Abstract
Nowadays online customer reviews play a vital role in the success of various businesses. Authentic reviews for businesses help customers make informed decisions. However, with the recent advances in ML/AI technologies, businesses may be able to use crowdturfing systems to fabricate machine generated fake reviews to misguide customers by: (1) posting positive fake reviews to attract more customers and promote their businesses, or (2) posting negative fake reviews to demote their competitors. In this project, we propose a framework to detect machine-generated fake reviews using deep learning techniques. This project consists of two phases. In the first phase, based on the human genuine reviews collected from Yelp, we employ word-level recurrent neural networks (RNNs) to generate a benchmark dataset including both machine-generated (fake) positive and negative reviews. In the second phase, we develop a word-level RNN model to effectively classify online reviews into human reviews and machine-generated reviews. Extensive experiments show that our approach outperforms the baseline RNN method in terms of various effectiveness metrics under all different settings.