We leverage a BLSTM with attention to identify toxic spans in texts. We explore different dimensions which affect the model's performance. The first dimension explored is the toxic set the model is trained on. Besides the provided dataset, we explore
the transferability of 5 different toxic related sets, including offensive, toxic, abusive, and hate sets. We find that the solely offensive set shows the highest promise of transferability. The second dimension we explore is methodology, including leveraging attention, employing a greedy remove method, using a frequency ratio, and examining hybrid combinations of multiple methods. We conduct an error analysis to examine which types of toxic spans were missed and which were wrongly inferred as toxic along with the main reasons why they occurred. Finally, we extend our method via ensembles, which achieves our highest F1 score of 55.1.
In this work, we present our approach and findings for SemEval-2021 Task 5 - Toxic Spans Detection. The task's main aim was to identify spans to which a given text's toxicity could be attributed. The task is challenging mainly due to two constraints:
the small training dataset and imbalanced class distribution. Our paper investigates two techniques, semi-supervised learning and learning with Self-Adjusting Dice Loss, for tackling these challenges. Our submitted system (ranked ninth on the leader board) consisted of an ensemble of various pre-trained Transformer Language Models trained using either of the above-proposed techniques.
The Toxic Spans Detection task of SemEval-2021 required participants to predict the spans of toxic posts that were responsible for the toxic label of the posts. The task could be addressed as supervised sequence labeling, using training data with gol
d toxic spans provided by the organisers. It could also be treated as rationale extraction, using classifiers trained on potentially larger external datasets of posts manually annotated as toxic or not, without toxic span annotations. For the supervised sequence labeling approach and evaluation purposes, posts previously labeled as toxic were crowd-annotated for toxic spans. Participants submitted their predicted spans for a held-out test set and were scored using character-based F1. This overview summarises the work of the 36 teams that provided system descriptions.
