Back to results
US 11,775,886 B2Granted
Combining group-level and item-level information for forecasting by empirical Bayesian deconvolution based technique
AssigneeAmazon Technologies, Inc.
InventorsRamesh Natarajan, Jonathan Hosking
Life of this patent
Estimating the term…
Abstract
A data set comprising records of state change events of items of an item collection, as well as records of asynchronous operations associated with the items, is obtained. The numbers of records in the data set may differ from one item to another. Using the data set, a Bayesian forecasting model employing a deconvolution algorithm is trained. The model generates estimates of metrics of a type of asynchronous operation using a combination of a category-level distribution of the asynchronous operation, an item-level distribution, and a category-versus item adjustment. A trained version of the model is stored.
Claims20
A system, comprising:
- one or more computing devices of a network-accessible forecasting service;
- wherein the one or more computing devices include instructions that upon execution on or across one or more processors cause the one or more computing devices to:
- obtain a training data set, comprising (a) records of a plurality of events, wherein individual ones of the events represent a state change of one or more items of an item collection, wherein at least one state change event represents a purchase transaction of an instance of an item, (b) indicators of respective categories to which individual ones of the items of the item collection belong, and (c) records of one or more types of asynchronous operations associated with at least some items of the item collection, wherein at least one type of asynchronous operation results in usage of customer support resources of an organization, wherein the item collection comprises at least a first item and a second item, and wherein, within the training data set, the number of represented state changes of the first item differs from the number of represented state changes of the second item;
- train, using the training data set, an empirical Bayesian forecasting model employing a deconvolution algorithm, wherein the empirical Bayesian forecasting model generates, corresponding to an item of the item collection whose state has changed, an estimate of a metric of a first type of asynchronous operation, wherein the estimate is based at least in part on (a) a category-level distribution of the first type of asynchronous operation, corresponding to a first category of the respective categories, (b) an item-level distribution of the first type of asynchronous operation, corresponding to the item, and (c) a category-versus-item adjustment between the category-level distribution and the item-level distribution;
- store a trained version of the empirical Bayesian forecasting model;
- in response to a forecasting request pertaining to at least the first item, obtain, using the trained version of the Bayesian forecasting model, a first estimated metric of a first type of asynchronous operation corresponding to the first item;
- and
- cause, based on the first estimated metric, one or more operations to be initiated with respect to at least some items of the item collection.
5 more elements
A method, comprising:
- performing, by one or more computing devices:
- obtaining a data set, comprising (a) records of a plurality of events, wherein individual ones of the events indicate changes to one or more items of an item collection, and (b) records of one or more types of asynchronous operations, wherein an asynchronous operation is related to one or more events of the plurality of events, wherein the item collection comprises a first item and a second item, and wherein, within the data set, the number of records indicating asynchronous operations related to state changes of the first item differs from the number of records indicating asynchronous operations related to state changes of the second item;
- training, using the data set, an empirical Bayesian forecasting model employing a deconvolution algorithm, wherein the empirical Bayesian forecasting model generates, corresponding to an item of the item collection for which a record of an event is included in the data set, an estimate of a metric of a first type of asynchronous operation, wherein the estimate is based at least in part on (a) a category-level distribution of the first type of asynchronous operation, corresponding to a first category of a plurality of item categories of the item collection, (b) an item-level distribution of the first type of asynchronous operation, corresponding to the item, and (c) a category-versus-item adjustment between the category-level distribution and the item-level distribution;
- storing a trained version of the empirical Bayesian forecasting model;
- in response to a forecasting request pertaining to at least the first item, obtaining, using the trained version of the Bayesian forecasting model, a first estimated metric of a first type of asynchronous operation corresponding to the first item;
- and
- causing, based on the first estimated metric, one or more operations to be initiated with respect to at least some items of the item collection.
4 more elements
One or more non-transitory computer-accessible storage media storing program instructions that when executed on or across one or more processors cause one or more computer systems to:
- obtain a data set, comprising (a) records of a plurality of state change events, wherein individual ones of the state change events represent a state change of one or more items of an item collection, and (b) records of one or more types of asynchronous operations associated with state changes of at least some items of the item collection, wherein the item collection comprises a first item and a second item, and wherein, within the data set, the number of records indicating changes to the first item differs from the number of records indicating changes to the second item;
- train, using the data set, a Bayesian forecasting model employing a deconvolution algorithm, wherein the Bayesian forecasting model generates, corresponding to an item of the item collection whose state has changed, an estimate of a metric of a first type of asynchronous operation, wherein the estimate is based at least in part on (a) a category-level distribution of the first type of asynchronous operation, corresponding to a first category of a plurality of item categories of the item collection, (b) an item-level distribution of the first type of asynchronous operation, corresponding to the item, and (c) a category-versus-item adjustment between the category-level distribution and the item-level distribution;
- store a trained version of the Bayesian forecasting model;
- in response to a forecasting request pertaining to at least the first item, obtain, using the trained version of the Bayesian forecasting model, a first estimated metric of a first type of asynchronous operation corresponding to the first item;
- and
- cause, based on the first estimated metric, one or more operations to be initiated with respect to at least some items of the item collection.
3 more elements
Dependencies are inferred from the claim text.