Amazon Technologies, Inc.

Rohan Gurappagouda Patil

Han Wu

Pragyana K Mishra

Michael Arthur Ten-Pow

Rohit K Gupta

Purak Jain

Pedro Henrique Almeida Tanaka

Ayman Omar Farahat

Douglas Wong

Amit Goyal

Sudheer Babu Tubati

Ehtesham Sam Haque

Kimberly Michele Wike

Himangshu Ranjan Borah

Systems and techniques are disclosed for detecting fraudulent activity by users of a media service, such as a music or video streaming service. Fraudulent activity may artificially modify metrics tracked by the media service, thus impacting chart positions, media recommendations, royalty amounts, and other elements determined by the media service. Media usage data associated with user activity samples is labeled to identify fraud samples and non-fraud samples. Because the majority of user activity on the media service may be legitimate such a number of labeled non-fraud samples exceeds a number of labeled fraud samples, a balanced training data set is created via undersampling to reduce the number of non-fraud samples to be equal or closer to the number of fraud samples. The balanced training data set is used to train a machine learning model to generate fraud predictions associated with new user activity on the media service.

Media fraud detection system

Bid values submitted for various keywords can take into account the recovery propensity between paid search and organic search. When submitting a bid to a search engine provider for a keyword, an entity may get a certain level of performance in return. If not submitting a bid, however, the entity will likely still get some level of performance, although likely less than for paid search. In order to optimize for a parameter such as impressions, purchases, or profit, the recovery propensity can be taken into account in order to adjust the bid price, taking into account the relative performance of paid and organic search and then optimizing for the determined goal. Organic search data in some embodiments can be obtained through testing or modeling, or a combination thereof.

Recovery-aware content management

Data security and privacy are improved by a client providing a hashed version of collected data to a remote analysis service, and having the analysis service determine the relevancy of the data from the hashes before requesting the plaintext of the data. In one example, a browser plug-in obtains data which is divided into overlapping three-character sequences, and the sequences are hashed to produce a sequence of hashes. The sequence of hashes is sent by the plug-in to the remote service, which uses the hashes to determine if the associated data is relevant to the analysis performed by the remote service, without requiring access to the associated plaintext. After making the determination, the remote service may request that relevant data be provided to the service in plaintext form, while data that is not relevant need not be sent to the remote service.

Cryptographic data encoding method with enhanced data security

Methods, systems, and computer-readable media for optimized management of online advertising auctions are disclosed. An auction management system sends bids for one or more bidding units (e.g., keywords or product identifiers) to an advertising vendor. A bid comprises a value per click associated with one or more search results pages or web pages for one or more search terms. At least a portion of the bids represent winning bids in one or more auctions for the one or more bidding units. The system determines success metrics associated with the bids based (at least in part) on user traffic for links associated with the winning bids. The system determines an optimal bid for a particular bidding unit based (at least in part) on the success metrics. The system sends the optimal bid for the particular bidding unit to the advertising vendor for an additional auction.

Optimized management of online advertising auctions

Growing community of inventors

Seattle, WA, United States of America

Rohan Gurappagouda Patil