Countercraft: counter-speech retrieval with template-based ingenuity

Abstract

To address the issue of hate speech, the study proposes a novel method to retrieve counter speeches using templates. The templates are created by analyzing instances of hate speech in the CONAN dataset, which contains examples of online conversations that include hate speech. The similarity analysis approach compares new instances of hate speech to those already present in the dataset. Our experiments identified intent and implication as factors that can help retrieve the right hate speech from the training dataset whose counterspeech can be used for the input hate speech. In conclusion, the study proposes an innovative method to combat hate speech by retrieving counter speeches using templates.