Effective Cyberbullying Detection with SparkNLP
A Rishab Vanigotha, M R Naveen Kumar, Shraddha B. Hiremath, Sujay Sukumaran Adityan, M John Basha · International Journal for Research in Applied Science and Engineering Technology · 2023
Abstract: People of all ages are affected by cyberbullying, which has become a serious problem that can have negative effects like depression and even suicide. As a result, social media content regulation is becoming more and more necessary. Our research project uses SparkNLP, a potent and scalable natural language processing library, to address the problem of cyberbullying. A model for detecting cyberbullying was developed by us using a dataset of 48,000 tweets from Kaggle for training. We used the Universal Sentence Encoder (USE) to extract text data, and ClassifierDL, which employs deep neural networks and uses USE as an input for text classification, was used to build the classification model. Class imbalance is a problem, so we used text augmentation to address it. Age, ethnicity, gender, religion, other cyberbullying, and not cyberbullying are six categories into which our methodology accurately detects and categorizes cyberbullying. We evaluated the model on a held-out set of data to assess its accuracy and resilience, and it produced remarkable results of 93.34% on training data and 89% on test data. The goal of our project is to support and reinforce the ongoing initiatives that are working to stop cyberbullying and to advance a secure and positive online environment for everyone.