The 3DSeTwitch corpus – A three-dimensional corpus annotated for sexist phenomena

In this paper we present our work on the creation of the 3DSeTwitch corpus, a multimodal corpus aligning the representation of chats, audios and videos from Twitch, annotated for hate speech phenomena. Twitch is a platform for sharing live multimedia streaming experiences. It brings together internet users who interact live with each other – textually via chat, visually via video and verbally via audio. This Platform addresses cyber violence and sexist hate in particular. The creation of the corpus follows these different stages: (i) the data collection; (ii) the automatic extraction of data and metadata; (iii) the manual identification of samples associated with sexist language; (iv) the representation of the corpus with the CMC-core scheme; (v) the original annotation scheme of explicit and implicit sexist discourse; (vi) the inter-annotator agreement, carried out using a perspectivist approach.