Classification of the Offensive Tweets on Twitter Based on s127 of the Communications Act 2003 in the UK

Abstract

Social media sites such as Facebook, Twitter, YouTube, etc. have been used globally in recent years. Twitter is one of the most popular networking sites that allow individuals to write and share their opinions, regardless of whether they are positive or negative. However, posting illigal comments has many disadvantages for Twitter users, which motivates the law authorities in the UK to employ an automatic classifier tool that can detect any aggressive comments on Twitter. In this dissertation, I use a supervised machine learning approach for the classification process, and aim to solve the issue of having such these bad comments on Twitter by experimenting with different useful techniques in order to determine which one was the best. The dataset created for this dissertation was a combination of two existing datasets; the first dataset’s instances were labelled in advance as offensive posts in accordance with s127 of the Communication Act 2003 in the UK, and the other dataset’s inputs were classified as “normal” tweets. The results show that using an ensemble classifier of various baseline classifiers, with the BoW method as a feature technique, improves the performance – as high as 99% or 100%. This tool would help the police to prosecute the authors of offensive tweets and save online communication from these types of crimes in the future.