OffendES

OffendES is a large-scale annotated corpus for offensive language analysis in Spanish, developed by the SINAI research group at the Universidad de Jaén and presented at RANLP 2021. It collects over 50,000 comments from YouTube, Instagram, and Twitter, gathered from posts directed at well-known young Spanish influencers. Comments were manually annotated via Amazon Mechanical Turk using a four-category schema that captures both the nature and target of offensive content: offensive targeting an individual (OFP), offensive targeting a group (OFG), non-offensive with expletive language (NOE), and non-offensive (NO).
The corpus is distributed in two subsets: a 3-annotator subset (44,951 instances) for multiclass classification, and a 10-annotator subset (13,706 instances) that additionally supports multioutput regression using annotator confidence values. Each instance includes metadata on source platform, influencer, and influencer gender.