Amazon Technologies, Inc.

Samuel Theodore Sandler

Jon T Hanlon

Giovanni Zappella

David Spike Palfrey

Rodolphe Jenatton

Karthik Mohan

Monica T McCann

Marcel Ackermann

Daniel Tarekegn

The present disclosure is directed to training, and providing recommendations via, a machine learning model architected to balance relevance and diversity of sets of recommendations. For example, a neural network can be provided with user profile features and can output probabilities for each of a number of recommendations. This can be converted into a ranked list of recommendations. The ranked list of recommendations is provided to a diversity model that maximizes an optimization objective having a first objective that quantifies relevance of a recommendation and a second objective that measures diversity of a set of recommendations. The output of the diversity model is a set of recommendations that have both high relevance and high diversity.

Artificial intelligence system for balancing relevance and diversity of network-accessible content

A system is disclosed that identifies early adopter users by creating a directed graph of item access information for an item category and performing a page rank type process on the item access information. This directed graph may be created in a reverse temporal order. The early adopter users can be identified as the users with nodes in the directed graph that have a threshold number or rate of incoming links directly or indirectly pointing towards the nodes. Using the early adopter users as a sample, systems herein can determine whether to recommend an item based on the popularity of the item with respect to the early adopter users. Further, systems herein can determine an inventory level to maintain for an item based on the popularity of the item with respect to the early adopter users.

Automated graph-based identification of early adopter users

A network-based enterprise or other system that makes items available for selection to users may implement real-time updates to item recommendation models based on matrix factorization. An item recommendation model may be maintained that is generated from a singular value decomposition of a matrix indicating selections of items by users. A user-specific update to the item recommendation model may be calculated in real-time for a particular user such that the calculation may be performed without performing another singular value decomposition to generate an updated version of the item recommendation model. Item recommendations may then be made based on the user-specific update and the item recommendation model. In various embodiments, the item recommendations may be made in response to an indication or request for item recommendations for the particular user.

Real-time updates to item recommendation models based on matrix factorization

Disclosed herein is a platform for automating the discovery of product bundles that are promising for sale to customers (i.e., they are relevant, valuable and/or timely). Techniques are disclosed for generating a set of potential product bundles according to criteria, thresholds and/or scores which indicate a propensity for items to be purchased together by customers. The generated set of potential product bundles may be ranked according to opportunity metrics and/or scores which indicate value to a merchant and/or customers. A subset of product bundles is selected from the set of potential product bundles based at least in part on the ranking, and each product bundle in the selected subset is matched to one or more entities, such as merchants. These product bundles may then be presented to the associated (i.e., matched) entity based on some criteria, such as offerings of the merchants.

Growing community of inventors

Seattle, WA, United States of America

Samuel Theodore Sandler