A Hybrid Deep Learning Approach for Spam Detection in Twitter

Detecting malicious user accounts on Twitter has become an active area of research in social network analysis.This kind of ill-intentioned users send undesired tweets to other users to promote products, services, rumors, fake news, or any abusive content.Hence, the detection of those spammers and their originators will prevent deterioration in the quality of communication services and legitimate users from being affected.Traditional machine learning techniques have been proposed to tackle the problem of spammers detection.However, many researchers have pointed out that the majority of machine learning based models that rely on supervised classification didn't perform well in noisy and short message platforms like Twitter.Recently, deep learning-based alternatives have shown remarkable performance in this area because of their competitive training speed and low implementation cost.In this paper, we propose a new hybrid architecture that combines Principal Component Analysis (PCA) with Convolutional Neural Network (CNN) to give birth to a more reliable and robust model for spammers detection in Twitter.Unlike other hybridizations, the convolutional layer in the CNN module is not fed traditionally by raw feature vectors, rather, we use very low dimensional vectors containing high-order features provided by PCA module.A series of nicely conducted experiments over benchmark datasets have shown that the hybridization proved to be effective for the detection of spammers.The results show that PCA-CNN model can achieve better classification performance with 94.91% precision, 96.76% recall, and 95.83% F-score when compared to baseline benchmarks like CNN, ANN and SVM.