This paper addresses several of the key issues facing creation of a classifier for hate-speech on forums, blogs, or other areas of web discourse.
This article deals with a supervised machine learning text classifier, trained and tested to distinguish between hateful and/or antagonistic response with a focus on race, ethnicity or religion; and more general responses. This study operated by using human annotated data collected from Twitter in the immediate aftermath of the Lee Rigby’s murder in 2013, to train and test the classifier. Rigby’s murder led to an explosive public social media reaction, and with the notable terrorist motive behind his death in addition to the public actions that accompanied it, a public response laden with emotion was inevitable. This study also discusses the results of the classifier, and its ability to optimally catch and flag heated trends in social media. Overall, the study achieves an F-measure of 0.95 using features derived from the content of each tweet, including syntactic dependencies between terms to recognise “othering” terms, incitement to respond with antagonistic action, and claims of well-founded or justified discrimination against social groups.
This article will be useful to researchers looking to build models using probabilistic, rule-based and spatial-based classifiers to forecast the likely spread of hate speech.