Semantic analysis of offensive language categories from existing annotated corpora

There exists a vast amount of different offensive language corpora for English language, annotation criteria and category naming. In this paper, we explore 21 categories language. We use natural processing techniques to find correlations between the based on seven data sets. employ several traditional (TF–IDF) advanced (fastText, GloVe, Word2Vec, BERT, other deep NLP methods) uncover similarities among categories. The findings reveal that most are densely interconnected, while two-level hierarchical representation them can be provided. also transfer analysis Slovenian compare both researched languages.