sarkarghya/amazon-suspicious-review-detector

Amazon Suspicious Review Detector

ModernBERT fine-tuned to identify suspicious review-reuse patterns in Amazon Reviews 2023.

This is a weakly supervised pattern detector. Amazon Reviews 2023 does not provide verified fake/real labels. Label suspicious_pattern means that a review matched high-confidence reuse heuristics; it is not proof of fraud.

Labels

• ordinary: conservative clean example
• suspicious_pattern: suspicious reuse pattern

Evaluation

The model was trained for 2 epochs on 34 Amazon categories with 28,321 examples per class. The held-out validation metrics are included in metrics.json.

• Accuracy: 0.8177
• F1: 0.8097
• ROC AUC: 0.8980
• Precision: 0.8474
• Recall: 0.7752

Usage

from transformers import pipeline

classifier = pipeline(
    "text-classification",
    model="sarkarghya/amazon-suspicious-review-detector",
)
print(classifier("The product arrived on time and works as described."))

Use the score as a review-quality signal, not as a standalone moderation or fraud decision.