Towards Detecting Fake Spammers Groups in Social Media: An Unsupervised Deep Learning Approach
Book Chapter
Soni, J, Prabakar, N, Upadhyay, H. (2022). Towards Detecting Fake Spammers Groups in Social Media: An Unsupervised Deep Learning Approach
. 113 237-253. 10.1007/978-3-031-10869-3_13
Soni, J, Prabakar, N, Upadhyay, H. (2022). Towards Detecting Fake Spammers Groups in Social Media: An Unsupervised Deep Learning Approach
. 113 237-253. 10.1007/978-3-031-10869-3_13
In recent years, the acceptance of social media has increased, where multiple users use various platforms to share different kinds of information. The current estimated number of online users is almost a billion. Today's web traffic has increased by more than a half within a year because of the increase in online activities. More data can be mined with more users performing online activities. Therefore, advertising and marketing administrations need to know how people express their views and their response can improve their business efficiency. Since social media are rooted at diverse scales in various data sources, they are often quite large. Nowadays, people review products online, which plays a vital role in making buying choices by most consumers in digital consumer markets. Spammers often write fake reviews to increase/decrease the value of specific products by taking advantage of online reviews. Past studies have focused on spotting distinct fake reviewer-ids of fake reviews. However, to target and control a specific product's sentiment, fake reviewers create multiple fake ids and work together in groups to write reviews. In this chapter, we address the problem of detecting such groups of fake reviewers. Explicitly, we propose a deep learning-based framework for fake reviewer groups detection data of reviewers using various unsupervised deep learning algorithms such as Self Organizing Maps and Restricted Boltzmann Machine. We perform the hyperparameter optimization of these algorithms on this reviewer graph data using advanced deep learning frameworks such as TensorFlow and popular machine learning frameworks such as scikit-learn. In the end, we discussed the practical implementation of grouping the fake reviewers on the real-world Google Play Store app review dataset, which has fractional ground-truth data on about 2207 fake reviewer-ids out of all 38,123 reviewer-ids in the original dataset. We validate with the experimental results that the projected method can detect the group of fake reviewers with reasonable accuracy. It can also be extended to perceive opinion spammers groups in social media with semantic features, sequential affinity, and emotion analysis.
