Detection of Cyberbullying on Social Media Using
Machine Learning
Under the Guidance of Presented by
Mr. [Link] Sekhar [Link] [Link] Kusuma Priya 216M1A05A9
Assistant Professor [Link] 216M1A05A0
[Link] Datta 216M1A0593
[Link] 216M1A0570
Detection of Cyberbullying
on Social Media Using
Machine Learning
Cyberbullying is a significant issue affecting teenagers and adults,
leading to negative outcomes like suicide and depression. Regulating
content on social media platforms is increasingly necessary. This
study uses data from hate speech tweets and personal attacks from
Wikipedia to build a model for detecting cyberbullying in text data
using Natural Language Processing and Machine Learning.
The study explores three feature extraction methods and four
classifiers to identify the best approach. The model achieves
accuracies above 90% for tweet data and above 80% for Wikipedia
data.
Existing Cyberbullying Detection Systems
Keyword Matching & Opinion Mining Troll Identification
Hsien used keyword matching, opinion mining, and social Gal´an-Garc´ıa et al. proposed identifying trolls by linking
network analysis, achieving a precision of 0.79 and recall fake profiles to real ones, using machine learning to
of 0.71 using datasets from four websites. analyze tweets and profile features, achieving 68%
accuracy.
These existing systems have limitations, such as the inability to detect trolls without real profiles or those who can
change writing styles to avoid detection. More efficient algorithms are needed to address these shortcomings.
Disadvantages of Existing
Systems
1 Limited Vocabulary
Existing systems often fail to design a comprehensive
vocabulary from all available documents, which can limit their
effectiveness.
2 Tf-Idf Similarity
The Tf-Idf method shares similarities with the bag-of-words
model, using the same approach to create a vocabulary for
feature extraction, which can be a disadvantage.
These limitations highlight the need for improved methods in
cyberbullying detection to overcome the shortcomings of current
approaches.
Proposed System: Binary
Classification
Hate Speech Detection
Detecting hate speech on Twitter.
Personal Attack Detection
Identifying personal attacks on Wikipedia.
The proposed system addresses cyberbullying detection as a binary
classification problem, focusing on hate speech on Twitter and
personal attacks on Wikipedia. The goal is to classify text data as
either containing cyberbullying or not.
Tokenization and Regular
Expressions
Tokenization
Splitting raw text into meaningful words or tokens. For
example, "we will do it" becomes 'we', 'will', 'do', 'it'.
Regex Tokenizer
Tokens are decided based on a rule, which in this case is
a regular expression. Tokens matching the regular
expression '\w+' are extracted.
Tokenization is a crucial step in Natural Language Processing,
breaking down text into manageable units for analysis. The use of
regular expressions allows for flexible and precise token extraction.
Stemming and Stop Word Removal
Porter Stemmer
2 Using PorterStemmer for stemming.
Stemming
1
Converting words into their root
form. For example, 'eating', 'eats',
'eaten' become 'eat'. Stop Word Removal
Removing irrelevant words like
'what', 'is', 'at', 'a' to improve model
3
performance.
Stemming and stop word removal are essential for refining text data, reducing redundancy, and improving the efficiency
of machine learning models.
Advantages of the
Proposed System
Contextual Prediction Semantic Nuance
Common Bag of Words (CBOW) CBOW model can capture
model takes multiple words as multiple semantics for a single
input and predicts the word word, such as different meanings
based on the context. of "Apple".
The proposed system leverages the CBOW model to provide
contextual predictions and capture semantic nuances, enhancing the
accuracy of cyberbullying detection.
System Requirements
Hardware Processor: Pentium –IV, RAM: 4
GB (min), Hard Disk: 20 GB,
Keyboard, Mouse, Monitor:
SVGA
Software OS: Windows 7 Ultimate,
Coding Language: Python,
Front-End: Python, Back-End:
Django-ORM, Designing: Html,
css, javascript, Data Base:
MySQL (WAMP Server)
The system requires standard hardware components and utilizes
Python for coding, Django-ORM for the back-end, and MySQL for the
database. HTML, CSS, and JavaScript are used for designing the front-
end.
Thank you