Toxic Comment Classification For French Online Comments - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2020

Toxic Comment Classification For French Online Comments

Résumé

In this paper, we propose a supervised approach for toxic comment classification for French language. We choose a set of features proposed for toxic comment detection for English and use it for French toxic comment detection. Our approach is based on N-gram features, linguistic features and a dictionary of insulting words and expressions. We obtain a F1-score of 78% with N-grams, linguistic and lexicon features, a precision of 87% with N-gram features and a recall of 83% with N-gram, linguistic and lexicon features. Classifiers used are linear SVM and decision tree.
Fichier non déposé

Dates et versions

hal-03161388 , version 1 (06-03-2021)

Identifiants

Citer

Nadira Boudjani, Yannis Haralambous, Inna Lyubareva. Toxic Comment Classification For French Online Comments. ICMLA 2020 : 19th IEEE International Conference on Machine Learning and Applications, Dec 2020, Miami, United States. pp.1010-1014, ⟨10.1109/ICMLA51294.2020.00164⟩. ⟨hal-03161388⟩
139 Consultations
0 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More