An Interpretable Siamese Bi-LSTM Framework for Addressing Class Imbalance in the Detection of False Reviews

¹ Department of Computer Science, Central University of Rajasthan, Ajmer 305817, India
² Department of Computer Science, Central University of Rajasthan, Ajmer 305817, India

^* This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

Fake online reviews undermine the credibility of digital markets by swaying consumers and eroding trust in businesses. Detecting fake reviews is essential for online platforms to ensure user trust. Advanced fake review detection capabilities are required to understand sophisticated fake reviews since they are written in refined speech styles and exist in enormous and highly imbalanced review datasets. Machine learning models trained on such datasets tend to be biased in their predictions. Traditional approaches, such as undersampling and oversampling, struggle to mitigate these problems and often introduce missing information or prediction errors. A new handling method emerged using a Siamese Bidirectional Long Short-Term Memory (SBiLM) framework that addresses class imbalance without resorting to sampling procedures. The proposed Enhanced SBiLM model uses a robust similarity measure to handle semantic relationships better and employs embedding magnitude analysis to provide interpretability that reveals the key features driving predictions. We preprocessed the Yelp filtered dataset and evaluated the performance of the proposed model using the Yelp filtered dataset; the model demonstrates superior performance compared to SBiLM and baseline approaches (DNN, CNN, SMOTE-DNN), achieving improved precision (0.99), recall (0.98), and F1-score (0.98), particularly for detecting fake reviews. Our robust and interpretable model addresses critical challenges in fake review detection and has the potential to contribute to a more trustworthy online environment.