Study on Combined-CNN Model for Classification of Terrorism Text

Ascertaining the type of behaviors in terrorist attacks is one of the most critical steps in counter-terrorism text analysis. Deep learning methods are the best solution to unstructured and complex big data for counter-terrorism. This study proposes a classification model based on a combined Convolutional Neural Network (CNN) named Combined-CNN model to address the difficulty of incomplete feature extraction from localized textual segments within terrorism text. This model uses Word2vec vectors to construct word embeddings, facilitating the representation of lexical semantics. By putting these word embeddings into the combined convolutional and pooling layers, discernible vector features are extracted to capture linguistic patterns within the text. The empirical findings obtained from the analysis of the Global Terrorism Database (GTD) demonstrate the superior efficacy of the Word2vec+combined-CNN model in the context of text classification tasks. Notably, this model exhibits heightened levels of accuracy and recall when contrasted against benchmark models, underscoring its proficiency in discerning nuanced patterns within terrorism-related textual data.