This document presents a method for detecting fake reviews based on multiple feature fusion and rolling collaborative training. The method establishes an evaluation index system using multiple features such as text, sentiment, and user behavior. It trains 7 classifiers on an initial sample set and selects the most accurate classifier. Newly classified reviews and their features are then added back to the sample set, automatically increasing its size. Experimental results on Yelp reviews show the method achieves 84.45% accuracy for detecting fake reviews, outperforming baseline methods by 3.5% and the latest deep learning model by 5.3%. Statistical tests found support vector machines and random forests to be the best classifiers.

Fake Review Detection Based On Multiple Feature Fusion and Rolling Collaborative Training PDF

The list of books you might like