Abusive content detection and classification system for online user generated data

Abstract

This thesis provides a comprehensive analysis of different forms of abusive content across various digital platforms, delving into its nature, prevalence, challenges encountered and impact on the users. The abusive behavior has been addressed under different categories in past including offensive content, hate speech, cyberbullying insults etc. Various Machine learning (ML) algorithms like Naïve Bayes (NB), Logistic Regression, Support Vector Machine (SVM) Random Forest etc. have been widely employed for over a decade for this task, leveraging the features from various categories based on content, user activity based, social network and sentiment etc. to enhance the efficiency of classifiers. Given the labour-intensive nature of feature engineering, there is a prevailing inclination towards incorporating deep learning models. This research study applies deep learning models to identify various types of abusive language on a Twitter dataset. The input data is classified into 3 groups: offensive language detection (Offensive or not Offensive), offensive language categorization (Targeted vs non-targeted), and offensive language target identification (Individual, group or others). The data undergoes various stages including preprocessing, tokenization and feature selection, prior to being input into a classifier. In the case of classifiers, the efficiency of five deep learning models, namely, Gated Recurrent Unit, Bidirectional Gated Recurrent Unit, Multi-Dense Long Short-Term Memory, Bidirectional Long Short-Term Memory, and Long Short-Term Memory, have been examined. These RNNs (Recurrent neural network) models are capable of handling long-term dependencies in the text. These models have the unique ability to maintain a hidden state, allowing the network to capture sequential dependencies by remembering previous inputs while processing.

Description

Keywords

Citation

item.page.endorsement

item.page.review

item.page.supplemented

item.page.referenced