Scribd
Upload
137 views

Detection of Cyberbullying On Social Media Using Machine Learning

The document discusses detecting cyberbullying on social media using machine learning. It summarizes an existing system that uses keyword matching, opinion mining and social network analysis to detect cyberbullying with 79% precision and 71% recall. The proposed system classifies tweets and Wikipedia comments as containing cyberbullying or not using natural language processing techniques like tokenization, stemming, stop word removal and machine learning classifiers, achieving over 90% accuracy for tweets and over 80% for Wikipedia comments. It describes the technical approach and system requirements to implement the proposed cyberbullying detection system.

Uploaded by

SLDFLAG
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
137 views

Detection of Cyberbullying On Social Media Using Machine Learning

The document discusses detecting cyberbullying on social media using machine learning. It summarizes an existing system that uses keyword matching, opinion mining and social network analysis to detect cyberbullying with 79% precision and 71% recall. The proposed system classifies tweets and Wikipedia comments as containing cyberbullying or not using natural language processing techniques like tokenization, stemming, stop word removal and machine learning classifiers, achieving over 90% accuracy for tweets and over 80% for Wikipedia comments. It describes the technical approach and system requirements to implement the proposed cyberbullying detection system.

Uploaded by

SLDFLAG
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 5

Detection of Cyberbullying on Social Media Using

Machine learning
ABSTRACT

Cyberbullying is a major problem encountered on internet that affects teenagers and


also adults.It has lead to mishappenings like suicide and depression. Regulation of
content on Social media platorms has become a growing need. The following study
uses data from two different forms of cyberbullying, hate speech tweets from
Twittter and comments based on personal attacks from Wikipedia forums to build a
model based on detection of Cyberbullying in text data using Natural Language
Processing and Machine learning. Three methods for Feature extraction and four
classifiers are studied to outline the best approach. For Tweet data the model
provides accuracies above 90% and for Wikipedia data it gives accuracies above
80%.

EXISTING SYSTEM

 Hsien[1] used an approach using keyword matching, opinion mining and


social network analysis and got a precision of 0.79 and recall of 0.71 from
datasets from four websites.Patxi Gal´an-Garc´ıa et al.[2] proposed a
hypothesis that a troll(one who cyberbullies) on a social networking sites
under a fake profile always has a real profile to check how other see the fake
profile. They proposed a Machine learning approach to determine such
profiles. The identification process studied some profiles which has some kind
of close relation to them.

 The method used was to select profiles for study, acquire information of
tweets, select features to be used from profiles and using ML to find the author
of tweets. 1900 tweets were used belonging to 19 different profiles. It had an
accuracy of 68% for identifying author. Later it was used in a Case Study in a
school in Spain where out of some suspected students for Cyberbullying the
real owner of a profile had to be found and the method worked in the case. The
following method still has some shortcomings.

 For example a case where trolling account doesnt have a real account to fool
such systems or experts who can change writing styles and behaviours so that
no patterns are found . For changing writing styles more efficient algorithms
will be needed.

 Mangaonkar et al. [3] proposed a collaborative detection method where there


are multiple detection nodes connected to each other where each nodes uses
either different or same algorithm and data and results were combined to
produce results. P. Zhou et al.[4] suggested a B-LSTM technique based on
concentration. Banerjee et al.[5]. used KNN with new embeddings to get an
precision of 93%.

Disadvantages
 A vocabulary is not designed from all the documents. The vocabulary may
consist of all words (tokens) in all documents or some top frequency tokens
 Tf-Idf method is not similar to the bag of words model since it uses the same
way to create a vocabulary to get its features.

PROPOSED SYSTEM

Cyberbullying detection is solved in this project as a binary classification problem


where we are detecting two majors form of Cyberbullying: hate speech on Twitter
and Personal attacks on Wikipedia and classifying them as containing Cyberbullying
or not.
 Tokenization: In tokenization we split raw text into meaningful words or tokens.
For example, the text “we will do it” can be tokenized into ‘we’, ‘will’, ‘do’, ‘it’.
Tokenization can be done into words called word tokenization or sentences called
sentence tokenization. Tokenization has many more variants but in the project we
use Regex Tokenizer. In regex tokenizer tokens are decided based on rule which in
the case is a regular expression. Tokens matching the following regular expression
are chosen Eg For the regular expression ‘\w+’ all the alphanumeric tokens
are extracted.

 Stemming: Stemming is the process of converting a word into a root word or stem.
Eg for three words ‘eating’ ‘eats’ ‘eaten’ the stem is ‘eat’. Since all three branch
words of root ‘eat’ represent the same thing it should be recognized as similar.
NLTK offers 4 types of stemmers: Porter Stemmer, Lancaster Stemmer, Snowball
Stemmer and Regexp Stemmer. The following project uses PorterStemmer.

 Stop word Removal: Stop words are words that do not add any meaning to a
sentence eg. some stop words for english language are: what, is, at, a etc. These
words are irrelevant and can be removed. NLTK contains a list of english stop words
which can be used to filter out all the tweets. Stop words are often removed from the
text data when we train deep learning and Machine learning models since the
information they provide is irrelevant to the model and helps in improving
performance.
Advantages
 Common Bag of Words model takes as input of multiple words and predicts
the word based on the context. Input can be one word or multiple words.
 CBOW model takes a mean of context of input words but two semantics can
be clicked for a single word. i.e. two vector of Apple can be predicted. First is
for the firm Apple and next is Apple as a fruit.
SYSTEM REQUIREMENTS

➢ H/W System Configuration:-

➢ Processor - Pentium –IV


➢ RAM - 4 GB (min)
➢ Hard Disk - 20 GB
➢ Key Board - Standard Windows Keyboard
➢ Mouse - Two or Three Button Mouse
➢ Monitor - SVGA

SOFTWARE REQUIREMENTS:

 Operating system : Windows 7 Ultimate.

 Coding Language : Python.

 Front-End : Python.

 Back-End : Django-ORM

 Designing : Html, css, javascript.

 Data Base : MySQL (WAMP Server).

You might also like