Abstract: Items within an index may be converted from classic geometry and embedded into a hyperbolic space. The hyperboloids within the hyperbolic space provide higher precision classifications of items within the index relative to their hierarchical structure. A received search query may also be converted to hyperbolic space and mapped as a query hyperboloid against an answer space that includes hyperboloids for items within the index. Distances or overlaps between the hyperboloids may be determined in order to generate a set of search results.

Abstract: Technologies are provided for reformulation of a tail query to a head query with the same purchase intent by mapping the tail query to the head query. In some of the technologies, a reasonable embedding can be learned on historical head queries. The embedding can then be refined by leveraging rewards generated from a persistently noisy oracle that compensates for the lack of historical behavioral signal for tail queries. Further, a contextual sampling technique that uses text-based rewards or oracle-based rewards, or both, can be implemented in order to avoid biases introduced by persistent noise in the oracle. Numerical experiments on large scale e-commerce datasets demonstrate that the provided technologies can outperform several conventional approaches to query reformulation.

Abstract: Various embodiments attempt to predict items in a new product category that may be of interest to a particular user when the user has not interacted with or made purchases in that product category. In accordance with various embodiments of the present disclosure, this is accomplished by training machine learning models, such as deep neural networks, using both the historical purchase data from the particular user as well as the historical purchase data from many other users to infer what new category of items the particular user might be interested in. In some embodiments, the previous purchase data of the other users are used to construct a graph, and the graph is used to train the machine learning model rather than the raw data. The graph can represent the entirety of the previous purchase data with less actual data, thereby requiring less memory and other computational resources.

Abstract: Technologies are provided for reformulation of a tail query to a head query with the same purchase intent by mapping the tail query to the head query. In some of the technologies, a reasonable embedding can be learned on historical head queries. The embedding can then be refined by leveraging rewards generated from a persistently noisy oracle that compensates for the lack of historical behavioral signal for tail queries. Further, a contextual sampling technique that uses text-based rewards or oracle-based rewards, or both, can be implemented in order to avoid biases introduced by persistent noise in the oracle. Numerical experiments on large scale e-commerce datasets demonstrate that the provided technologies can outperform several conventional approaches to query reformulation.