LISAC FSDM USMBA at SemEval-2021 Task 5: Tackling Toxic Spans Detection Challenge with Supervised SpanBERT-based Model and Unsupervised LIME-based Model

Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021), pages 865–869 Bangkok, Thailand (online), August 5–6, 2021. ©2021 Association for Computational Linguistics 865 LISAC FSDM USMBA at SemEval-2021 Task 5: Tackling Toxic Spans Detection Challenge with Supervised SpanBERT-based Model and Unsupervised LIME-based Model Abdessamad Benlahbib1∗, Ahmed Alami2, Hamza Alami2∗, 1 LISAC Laboratory, Faculty of Sciences Dhar EL Mehraz (F.S.D.M), Sidi Mohamed Ben Abdellah University (U.S.M.B.A) 2 Laboratory of Engineering Sciences, National School of Applied Sciences, Ibn Tofail University, Kenitra, Morocco abdessamad.benlahbib@usmba.ac.ma, alami.alami1996@gmail.com, hamza.alami1@usmba.ac.ma Abstract Toxic spans detection is an emerging challenge that aims to find toxic spans within a toxic text. In this paper, we describe our solutions to tackle toxic spans detection. The first so- lution, which follows a supervised approach, is based on SpanBERT model. This latter is intended to better embed and predict spans of text. The second solution, which adopts an un- supervised approach, combines linear support vector machine with the Local Interpretable Model-Agnostic Explanations (LIME). This last is used to interpret predictions of learning- based models. Our supervised model outper- formed the unsupervised model and achieved the f-score of 67,84% (ranked 22/85) in Task 5 at SemEval-2021: Toxic Spans Detection. 1 Introduction By dint of the massive production of user-generated content in social media, moderation becomes cru- cial to promote healthy online discussions by re- moving toxic posts and contents. However, it is nearly impossible for a human to keep tracking user-generated content. Thus, the need for the right tools and technologies to help in such a task be- comes a necessity. The Toxic Spans Detection task aims to detect the spans that make a text toxic. While several toxicity detection datasets (Wulczyn et al., 2017; Borkan et al., 2019) and models (Pavlopoulos et al., 2017a, 2019; Schmidt and Wiegand, 2017; Pavlopoulos et al., 2017b; Zampieri et al., 2019; Alami et al., 2020) have been released. However, these works estimate the likelihood of a document being toxic with weak interpretability. In fact, high- lighting toxic spans can assist human moderators who often deal with lengthy comments, and who prefer attribution instead of just a system-generated unexplained toxicity score per post. ∗contributed equally In this paper, we propose two solutions to tackle toxic spans detection (Pavlopoulos et al., 2021). The first solution, which follows a supervised ap- proach, is based on SpanBERT (Joshi et al., 2020) model that is pre-trained on span boundary objec- tive and considers masks contiguous spans. There- fore, SpanBERT gives better spans representations and predictions. The second solution, which adopts an unsupervised approach, combines linear support vector machine (Fan et al., 2008) with the Local In- terpretable Model-Agnostic Explanations (LIME) (Ribeiro et al., 2016). LIME is an explanation technique that seeks to faithfully interpret the pre- dictions of any classifier. This paper is organized as follows: Section 2 describes the proposed methods; Section 3 presents the experimental results; Finally, Section 4 con- cludes and outlines future directions. 2 Methods In this section, we describe the proposed solu- tions including SpanBERT-based method …

Content truncated to 500 words