ur-iw-hnt at GermEval 2021: An Ensembling Strategy with Multiple BERT Models

TL;DR

This paper presents an ensembling approach using multiple BERT models for classifying comments in GermEval 2021, demonstrating that ensemble models outperform individual models across toxic, engaging, and fact-claiming comment detection tasks.

Contribution

The paper introduces an ensembling strategy combining diverse BERT models for comment classification, showing improved performance over single models.

Findings

Ensemble models outperform individual BERT models.

Twitter-based BERT models perform best among all models.

Multilingual models perform slightly worse than German and Twitter models.

Abstract

This paper describes our approach (ur-iw-hnt) for the Shared Task of GermEval2021 to identify toxic, engaging, and fact-claiming comments. We submitted three runs using an ensembling strategy by majority (hard) voting with multiple different BERT models of three different types: German-based, Twitter-based, and multilingual models. All ensemble models outperform single models, while BERTweet is the winner of all individual models in every subtask. Twitter-based models perform better than GermanBERT models, and multilingual models perform worse but by a small margin.