BERT's Auxiliary Sentence focused on Word's Information for Offensiveness Detection

This paper describes the participation of the CIMAT-GTO team in the MeOffendEs 2021 competition. Our main goal is to evaluate an auxiliary sentence scheme for classification with BERT in the offensiveness identification task. The use of the auxiliary sentence has been shown to increase the efficiency of classifiers based on pre-trained BERT models in various tasks. We propose two new approaches to obtain the auxiliary sentence, the objective of the proposals is to remark the available information on the use of the words along the classes in the training corpus. The proposals S2KNNC and S2ChiN use techniques related to Nearest Neighbor and Attribute Selection by Chi-square, respectively, to construct the auxiliary sentence. Our results indicate that the auxiliary sentence scheme allows to improve the performance of the BERT-based classifier or even BERT classifier ensembles.