Preprocessing multilingual text for the detection of extremism and radicalization in social networks using deep learning

Due to the lack of strict controls on social networks, extremist groups like ISIS, Al-Qaeda, and white supremacists have taken advantage of these platforms to spread their ideas, distribute harmful content, and recruit new members. The study of online extremism and radicalization is a multifaceted and intricate area of research. Although the majority of research in this field focuses on the analysis of data in a single language, there needs to be more studies on the analysis of multilingual data, specifically about detecting multi-ideology extremism in social media content. This research paper introduces the building of an artificial intelligence system that identifies instances of extremism and radicalization from data extracted from social networks. We utilize natural language processing (NLP) linguistic methods and text classification to process the textual data. Our study results show significant progress in multiclass multilingual text classification and the detection of extremism and radicalization within social networks. The Bi-LSTM (Bidirectional et al.) model demonstrates a binary classification accuracy of 97.33%, and the multiclass classification accuracy of the Transformer-based model, which employs the DistilBERT-multi (Distilled version of the Multilingual Bidirectional Encoder Representations from Transformers) pre-trained model, is 91.07%.