Advanced Data Science and Computational Intelligence

(Annual (One issue per year))

ABSTRACT

Without us realizing it, online reviews have emerged as one of the most influential factors in directing consumer behavior. A few star ratings with brief user feedback on platforms like Amazon, Flipkart, Yelp, and TripAdvisor can make or break a product. Yet, this great power has led to the emergence of review manipulation: fake reviews created by bots, writers for hire, or more and more advanced AI tools, are degrading the trust on which these platforms rely. Human checking of these is just not possible as every day millions of reviews are coming in. This paper describes a system we built to tackle this problem from multiple angles simultaneously. Rather than treating fake review detection as a pure text classification task, we combined two complementary deep learning models a CNN-LSTM network and a fine-tuned BERT model-with a behavioral analysis layer that tracks user reputation over time. We also went a step further than most academic work by building a Chrome extension that intervenes in real time, visually blocking fake reviews before a consumer even reads them and striking through ratings that have been artificially inflated. The ensemble of BERT and CNN-LSTM achieves 95.32% accuracy, with BERT alone reaching 94.58%. More practically, the system successfully prevents 99.7% of flagged fake reviews from ever appearing on screen—turning detection into genuine consumer protection.

Keywords: Fake Reviews; Machine Learning; Deep Learning; BERT; CNN-LSTM; Natural Language Processing; Sentiment Analysis; Fraud Detection; Browser Extension; User Blacklisting; Real-Time Detection; Review Analysis; Rating Invalidation