Note: The contents of this repository include preliminary research on fake review detection using MCD-embedded BERT and, LSTM models conducted during our academic years in BRACU.
In the e-commerce landscape of today, online shopping reigns supreme, largely due to customers relying heavily on reviews. However, since customers cannot physically examine products, online reviews hold immense sway over their purchasing decisions. Sellers take advantage of this by employing deceptive tactics like fake reviews to sway opinions. Despite the proliferation of fake review detection models, ensuring robustness remains a challenge. Our study addresses this gap by utilizing BERT and LSTM models alongside Monte Carlo Dropout (MCD) on the Yelp Labelled Dataset. MCD bolsters robustness by introducing uncertainty through neuron dropout. The BERT-embedded MCD achieves an impressive 91.75% accuracy, surpassing the LSTM model, thus rendering it more dependable for detecting fake reviews.
The official paper on "Robust Fake Review Detection using Uncertainty-Aware LSTM and BERT" has been published in the 2023 IEEE 15th International Conference on Computational Intelligence and Communication Networks (CICN).