Profiling Spreaders of Hate Speech with N-grams and RoBERTa

Christopher Bagdon
2021 Conference and Labs of the Evaluation Forum  
This paper outlines our approach to the 2021 CLEF Conference Shared Task, Profiling Hate Speech Spreaders on Twitter. Our approach uses the probability output of a logistic regression classifier and a RoBERTa based classifier as features for a linear support vector classifier. During a final cross validation analysis the Spanish meta-classifier performed better than any other single classifier. For English the meta-classifier performed slightly worse than the RoBERTa classifier. On the test set
more » ... our system performed moderately well in comparison to other submissions, with 81% accuracy for Spanish and 67% for English. Overall our system placed 15 th of 66 entries.
dblp:conf/clef/Bagdon21 fatcat:qq6pffairrdhlb6ytuqpdecsze