Image spam detection on instagram using convolutional neural network

Abstract

Instagram is a social media to share moments in the form of photos and videos that is currently in great demand. However, the popularity of Instagram also widely used by certain people to spread spam for their personal interests such as advertising. Therefore, it requires a system for detecting spam on Instagram to obtain useful information and expected by users. The previous researches on image spam detection have been done to filter out inappropriate content on email using conventional classification methods. Recently, the Convolutional Neural Network (CNN) is a method that obtains higher accuracy than conventional classification methods for image classification problems without prior feature extraction process. We propose image spam detection using CNN on social media Instagram. The four architectures of CNN are used to compare the performance of each architecture, i.e., 3-layer, 5-layer, AlexNet, and VGG16. The performance of the system is evaluated by 8000 images taken from Instagram using web crawler. The results show that the highest accuracy achieves 0.842 by using a VGG16 architecture.