Design of an Intelligent Model Selection Framework for Unified Review Authenticity Detection with Real-Time E-Commerce Integration

Abstract

Objectives: Develop a machine learning system with online classification ability for detecting fake reviews of hotels using a web-based application.

 Methods: The analysis was done on 11,912 samples of hotel reviews. Preprocessing of the data set and extraction of the features were done using the TF-IDF method, after which seven machine learning algorithms including Decision Tree, K-Nearest Neighbor (KNN), Gaussian Naive Bayes, Multinomial Naive Bayes, Logistic Regression, Stochastic Gradient Descent, and Support Vector Machine (SVM) were trained using various kernels. The most efficient model was implemented using API on a demo hotel website to provide real-time prediction. Findings: An SVM model using Radial Basis Function (RBF) kernel yielded the highest prediction accuracy of 88% among the evaluated  classifiers such as Decision Tree (63%), KNN (73%), Gaussian Naive Bayes (74%), Multinomial Naive Bayes (79%), Logistic Regression (81%) and SGD (83%). It can be inferred from the results that the RBF kernel is very effective in handling non-linearity present in the data set. Contrary to offline models, the proposed system provides real-time detection of fake hotel reviews using API.

Novelty: This research has presented a unified framework that involves multi-model comparison, SVM kernel optimization, and real-time API-based deployment, enabling the practical implementation of machine learning algorithms for fake reviews detection applicable in online applications.