Cyberbullying detection: a machine learning approach
Machine learning is a hot topic and it is widely implemented in software, web application and more. Those algorithms are used in the classification or regression model to predict an input. Nowadays, the cases of cyberbullying have been increasing over the years. It causes distress to those that are...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Year Project / Dissertation / Thesis |
| Published: |
2022
|
| Subjects: | |
| Online Access: | http://eprints.utar.edu.my/4698/1/fyp_CS_2022_YSY.pdf http://eprints.utar.edu.my/4698/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Machine learning is a hot topic and it is widely implemented in software, web application and more. Those algorithms are used in the classification or regression model to predict an input. Nowadays, the cases of cyberbullying have been increasing over the years. It causes distress to those that are involved, even though they are not hurt physically but they are mentally affected. Even though the social media sites have been taking measures to control the situation, and it helped to decrease the cyberbullying cases. However, it might not be enough because not every social media site has a cyberbullying detector machine. In this project, a model was created to classify the text as cyberbullying message or non-cyberbullying message. This model combines a rule-based approach of sentiment analysis and a supervised machine learning algorithm to classify the text. This model used sentiment analysis to label the datasets and these data are fed into the classifier for training. TextBlob was used to determine the polarity of the text. After labelling the data, these labels will act as the target feature for the model. Bag of Words model was used to convert text into numerical inputs. The machine learning algorithm, Support Vector Machine was chosen after comparing it with other algorithms such as Multinomial Naïve Bayes, Decision Tree Classifier, and Random Forest Classifier. The model has a high accuracy score, 0.93. The F1-score for both classes were high, 0.92 for non-cyberbullying class, 0.93 for cyberbullying class. Finally, the model was pickled and loaded into the web application. The web application was created to test the effectiveness of the model, it would simulate the process of cyberbullying that will occur in a social media site. |
|---|