E3S Web of Conferences 507, 01069 (2024) https://doi.org/10.
1051/e3sconf/202450701069
ICFTEST-2024
Detecting cyberbullying in social media using
text analysis and ensemble techniques
Y. Jeevan Nagendra Kumar1*, Rohith Reddy Vanapatla1, Vamshi Krishna Pinamoni1,
Jaswanth Kandukuri1, Muntather Almusawi2, Aravinda K3, Lavish Kansal4 and Ravi Kalra5
1Department of Information Technology, Gokaraju Rangaraju Institute of Engineering and
Technology, Hyderabad, India.
2The Islamic university, Najaf, Iraq
3Department of Electronics and Communication Engineering, New Horizon College of Engineering,
Bangalore, Karnataka, India.
4Lovely Professional University, Phagwara, Punjab, India
5Lloyd Institute of Engineering & Technology, Knowledge Park II, Greater Noida, Uttar Pradesh,
India.
Abstract. In the dynamic landscape of our hyper-connected digital world,
social media platforms play a dual role as facilitators of global interaction
and breeding grounds for harmful behaviors. Cyberbullying, an insidious
online menace, inflicts emotional distress and psychological trauma on
numerous individuals, underscoring the urgent need for advanced tools to
detect and prevent such malevolent actions. This innovative project
harnesses the power of artificial intelligence and text analysis to illuminate
the dark corners of social media where cyberbullying thrives, offering hope
to countless victims. At its core, this endeavor utilizes cutting-edge
ensemble techniques, a fusion of diverse machine learning algorithms, to
analyze textual content across social media platforms. This approach ensures
unparalleled accuracy in identifying and flagging cyberbullying instances,
enhancing the efficiency of the detection process while minimizing false
positives. The project adopts a multifaceted approach to text analysis,
examining explicit language, sentiments, context, and behavioral patterns in
online interactions. By delving into the intricacies of human communication,
the system distinguishes between genuine expressions and malicious intent,
providing a nuanced and accurate assessment.
1 Introduction
The project "Detecting Cyberbullying on Social Media using Text Analysis and Ensemble
Techniques" is a pioneering initiative addressing escalating concerns about cyberbullying in
the era of widespread social media usage. Acknowledging the dual nature of social platforms,
the project employs ensemble techniques, such as Random Forests and Gradient Boosting, to
© The Authors, published by EDP Sciences. This is an open access article distributed under the terms of the Creative
Commons Attribution License 4.0 (https://creativecommons.org/licenses/by/4.0/).
E3S Web of Conferences 507, 01069 (2024) https://doi.org/10.1051/e3sconf/202450701069
ICFTEST-2024
enhance the accuracy of cyberbullying detection models and mitigate overfitting risks.By
incorporating text analysis, a component of natural language processing (NLP), the project
goes beyond keyword detection. It explores linguistic patterns, sentiment analysis, and
contextual understanding, providing a nuanced perspective crucial for distinguishing
between harmless communication and potential cyberbullying instances.
The societal impact is significant, showcasing technology's proactive role in safeguarding
digital spaces and fostering positive online interactions. As social media continues to be
integral to our lives, the project emphasizes the responsibility of technological innovation to
contribute to the well-being of digital communities. It underscores the need for ethical
considerations, promoting an environment where users can engage online without fear or
intimidation.
In conclusion, the project goes beyond technical aspects, representing a conscientious
effort to responsibly leverage technology for the greater good. It serves as a beacon for the
ethical application of machine learning and NLP, setting a positive precedent for digital
solutions that prioritize user safety, inclusivity, and the development of a respectful online
community.
2 Literature Survey
[1] This research delves into cyberbullying detection on Twitter, employing ensemble
stacking learning and a customized BERT model (BERT-M). Using a Twitter dataset, the
study preprocessed the data and utilized word2vec-CBOW for feature extraction. The stacked
model exhibited impressive performance, showcasing high precision (0.950), recall (0.92),
and F1-score (0.964), along with swift detection speed (3 minutes). It outperformed standard
BERT and other NLP detectors, demonstrating its efficacy in combatting cyberbullying on
social media.
[2] This conduct contributes to a hostile online environment, resulting in various forms of
harassment such as privacy violations and sexual insults, impacting individuals globally.
Academic attention is growing towards identifying bullying behaviors in text. This study
aims to utilize machine learning and natural language processing to accurately detect online
bullying. An algorithm was developed and utilized to analyze and authenticate hostile
comments.
[3] This study highlights the insufficient awareness of netiquette and security among users,
particularly in social media, where platforms like Twitter have become prevalent for
Cyberbullying. Through Sentiment Analysis, this paper aims to discern between positive and
negative sentiments expressed in tweets using Machine Learning algorithms. Its primary
objective is to alleviate the emotional, mental, and physical toll caused by Cyberbullying.
[4] The review of literature delves into cyberbullying detection, particularly concerning the
prevalent technology usage among young individuals. It sheds light on how online
communication and social networking expose teenagers to bullying. The study proposes
methodologies that leverage supervised learning techniques to spot cyberbullying through
language pattern analysis. It underscores the growing acknowledgment of cyberbullying's
influence on youth and the application of machine learning for automated detection of
2
E3S Web of Conferences 507, 01069 (2024) https://doi.org/10.1051/e3sconf/202450701069
ICFTEST-2024
bullying content. With a dataset obtained from Kaggle containing a significant volume of
bullying-related content, the research conducts model training and validation.
[5] The survey of literature highlights the surge in cyberbullying through harmful online
content, significantly affecting the mental health of young individuals. Present machine
learning models lack a comprehensive feature set essential for efficient cyberbullying
detection. The study introduces a bidirectional deep learning model based on BERT,
integrating diverse features crucial for precise identification of cyberbullying. Its goal is to
alleviate the adverse impacts of cyberbullying.
[6] This study delves into cyberbullying detection across various social media platforms,
employing three approaches and five distinct models on a diverse dataset. By augmenting
Support Vector Machines, leveraging DistilBERT, and incorporating ensemble methods, it
explores their effectiveness. [7] The ensemble models consistently outperform individual
ones across most evaluation metrics, achieving the highest accuracy at 89.6%. DistilBERT
notably showcases its effectiveness with the highest precision at 91.17%. [8] Moreover,
enhancing feature granularity contributes to improved performance compared to basic TF-
IDF techniques.
3 Methodology
1) Dataset
The dataset utilized in this project serves as a foundational element for the development of
an effective cyberbullying detection model. Comprising 47,692 records with two essential
columns “tweet_text” and “cyberbullying_type”. The dataset was thoughtfully curated from
Kaggle, a reputable platform for data science and machine learning resources. The "text"
column contains textual data extracted from various online sources, while the "type of text"
column categorizes each entry into specific cyberbullying categories such as
`not_cyberbullying`, `religion`, `ethnicity`, `age`, `gender`, and `other_cyberbullying`. This
structured dataset not only facilitates the training and evaluation of the machine learning
model but also ensures a diverse representation of cyberbullying instances.
2) Architecture:
Fig. 1. System Architecture
3
E3S Web of Conferences 507, 01069 (2024) https://doi.org/10.1051/e3sconf/202450701069
ICFTEST-2024
3)Workflow:
A. Importing Libraries:
In this step, we bring in the necessary tools to work with data and perform analysis.`NumPy`
and `Pandas` are used for data manipulation and analysis. `Matplotlib` and `Seaborn` are for
data visualization. `Plotly Express` is employed for interactive and expressive visualizations.
`demoji` is a library for handling emojis in text data.
B. Loading Dataset:
We read our dataset, ‘cyberbullying_tweets.csv,’ into a Pandas DataFrame named `df`.
`df.head()` shows the first few rows of the dataset, and `df.info()` provides a summary of the
dataset’s structure.
C. Data Preprocessing:
Removing Hashtags, Mentions, and URLs.Making text lowercase,Stemming,Removing
punctuation,Removing stopwords,Handling Emojis,Combining Text and Applying Cleaning
Function
D. Exploratory Data Analysis (EDA):
We use visualization tools (`Seaborn`) to explore the distribution of different types of
cyberbullying in our dataset. The `sns.countplot` visualizes the count of each type, giving us
insights into the distribution.
E. Text Analysis and Feature Extraction:
We use the TF-IDF (Term Frequency-Inverse Document Frequency) vectorizer to convert
text into a numerical format that machine learning models can understand. Dimensionality
reduction is performed using PCA (Principal Component Analysis) to reduce the number of
features while retaining 90% of the variance.
Fig. 2. PCA
4
E3S Web of Conferences 507, 01069 (2024) https://doi.org/10.1051/e3sconf/202450701069
ICFTEST-2024
F. Model Training:
We train several machine learning models, including Logistic Regression, Support Vector
Machines, Neural Networks, Random Forests, Gradient Boosting, and Naive Bayes. Grid
search with cross-validation is used to find the best hyperparameters for each model.The
performance of each model is evaluated using classification reports and confusion matrices.
G. Model Evaluation:
The trained models are evaluated using the `classification_report` function, which provides
precision, recall, and F1-score for each class. Confusion matrices (`plot_confusion_matrix`)
visually represent the model's performance in terms of true positive, true negative, false
positive, and false negative predictions.
Fig. 3. Classification Report
Fig. 4. Confusion Matrix
5
E3S Web of Conferences 507, 01069 (2024) https://doi.org/10.1051/e3sconf/202450701069
ICFTEST-2024
H. Pipeline Creation:
The Random Forest classifier, identified as the best-performing model, is integrated into a
machine learning pipeline.This pipeline includes the TF-IDF vectorizer, ensuring a
streamlined process from data preprocessing to model training.
I. Model Serialization:
The entire pipeline, which encapsulates the preprocessing steps and the trained Random
Forest model, is saved to a file named 'CBDmodel1.pkl.'Serialization allows us to reuse the
trained model without going through the training process again.
4 Results
The cyberbullying detection model, employing ensemble techniques and text analysis,
demonstrates impressive accuracy in identifying online harassment. Its nuanced approach
distinguishes between harmless communication and potential cyberbullying instances. The
model's use of Random Forests and Gradient Boosting minimizes overfitting risks, ensuring
robust performance. Beyond technical success, the model significantly contributes to positive
online engagements, reflecting a responsible application of technology. Results highlight a
comprehensive understanding of linguistic patterns and sentiments in social media
interactions.
Fig. 5. UI
6
E3S Web of Conferences 507, 01069 (2024) https://doi.org/10.1051/e3sconf/202450701069
ICFTEST-2024
Fig. 6. Sample Output
5 Conclusion
In concluding our project on detecting cyberbullying in social media, we've embarked on a
comprehensive journey integrating text analysis, machine learning, and ensemble techniques.
Our objective extends beyond creating a one-time solution; it's about crafting a robust system
capable of adapting to the dynamic landscape of online interactions. Anchored in ethical
considerations, our approach prioritizes responsible and equitable cyberbullying detection.
In essence, our project is not just a technological achievement; it's a commitment to
fostering a safer and more inclusive digital environment. By combining technological
innovation with ethical considerations and a commitment to ongoing improvement, we aim
to contribute meaningfully to the ongoing efforts to combat cyberbullying and promote
positive online interactions. This project stands as a testament to the power of
interdisciplinary approaches in addressing complex societal issues within the digital
landscape. Furthermore, as we reflect on the impact of our project, we recognize the
imperative of knowledge-sharing and collaboration. Our commitment extends to providing
insights and resources that empower other researchers, developers, and stakeholders in the
collective fight against cyberbullying. By fostering a culture of shared learning and
cooperation, we aspire to create a ripple effect that amplifies the positive influence of our
work.
References
1. Bozzola, E.; Spina, G.; Agostiniani, R.; Barni, S.; Russo, R.; Scarpato, E.; Di Mauro, A.;
Di Stefano, A.V.; Caruso, C.; Corsello, G.The use of social media in children and
7
E3S Web of Conferences 507, 01069 (2024) https://doi.org/10.1051/e3sconf/202450701069
ICFTEST-2024
adolescents: Scoping review on the potential risks. Int. J. Environ. Res. Public Health
2022, 19, 9960. [CrossRef] [PubMed].
2. P. Nagaraj, Dr. M. Venkat Dass, E. Mahender “Breast Cancer Risk Detection Using
XGB Classification Machine Learning Technique “, IEEE International Conference on
Current Development in Engineering and Technology (CCET)-2022, Sageuniversity,
Bhopal, India, 23-24, Dec 2022.
3. Vyawahare, M., & Chatterjee, M., (2020), “Taxonomy of cyberbullying detection and
prediction techniques in online social networks”, In L. C. Jain, G. A. Tsihrintzis, V. E.
Balas, and D. Sharma (Eds.), Data communication and networks (pp. 21–37). Springer.
https://doi.org/10.1007/978-981-15-0132-6_3.
4. Chen, J., Yan, S., & Wong, K.-C., (2018), “Verbal aggression detection on Twitter
comments: convolutional neural network for short-text sentiment analysis”, Neural
Computing and Applications. https://doi.org/10.1007/s00521-018-3442-0.
5. S. Hnduja and J. W. Patchin “Cyberbullying: Identification, Prevention, &
Response,”Cyberbullying Res. Cent, no. October, pp. 1-9,2018.
6. Dr. Vijayakumar V and Dr Hari Prasad D , “Intelligent Chatbot Development for Text
based Cyberbullying Prevention” International Journal of New Innovations in
Engineering and Technology,2021.
7. G. A. León-Paredes et al., Presumptive Detection of Cyberbullying on Twitter through
Natural Language Processing and Machine Learning in the Spanish
Language,CHILECON pp. 1-7, doi: 10.1109/CHILECON47746.2019.8987684. (2019).
8. Murshed, A.H., Abawajy, J., Mallappa, S., Saif, M.A.N., Al-Ariki, H.D.E. (2022). DEA-
RNN: A Hybrid Deep Learning Approach for Cyberbullying Detection in Twitter Social
Media Platform. IEEE Access, 10: 25857
25871.https://doi.org/10.1109/ACCESS.2022.31536.
