Review AI detector (CSE 594 coursework)

DistilBERT (distilbert-base-uncased) fine-tuned to tell real guest reviews of Chicago hotels (truthful reviews of the Ott et al. Deceptive Opinion Spam Corpus) from AI-generated ones (Qwen3.8-27B and gpt-oss-120b). Label 1 = AI-generated. 3 epochs, lr 3e-5, batch 16, max length 384.

Built for a class human-AI study. It was evaluated on reviews from Claude Opus 5.5, a generator it never saw in training: 86.25% accuracy on a 400-review study set. It is not a general-purpose AI-text detector.

Downloads last month
46
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for Xinyueee/review-ai-detector

Quantized
(108)
this model