Abstract: Reviews submitted through a discovery application may be provided to a validation server and reviewed against indicators in the review data to determine whether a given review is potentially fraudulent, not fraudulent or legitimate, or should be flagged for administrative review through an administrative model. Reviews processed by the administrative model may be fed back to a machine leaning module on the validation server as additional positive or negative examples. Using these additional examples, the machine learning module may adjust weights of associated indicators and/or identify additional indicators in the review data for consideration in flagging fraudulent reviews. The additional indicators may be flagged for administrative review prior to implementation in reviewing and flagging reviews.