Hate Speech on Social Media – The Possibility of Automatic Detection and Elimination

The article deals with the issues of hate speech and other forms of verbal aggression on the Internet as well as the possibility of their automatic detection. The paper discusses the studies confirming the partial effectiveness of text mining methods in the automatic detection of hate speech on social media. Hate speech is related to verbal aggression resulting from belonging to a group (national, racial, religious, etc.) and has become a significant problem in the social and economic context. Automatic detection significantly support the management of online news websites and social media due to the moderation of the received content. Moreover, eliminating online hate speech reduces its negative social and economic effects. The linguistic and cultural specificity of the hate speech are the problem, and the gap so far is solving the problem in Polish conditions. The study used the Tweeter database. Then, methods such as artificial neural networks, naïve Bayes classifier and support vector machine were used. The obtained results confirm the thesis about the possibility of using text mining methods in the process of reducing hate speech, but at the moment the described methods do not allow for full automation of the elimination of such content. The issue was presented in the article primarily in the context of the significance and scale of the problem and the possibility of solving it, and less from the point of view of the technical details.

JEL: L86, L82, C40, O30

The article deals with the issues of hate speech and other forms of verbal aggression on the Internet as well as the possibility of their automatic detection. The paper discusses the studies confirming the partial effectiveness of text mining methods in the automatic detection of hate speech on social media. Hate speech is related to verbal aggression resulting from belonging to a group (national, racial, religious, etc.) and has become a significant problem in the social and economic context. Automatic detection significantly support the management of online news websites and social media due to the moderation of the received content. Moreover, eliminating online hate speech reduces its negative social and economic effects. The linguistic and cultural specificity of the hate speech are the problem, and the gap so far is solving the problem in Polish conditions. The study used the Tweeter database. Then, methods such as artificial neural networks, naïve Bayes classifier and support vector machine were used. The obtained results confirm the thesis about the possibility of using text mining methods in the process of reducing hate speech, but at the moment the described methods do not allow for full automation of the elimination of such content. The issue was presented in the article primarily in the context of the significance and scale of the problem and the possibility of solving it, and less from the point of view of the technical details.

JEL: L86, L82, C40, O30