Title: Detection of terrorism's apologies on Twitter using a new bi-lingual dataset

Authors: Khaled Bedjou; Faical Azouaou

Addresses: Laboratoire d'Informatique Médicale (LIMED), Faculté des Sciences Exactes, Université de Bejaia, 06000 Bejaia, Algeria ' Laboratoire LMCS, Ecole Nationale Supérieure d'Informatique, ESI ex INI Oued-smar-Alger, Algeria

Abstract: A lot of terrorist apology content is being shared on social media without being detected. Therefore, the automatic and immediate detection of these contents is essential for people's safety. In this paper, we propose a language independent process to detect and classify terrorism's apologies on Twitter into three classes (apology, no apology, and neutral). We tested the process on a bi-lingual (Arabic and English) dataset of 12,155 manually annotated tweets. We conducted two sets of experiments, one with imbalanced data and the other with oversampled data. We compared the classification performances of four machine learning algorithms (RF, DT, KNN, and NB) and five deep learning algorithms (GRU, SimpleRNN, LSTM, BiLSTM, and BERT). Our comparative study concluded that BERT achieves better classification performance than the others do, with an accuracy of 0.84 for Arabic and 0.81 for English on imbalanced data, and 0.88 for Arabic and 0.91 for English on oversampled data.

Keywords: terrorism's apology; social network analysis; Twitter; NLP; sentiment analysis; machine learning; deep learning; transfer learning.

DOI: 10.1504/IJDMMM.2023.134581

International Journal of Data Mining, Modelling and Management, 2023 Vol.15 No.4, pp.331 - 354

Received: 15 Jul 2022
Accepted: 08 Oct 2022
Published online: 30 Oct 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article