studies have shown that Twitter representations that capture semantic
data, due to its high volume and the relationships (Mikolov et al., 2013;
informal nature of its content, can serve Pennington et al., 2014). Sentiment analysis,
as a rich source for detecting which involves determining the emotional
cyberbullying but also presents tone of text, has also been applied to detect
challenges in terms of data complexity negative or abusive content, providing
and variability (Zhang et al., 2018). The additional context for identifying
platform's anonymity and the ability to cyberbullying (Pang & Lee, 2008).
create multiple accounts further
complicate efforts to identify and IV. Existing Models for Cyberbullying Detection
address abusive behavior.
Several studies have focused on
II. Machine Learning Techniques in Text Analysis developing models specifically for
cyberbullying detection. For example, the
Machine learning (ML) techniques have become work by Nobata et al. (2016) utilized a
increasingly important in analyzing and combination of linguistic features and
processing textual data. ML methods, such as machine learning algorithms to detect
supervised learning algorithms, have been abusive comments in social media. Their
successfully applied to various text model incorporated various features, such as
classification tasks, including sentiment analysis lexical and syntactic patterns, to enhance
and spam detection (Manning et al., 2008). For detection accuracy. Similarly, Xu et al.
cyberbullying detection, these techniques can be (2018) proposed a deep learning-based
employed to identify patterns indicative of approach that leveraged convolutional
abusive language. Algorithms such as Naive neural networks to capture contextual
Bayes, Support Vector Machines (SVM), and information in text, achieving notable
ensemble methods like Random Forests have improvements in classification performance.
been used to classify text data into categories of Despite these advancements, challenges
abusive or non-abusive content (Zhang et al., remain, including handling the variability in

3
 language use, detecting context-specific concerns are paramount, as the use of social
abuse, and managing the trade-off between media data raises questions about the handling
precision and recall (Sood et al., 2012). and storage of sensitive information. Ensuring
that data is anonymized and used in accordance
V. Datasets for Cyberbullying Detection with ethical guidelines is crucial for maintaining
user trust (Shadbolt et al., 2020). Furthermore,
The availability of annotated datasets is addressing potential biases in detection models
critical for training and evaluating is essential to avoid unfairly targeting specific
cyberbullying detection models. Publicly user groups and to ensure that the system is
available datasets, such as the equitable and inclusive (Bolukbasi et al., 2016).
Cyberbullying Dataset from the Kaggle
platform and the HatEval dataset from VIII. Recent Advances and Future Directions
SemEval, provide valuable resources for
researchers. These datasets include a Recent advancements in machine learning and
range of text samples, annotated for NLP continue to drive progress in cyberbullying
various types of abusive behavior, which detection. Techniques such as transformerbased
facilitates the development and models (e.g., BERT and GPT) have shown
benchmarking of detection algorithms promise in understanding context and nuance in
(Hate Speech Dataset, 2018; Basile et text, potentially improving detection accuracy
al., 2019). The quality and diversity of (Devlin et al., 2018; Radford et al., 2019).
these datasets are essential for ensuring Future research directions include exploring
that models generalize well to different these advanced models, integrating multi-modal
contexts and populations. data (e.g., images and text), and addressing the
ethical implications of automated detection
VI. Evaluation Metrics systems. Continued innovation and
interdisciplinary collaboration will be crucial for
Evaluating the performance of developing more effective and ethical solutions
cyberbullying detection models requires for combating cyberbullying.
the use of appropriate metrics. Common
evaluation metrics include accuracy,
precision, recall, and F1-score, each
providing different insights into the
model's performance (Manning et al.,
2008). Accuracy measures the overall
correctness of predictions, while
precision and recall provide insights into
the model's ability to identify positive
instances of cyberbullying. The F1score,
which combines precision and recall, is
particularly useful for balancing the RESEARCH METHODOLOGY
trade-offs between these metrics.
Additionally, considerations of false
positives and false negatives are crucial,
as high rates of either can impact the
effectiveness of the detection system
(Zhang et al., 2018).

VII. Ethical Considerations

The development and deployment of automated

cyberbullying detection systems involve
important ethical considerations. Privacy

4
 Tokenization refers to the process of segmenting
the text into individual units or tokens, such as words
or phrases. For instance, tokenizing the sentence
"Stop bullying others!" results in the tokens: ["Stop",
"bullying", "others"]. This step is essential as
machine learning models analyze patterns based on
these individual word units.
2. Lemmatization - Lemmatization involves
reducing words to their base or dictionary
forms. For example, "running" and "ran" are
normalized to the lemma "run." This process
Our working is divided into 2 phases mainly
ensures that different forms of a word are
(I) NLP
treated consistently, thereby simplifying the
(II) Machine learning
data and enhancing the performance of
machine learning algorithms.
[1] Phase 1: Natural Language Processing (NLP)
3. Vectorization - Following tokenization and
The initial phase of this project focuses on Natural
lemmatization, the text data is converted into
Language Processing (NLP), which is crucial for
numerical representations that can be
preparing raw tweet data for subsequent machine
interpreted by machine learning models. This
learning algorithms. This phase encompasses several
conversion is achieved through vectorization
key sub-steps that convert unstructured text into a
methods such as:
structured format suitable for analysis.
4. TF-IDF (Term Frequency-Inverse Document
Data Extraction
Frequency): This method assesses the
The first step in this phase involves data
significance of words in the context of the
extraction, where raw text data, specifically tweets,
dataset by evaluating their frequency in
is collected from the Twitter platform. The Twitter
individual documents relative to their
API, among other tools, is utilized to retrieve tweet
frequency across the entire dataset.
data along with metadata such as timestamps,
5. Word Embeddings (Word2Vec, GloVe):
usernames, and hashtags. The objective is to
These techniques capture semantic
assemble a comprehensive dataset that includes both
relationships between words by representing
cyberbullying and non-cyberbullying content.
them as dense vectors in a high-dimensional
Data Cleaning
space.
Following extraction, the data undergoes a
cleaning process to eliminate noise and irrelevant 6. Through these preprocessing techniques, the
information. The cleaning process involves: raw tweet data is transformed into a
• Removing special characters, numerical structured format suitable for machine
values, and URLs. learning, thereby facilitating the subsequent
• Converting all text to lowercase to phase of the project.
standardize the format.
• Filtering out non-textual elements such as
emojis and symbols, unless they hold specific
relevance for the analysis.
This cleaning process ensures that the data is
uniform and ready for further processing.
Preprocessing Techniques
1. Tokenization

5
 (SVM), K-Nearest Neighbors (KNN), Logistic
Regression, and Stochastic Gradient Descent
(SGD) Classifier.
Support Vector Machine (SVM)
B. Support Vector Machine (SVM) is a powerful
supervised learning algorithm employed for
classification and regression tasks. SVM operates
by constructing a hyperplane in a
highdimensional space that separates different
classes with the maximum margin. The primary
goal of SVM in cyberbullying detection is to
identify an optimal boundary that differentiates
cyberbullying tweets from non-cyberbullying
ones.
C. SVM’s effectiveness stems from its ability to
handle both linear and non-linear classifications

through the use of kernel functions. A kernel

function transforms the data into a
higherdimensional space where a linear
separation is possible. Commonly used kernels
include the linear, polynomial, and radial basis
function (RBF) kernels. The choice of kernel
significantly impacts the performance of the
SVM model. For this project, hyperparameter
tuning, including the selection of the kernel and
regularization parameters, is essential to
achieving optimal classification accuracy.
K-Nearest Neighbors (KNN)
D. K-Nearest Neighbors (KNN) is a straightforward,
instance-based learning algorithm used for both
classification and regression. The fundamental
principle of KNN is
to classify a data point based on the majority class
of its k-nearest neighbors in the feature space. In
the context of cyberbullying detection, KNN
evaluates the similarity of a tweet to its nearest
[2] Machine Learning Algorithms neighbors and assigns a class label based on a
A. In the subsequent phase of the project, machine majority vote.
learning algorithms are applied to classify E. The performance of KNN is highly dependent on
processed tweet data into cyberbullying or the choice of the parameter k, which determines
noncyberbullying categories. This phase utilizes the number of nearest neighbors considered.
various well-established algorithms, each with Additionally, the distance metric used to measure
distinct methodologies and strengths. The similarity—such as Euclidean distance,
following subsections provide a detailed Manhattan distance, or Minkowski distance—
examination of the Support Vector Machine can influence the results. KNN’s simplicity and

6
 interpretability make it a valuable tool, though its J. Each of these algorithms brings a unique set of
performance may degrade with highdimensional advantages to the cyberbullying detection task.
data and large datasets. Logistic Regression The selection of an appropriate algorithm,
F. Logistic Regression is a statistical method coupled with rigorous parameter tuning and
designed for binary classification problems. It evaluation, is vital for enhancing the accuracy
models the probability of a binary outcome by and robustness of the classification model.−
employing a logistic function to estimate the
probability that a given input belongs to one of the
two classes. In the domain of cyberbullying
detection, logistic regression estimates the [3] Evaluation Phase
likelihood that a tweet falls into either the The evaluation phase is crucial in assessing the
cyberbullying or non-cyberbullying category. performance of the machine learning models used
G. The logistic function, or sigmoid function, for cyberbullying detection. This phase involves
transforms the linear combination of the input comparing the predicted classifications with the
features into a probability value between 0 and true labels to determine the effectiveness of the
1. The model parameters are estimated using models. Key evaluation metrics used in this phase
maximum likelihood estimation. Logistic include precision, recall, F-measure, and accuracy.
Regression is advantageous for its simplicity, These metrics provide a comprehensive
interpretability, and efficiency, especially when understanding of how well the models perform in
the relationship between the predictors and the identifying cyberbullying content.
outcome is approximately linear. Regularization [4] Precision
techniques such as L1 (Lasso) and L2 (Ridge) Precision measures the accuracy of the positive predictions
can be employed to prevent overfitting and made by the model. It is the proportion of true positive
enhance model generalization. predictions out of all the instances that were predicted as
positive. TP
Stochastic Gradient Descent (SGD) Classifier
Precision=
H. The Stochastic Gradient TP + FP
Descent (SGD)
Classifier is an optimization method used to train where:
various types of models, including linear • TP(True Positives): The number of correctly predicted
classifiers. SGD operates by iteratively updating positive instances.
model parameters based on small, random • FP (False Positives): The number of instances
incorrectly predicted as positive.
subsets of the training data, known as [5] Recall
minibatches. This approach enables the classifier Recall (also known as Sensitivity or True Positive Rate)
to handle large-scale datasets and high- measures the model's ability to identify all relevant positive
dimensional feature spaces efficiently. instances. It is the proportion of true positive predictions out of
all the actual positive instances.
I. In the context of cyberbullying detection, the
. TP
SGD Classifier approximates the solution to the
Recall =
classification problem by minimizing the loss TP + FN
function through stochastic gradient descent. The
choice of loss function (e.g., hinge loss for linear where:
SVM, log loss for logistic regression) and the • TP (True Positives): The number of correctly predicted
learning rate are crucial for the convergence and positive instances.
• FN (False Negatives): The number of actual positive
performance of the model. SGD’s ability to
instances that were missed by the model.
process large datasets and adapt quickly to new
3. F-Measure (F1 Score)
data makes it particularly suitable for applications
The F-Measure, or F1 Score, is the harmonic mean
involving extensive tweet data.
of precision and recall, providing a single metric that

7
balances both precision and recall. It is particularly 4. Cross-Validation: To ensure the robustness of
useful when dealing with imbalanced datasets where the model performance, cross-validation
one class is more frequent than the other. The F1 techniques such as k-fold cross-validation
Score is given by the formula: are used to evaluate the models on different
subsets of the data.
F measure = 2 ∗ precision ∗ recall
By thoroughly evaluating the models using these
precision + recall
metrics, one can assess their effectiveness in
4. Accuracy
accurately identifying cyberbullying tweets and
Accuracy measures the proportion of correctly
ensure that the chosen model performs well across
classified instances (both true positives and true
various dimensions of classification quality.
negatives) among all instances in the dataset. It
provides an overall assessment of the model's
performance. Accuracy is given by the formula:

Accuracy = TP+TN
TP + FP + TN
+FN
where:
• TP (True Positives) refers to the number of
correctly predicted cyberbullying tweets.
• TN (True Negatives) refers to the number of
correctly predicted non-cyberbullying Comparison of Algorithms with Count Vectorizer
tweets.
• FP(False Positives) refers to the number of
tweets incorrectly classified as
cyberbullying.
• FN (False Negatives) refers to the number of
cyberbullying tweets that were missed by the
model.
Evaluation Process
To evaluate the performance of each machine
learning model, the following steps are typically
undertaken:
1. Confusion Matrix Calculation: A confusion
matrix is generated to summarize the results
of the classification model. It includes counts
of true positives, true negatives, false
positives, and false negatives.
2. Metric Computation: Precision, recall, F1
Score, and accuracy are calculated based on Comparison of Algorithms with Term Frequency-Inverse Document
Frequency
the confusion matrix values.
3. Model Comparison: The calculated metrics
are used to compare the performance of USER INTERFACE DESIGN
different models and select the
The user interface (UI) design of the cyberbullying
bestperforming one for the task of
detection project is essential for delivering a userfriendly
cyberbullying detection. and efficient experience. The interface leverages Tkinter

8
for desktop application development, NLTK for natural 3. Visual Design
language processing tasks, and Streamlit for interactive a. Colour Scheme and Typography
web-based visualizations. The design aims to provide an • Colour Scheme: The application uses a coherent
intuitive interaction model and effective data colour scheme to enhance readability and visual
presentation. appeal. The colours are chosen to provide clear
1. Overview contrast and highlight important elements such as
The UI design focuses on simplifying the user results and charts.
experience by enabling users to input text, view analysis • Typography: Clear and legible fonts are selected
results, and understand data visualizations seamlessly. to ensure that text is easily readable.
The application is designed to be straightforward and Consistent use of font sizes and styles contributes
accessible, allowing users to perform cyberbullying to a professional and cohesive look.
detection efficiently. b. Interaction Design
2. Layout and Components a. Main • User Feedback: Tkinter provides visual
Interface (Tkinter) feedback for interactive elements, such as buttons
• Input Area: A text entry field where users can and input fields, through color changes or
type or paste tweets for analysis. This input field messages to indicate actions. Streamlit enhances
is central to the application, allowing users to interaction with real-time updates and feedback.
enter data easily. • Error Handling: Informative error messages and
• Submit Button: A button that users click to prompts guide users in case of invalid inputs or
initiate the analysis. Upon clicking, the text is processing issues. This helps users correct errors
processed, and results are generated. and proceed with their analysis.
• Results Display: A section that shows the 4. Implementation The UI is
classification results, indicating whether the implemented using:
tweet is identified as cyberbullying or not. This • Tkinter: For the desktop application interface,
area may also include additional details such as including input fields, buttons, and results
confidence scores or a brief explanation of the display.
result. • NLTK: For processing the input text and
b. Visualization and Analysis (Streamlit) performing natural language analysis, such as
• Graphs and Charts: Streamlit is used to create tokenization and lemmatization.
interactive visualizations, such as bar charts or • Streamlit: For creating interactive web-based
pie charts, displaying metrics like the distribution visualizations and displaying performance
of cyberbullying and noncyberbullying content. metrics and charts.
These visualizations help users interpret the 5. User Experience Considerations
analysis results more effectively. • Usability: The design emphasizes ease of use,
• Summary Statistics: This section presents key ensuring that users can interact with the application
performance metrics of the model, including intuitively without needing extensive guidance.
precision, recall, F1 score, and accuracy. • Accessibility: The UI is designed to be accessible,
Streamlit enables dynamic updates of these with features like adjustable font sizes and clear
metrics based on the latest analysis. navigation paths, to cater to users with different
c. Navigation and Accessibility needs.
• Navigation Menu: Tkinter’s menu system or By integrating Tkinter, NLTK, and Streamlit, the
Streamlit’s sidebar can be used to navigate cyberbullying detection project achieves a well-rounded
between different functionalities of the user interface that supports efficient interaction, data
application, such as input analysis, historical processing, and visualization.
data, and settings. This provides a streamlined
way to access various features.
• Responsive Design: While Tkinter is primarily
used for desktop applications, careful design
ensures that the UI remains responsive and
functional across different screen sizes and
resolutions.

9
 Chat Window
The chat window is a central component of the
• User Authentication and Login Window cyberbullying detection platform, designed to
• The cyberbullying detection platform includes a facilitate real-time interaction between users and
user authentication mechanism to ensure secure the system. This feature allows users to input text,
and personalized access to the application. This is view analysis results, and engage with the
facilitated through a login window, which serves platform in a conversational manner. The chat
as the entry point for users to access the platform's window enhances user experience by providing a
features. The design and implementation of this dynamic and intuitive interface for cyberbullying
login window are crucial for managing user detection and analysis.
sessions and safeguarding sensitive data.
• User Authentication and Login Window
• The cyberbullying detection platform includes a a. Chat Interface
user authentication mechanism to ensure secure • Text Input Field: A text entry field where users can
and personalized access to the application. This is type or paste their messages (tweets) for analysis.
facilitated through a login window, which serves This input field is designed to be easily accessible,
as the entry point for users to access the platform's allowing users to quickly enter text.
features. The design and implementation of this • Send Button: A button that users click to submit
login window are crucial for managing user their input for processing. The send button
sessions and safeguarding sensitive data. triggers the analysis of the entered text and
updates the chat window with the results.
• Message Display Area: A section that shows the
conversation history, including user inputs and
system responses. This area displays the results of
the cyberbullying detection, such as whether the
tweet is classified as cyberbullying or not, along
with any additional comments or analysis.
b. Interaction Flow
• User Input: Users enter their messages into the
text input field and click the send button. The
system processes the input using the natural
language processing (NLP) and machine learning
algorithms implemented in the platform.
• System Response: After processing the input, the
system generates a response that is displayed in
the message area. This response includes the
results of the analysis and any relevant
information, such as confidence scores or
feedback.
• Conversation History: The chat window
maintains a history of interactions, allowing users
to review previous inputs and responses. This
feature helps users track the results of their
analyses and provides context for ongoing
interactions.

10
 • Action Buttons: Each flagged message includes
action buttons that allow administrators to:
o Review: View the full content of the
flagged message and any associated
analysis or comments.
o Disable: Mark the content as
inappropriate and disable it from being
visible to users. This action helps in
managing and moderating content
effectively.
b. Review and Management
• Detailed View: Administrators can click on
individual flagged messages to access a detailed
view, including the full content, detection results,
and any notes or contextual information provided
by the system.
• Content Disabling: Administrators have the
option to disable inappropriate content, which
removes the flagged messages from user
Admin Page for Cyberbullying Detection Platform interactions and prevents them from being
The admin page is a critical component of displayed on the platform.
the cyberbullying detection platform, c. User Management
designed to provide administrators with tools • User Profiles: The admin page provides access to
to manage and monitor the platform's user profiles, allowing administrators to review
activities. This page offers functionalities to user activity and manage user permissions. This
review detected instances of cyberbullying, feature helps in identifying users who may be
disable inappropriate content, and oversee repeatedly involved in cyberbullying.
user interactions. The admin page enhances • Account Actions: Administrators can perform
the system's control and moderation actions such as suspending or deactivating user
capabilities, ensuring a safer and more accounts based on their behaviour or involvement
manageable environment. in flagged content.
3. Implementation
The admin page is implemented using a
combination of Tkinter for desktop-based
a. Admin Interface
management and Streamlit for web-based
• Dashboard Overview: The admin page includes
visualization and interaction:
an overview dashboard that summarizes key
• Tkinter: Provides the graphical interface for the
metrics, such as the number of flagged instances,
admin page, including the list of flagged content,
active users, and recent activities. This overview
action buttons, and user management tools.
helps administrators quickly assess the platform's
Tkinter’s widgets are used to create a functional
status.
and organized layout for administrators.
• Flagged Content List: A list or table displaying all
• Streamlit: Enhances the admin page with
chats or messages flagged by the system as
interactive elements and real-time updates.
potential instances of cyberbullying. This list
Streamlit’s capabilities are used to display
includes details such as the message content, user
metrics, update flagged content status, and
information, and the reason for flagging.
visualize data related to user interactions and flagged
messages.
4. Security and Access Control

11
• Authentication: Access to the admin page is restricted to
authorized personnel only. Administrators must log in with
special credentials to access the management tools and features.
• Data Security: Sensitive information displayed on the admin
page, such as user data and flagged content, is protected
through encryption and secure data handling practices.

12