First large-scale, gold-standard dataset of deceptive opinion spam in product reviews. Contains 800 hotel reviews: 400 authentic 5-star reviews scraped from TripAdvisor, and 400 deceptive reviews created via Amazon Mechanical Turk where workers role-played as marketing employees writing fake positive reviews.
Details
- Truthful reviews: 5-star reviews from 20 popular Chicago hotels on TripAdvisor (balanced across hotels)
- Deceptive reviews: Written by crowdworkers on AMT, instructed to pose as hotel marketing employees writing positive reviews (workers assumed their "boss" wanted the hotel marketed favorably)
- Quality control: All reviews at least 150 characters; submissions checked for plagiarism; workers restricted to US-based Turkers with 90%+ approval rating
- Collection time: ~14 days to gather 400 satisfactory deceptive opinions
- Balanced design: Similar document lengths and realistic composition across both classes
Applications
Used extensively in research on deceptive opinion spam detection, fake review generation, and linguistic markers of deception. Benchmark dataset for evaluating machine learning and deep learning approaches to fraud detection in online reviews.
Related datasets
- FakeNewsAMT (Pérez-Rosas et al. 2017) — similar gold-standard approach applied to fake news
- Liar (Wang 2017) — political fact-checking claims and verdicts
Citation
Ott, M., Choi, Y., Cardie, C., & Hancock, J. T. (2011). Finding deceptive opinion spam by any stretch of the imagination. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics, 309–319.