FHAC at GermEval 2021: Identifying German toxic, engaging, and fact-claiming comments with ensemble learning

TL;DR

This paper explores ensemble learning with fine-tuned German BERT and ELECTRA models to classify toxic, engaging, and fact-claiming comments in Facebook data, achieving high macro-F1 scores in the GermEval 2021 competition.

Contribution

It introduces an ensemble approach combining German BERT and ELECTRA models for multi-label comment classification, analyzing the impact of ensemble size and composition.

Findings

Best ensemble achieved macro-F1 of 0.73 on out-of-sample data.

F1 scores of 0.72, 0.70, and 0.76 for the three subtasks.

Ensemble size and composition influence classification performance.

Abstract

The availability of language representations learned by large pretrained neural network models (such as BERT and ELECTRA) has led to improvements in many downstream Natural Language Processing tasks in recent years. Pretrained models usually differ in pretraining objectives, architectures, and datasets they are trained on which can affect downstream performance. In this contribution, we fine-tuned German BERT and German ELECTRA models to identify toxic (subtask 1), engaging (subtask 2), and fact-claiming comments (subtask 3) in Facebook data provided by the GermEval 2021 competition. We created ensembles of these models and investigated whether and how classification performance depends on the number of ensemble members and their composition. On out-of-sample data, our best ensemble achieved a macro-F1 score of 0.73 (for all subtasks), and F1 scores of 0.72, 0.70, and 0.76 for subtasks 1, 2, and 3, respectively.