Detection of Arabic Spam Tweets Using Word Embedding and Machine Learning

Twitter has become one of the most popular social networking platforms for sharing activities and opinions. In this study, we explore the idea of applying word embedding based features with machine-learning techniques to detect Arabic spam tweets. In addition, the effects of text domain of the collected corpus to learn word embedding is analyzed. This is evaluated using a publicly available dataset of 3503 tweets alongside with three popular classifiers for binary classification, namely: Naïve Bayes, Decision trees and SVM. The experimental results reveal that the proposed method outperforms the baseline approach in distinguishing between machine-generated tweets and human-generated tweets. An accuracy rate of 87.33% is achieved using skip-gram word2vec technique with SVM.