Detect Fake Review Farms in a Marketplace
A marketplace with 40,000 sellers suspects that some are buying 5-star reviews, now written with generative AI and hard to distinguish by the text. The tactics change every month and there are hardly any confirmed cases. Buyers are starting to distrust ratings and consumer authorities are watching this fraud.
Where the money comes from
What to do now
- 1Score accounts and sellers by how unusual their behaviour is.
- 2Investigation team on the weekly top 50.
- 3Keep confirmed cases for a supervised model.
For your committee
One page with the decision, the impact, the risks, the steps and the KPI to track.
The print dialog opens: choose "Save as PDF".
Watch out: more volume is not always more profit
Fake reviews inflate sales in the short term and the marketplace earns commission on them; tolerating them looks profitable until the trust of all buyers falls.
KPI for the committee
% of investigations with confirmed fraud and trust in reviews.
Detail for your data team
Proposed technique: Isolation Forest
Why: There are hardly any confirmed cases and the tactics change; moreover, AI-generated text is indistinguishable. Behaviour, however, gives it away: new accounts that review 30 products from the same seller in one hour. Isolation Forest isolates those rare patterns without needing labels.
Cost of getting it wrong: Letting review farms through erodes trust in the whole marketplace; accusing an honest seller creates conflict and possible claims.
Data needed: Reviews: date, rating, length, product, seller.; Behaviour of each account: age, purchases, reviews per day, devices, IP addresses.; Seller patterns: review spikes after launch, reviews/sales ratio.