Debunking Online Reputation Rumours Using Hybrid of Lexicon-Based and Machine Learning Techniques

Abstract

The scale and scope of virality of a malicious rumour hurt the victim more as so many people see it quickly and it is also hard to respond to it. Quintessentially, reputation rumour is a tool for denigration bullying which can sully reputations and ruin the image of a person or an entity in public. This work is the primary study to demonstrate a correlation between denigration cyber-bullying and rumour. A novel ReputeCheck model is put forward to detect and debunk rumours sourced online on Twitter about global celebrities and world leaders/politicians. The model uses a lexicon to check for the presence of derogatory words and extracts message-based and user-account-based features to train a learning model. Supervised learning is used to train and test the model on a data set created from the Twitter post. A performance accuracy of 83.4% using support vector machine with Gaussian radial basis function kernel is observed. This study validates the vulnerabilities of denigration associated with a public figure or celebrity are the highest and rumour detection can be used as regulatory mechanism to inhibit the production, dissemination and impact of hateful messages online.