The rise in hateful and offensive language directed at other users is one of the adverse side effects of the increased use of social networking platforms. This could make it difficult for human moderators to review tagged comments filtered by classification systems. To help address this issue, we present the ViHOS (Vietnamese Hate and Offensive Spans) dataset, the first human-annotated corpus containing 26k spans on 11k comments. We also provide definitions of hateful and offensive spans in Vietnamese comments as well as detailed annotation guidelines. Besides, we conduct experiments with various state-of-the-art models. Specifically, XLM-R_Large achieved the best F1-scores in Single span detection and All spans detection, while PhoBERT_Large obtained the highest in Multiple spans detection. Finally, our error analysis demonstrates the difficulties in detecting specific types of spans in our data for future research. Our dataset is released on GitHub.

We describe our systems participated in the SemEval-2023 shared task for Named Entity Recognition (NER) in English and Bangla. In order to address the challenges of the task, where a large number of fine-grained named entity types need to be detected with only a small amount of training data, we use a method to augment the training data based on BabelNet conceptsand Wikipedia redirections to automatically annotate named entities from Wikipedia articles. We build our NER systems based on the powerful mDeBERTa pretrained language model and trained on the augmented data. Our approach significantly enhances the performance of the fine-grained NER task in both English and Bangla subtracks, outperforming the baseline models. Specifically, our augmented systems achieve macro-f1 scores of 52.64% and 64.31%, representing improvements of 2.38% and 11.33% over the English and Bangla baselines, respectively.

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’s library and Question-Answering with RoBERTa combining with ToxicBERT, and the former gains the highest F1-score of 66.99%.