Henry Jackson Undergraduate Dissertation 2014/15

Satire Detection

Supervised by M.Stevenson

Abstract

Many people rely on the internet as a source of news. This can cause issues when they come to satirical news articles, designed for humour or to send a political message, but with little or no factual basis. There have been a number of proposed solutions to this problem, from utilising the sentiment in the articles (Riloff et al., 2013) to the use of big data, such as in Burfoot and Baldwin's work on satire detection (2009).

This work builds on Burfoot and Baldwin's satire detection experiments, testing the usefulness of their additional features by combining them to try and maximise the success. The effects of bias in their data set were also investigated. It was found that their 'semantic validity' feature significantly improved the results of a bag-of-words model, although the effects of combining the features showed no significant improvement. It was also found that reducing the bias in the data set improved the system.