UIT-E10dot3 at SemEval-2021 Task 5: Toxic Spans Detection with Named Entity Recognition and Question-Answering Approaches

TL;DR

This paper presents two approaches for toxic span detection in online comments, using NER with spaCy and QA with RoBERTa and ToxicBERT, achieving a highest F1-score of 66.99%.

Contribution

It introduces a novel combination of NER and QA methods for toxic span detection and compares their effectiveness on the SemEval-2021 dataset.

Findings

NER approach achieved 66.99% F1-score.

QA approach provides a competitive alternative.

Analysis of toxic span structure informed method selection.

Abstract

The increment of toxic comments on online space is causing tremendous effects on other vulnerable users. For this reason, considerable efforts are made to deal with this, and SemEval-2021 Task 5: Toxic Spans Detection is one of those. This task asks competitors to extract spans that have toxicity from the given texts, and we have done several analyses to understand its structure before doing experiments. We solve this task by two approaches, Named Entity Recognition with spaCy library and Question-Answering with RoBERTa combining with ToxicBERT, and the former gains the highest F1-score of 66.99%.