Verifai: AI Review Intelligence and Trust Analyzer for Fake Review Detection Using Machine Learning
Online reviews influence purchasing decisions but can also contain misleading or suspicious content. This paper presents Verifai, a review intelligence system that extends a fake-review classifier with supporting, interpretable signals. Its authenticity model uses TF-IDF features with unigrams and bigrams and Logistic Regression, selected after comparison with Multinomial Naive Bayes and Random Forest using five-fold cross-validation on weighted F1. The labelled corpus was cleaned from 381,734 raw entries to 28,530 reviews and divided into training and held-out sets, with 5,706 reviews reserved for testing. On that test set, the confusion matrix yields 72.59% accuracy, 72.49% fake-class precision, 73.45% fake-class recall, and 72.97% fake-class F1; ROC-AUC was reported as 80.59%. Beyond classification, Verifai analyzes suspicious writing patterns, sentiment, exact duplicates, rating-text consistency, and product aspects, then presents an indicative 0-100 reliability score. A Flask API serves a React and TypeScript interface; all 20 reported application regression cases passed. The score is a decision support signal, not proof that a review or seller is fraudulent.
Authors
- Suthar Soham
- Gajera Yagnik
- Patel Yash
Institutions
- Parul University (IN)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-13
- DOI
- https://doi.org/10.5281/zenodo.24190446
- Primary Topic
- Spam and Phishing Detection
- Type
- article
- Field-Weighted Citation Impact
- 0.00