Abstract: Systems and methods are disclosed for predicting a product's (e.g., a book's) performance prior to its availability. An example embodiment is a system for machine learning classification that includes representations of characteristics of products, a pre-processor, and a machine learning classifier. The pre-processor can determine (i) representations of comparative intrinsic characteristics of the products based on the representations of characteristics of products and (ii) representations of corresponding comparative extrinsic characteristics of the products. The pre-processor can generate a data structure representing relationships between the comparative intrinsic characteristics and the comparative extrinsic characteristics. The machine learning classifier is trained with the data structure. The classifier can return representations of comparative extrinsic characteristics in response to given comparative intrinsic characteristics.

Abstract: Methods, systems, and media for determining location information from real-time bid requests are provided. In some implementations, a method for determining device locations is provided, the method comprising: receiving a real-time data stream that includes location proxies corresponding to devices; generating a movement graph of interconnected nodes and edges, wherein each node represents one of the location proxies and wherein each edge represents movement information between the location proxies; calculating, for a node in the movement graph, location information by applying a predictive model having weights based on the movement information; and assigning, for the node in the movement graph, a census-based identifier to the calculated location information. The census-based identifier can be used, for example, to supplement or enrich bid requests within the real-time data stream.

Abstract: Methods, systems, and media for determining location information from real-time bid requests are provided. In some implementations, a method for determining device locations is provided, the method comprising: receiving a real-time data stream that includes location proxies corresponding to devices; generating a movement graph of interconnected nodes and edges, wherein each node represents one of the location proxies and wherein each edge represents movement information between the location proxies; calculating, for a node in the movement graph, location information by applying a predictive model having weights based on the movement information; and assigning, for the node in the movement graph, a census-based identifier to the calculated location information. The census-based identifier can be used, for example, to supplement or enrich bid requests within the real-time data stream.