Determining user-personalized target values of products using machine learning models
A set of inputs from a plurality of inputs are selected based on a user identifier of a user. The set of inputs are apportioned and recombined to produce one or more grouped inputs. The one or more grouped inputs are input to one or more machine learning models to provide individualized information output by the one or more machine learning models to the user.
1 . A computer-implemented method performed by a service provider to estimate target values for items, comprising:
obtaining, from a user of the service provider, a request to estimate a set of target values for a set of items that the user provides to clients of the user, the request including a user identifier usable by the service provider to:
identify the user from a plurality of users of the service provider;
identify, from a plurality of machine learning models of the service provider trained to output target value estimations, one or more machine learning models associated with the user; and
identify a set of input types used by the one or more machine learning models to estimate the target values;
obtaining a set of input values corresponding the set of input types;
apportioning and combining the set of inputs according to predetermined groups to produce one or more grouped inputs;
monitoring performance of the one or more machine learning models associated with the user based on feedback signals and measurements from prior estimations to detect model performance drift;
updating the one or more machine learning models associated with the user by retraining with updated inputs and deploying an updated model instance responsive to the detected model performance drift;
inputting the one or more grouped inputs to the updated model instance;
obtaining, as output from the updated model instance, a set of target values for the set of items; and
providing the set of target values to the user.
2 . The computer-implemented method of claim 1 , wherein the set of input types includes one or more of:
a list of the set of items,
geographical information relating to a geographical location of the user, client-specific information, or
a set of initial values for the set of items.
3 . The computer-implemented method of claim 1 , wherein the updated model instance is trained to output one or more of:
an estimated target value for an item,
a recommendation for the item,
a predicted side effect of the item, or
a predicted interaction between the item and another item.
4 . The computer-implemented method of claim 1 , wherein:
obtaining the request includes receiving the request at one or more application programming interfaces (APIs); and
the APIs assign the set of inputs and the user identifier to one or more components of a system implementing the computer-implemented method.
5 . A system, comprising:
one or more processors; and
memory that stores computer-executable instructions that, if executed by the one or more processors, cause the system to:
select a set of inputs from a plurality of inputs based on a user identifier of a user;
apportion and recombine the set of inputs to produce one or more grouped inputs;
monitor performance of one or more machine learning models associated with the user based on feedback signals and measurements from prior predictions to detect model performance drift;
update the one or more machine learning models associated with the user by retraining with updated inputs and deploying an updated model instance responsive to the detected model performance drift;
input the one or more grouped inputs to the updated model instance, the updated model instance trained to generate predictions using the plurality of inputs; and
provide individualized information that is output by the updated model instance to the user.
6 . The system of claim 5 , wherein the executable instructions that apportion and recombine the set of inputs include instructions that apportion and recombine the set of inputs using criteria included in the plurality of inputs.
7 . The system of claim 5 , wherein the updated model instance further:
receive outputs from one or more additional machine learning models; and
use the outputs, in addition to the one or more grouped inputs, to determine the individualized information.
8 . The system of claim 7 , wherein the outputs from the one or more additional machine learning models include one or more of:
an item cluster,
a characterization map, or
an initial value of an item.
9 . The system of claim 5 , wherein the individualized information includes one or more of:
a target value for an item,
a comparison of other values for the item, or
a set of alternative items to the item.
10 . The system of claim 5 , wherein the individualized information includes a target item value generated by the updated model instance using an iterative process, the iterative process being used to re-compute a computed item value until a distance between the computed item value and optimization criteria no longer decreases.
11 . The system of claim 5 , wherein the individualized information includes a target item value generated using historical values that are included in the plurality of inputs.
12 . The system of claim 5 , wherein the individualized information includes a target item value generated using user similarity criteria included in the plurality of inputs.
13 . The system of claim 12 , wherein the system further implements an additional one or more machine learning models to:
estimate associated inputs from the plurality of inputs, the associated inputs being applicable to users assigned to a cluster of the user; and
input the associated inputs to the updated model instance.
14 . A non-transitory computer-readable storage medium storing executable instructions that, as a result of being executed by one or more processors of a computer system, cause the computer system to at least:
obtain a request from a requestor to generate a target value for an item, the request including a user identifier and a plurality of inputs;
select one or more machine learning models and a set of inputs from the plurality of inputs using the user identifier, the one or more machine learning models having been trained to generate target values using the plurality of inputs;
apportion and recombine the set of inputs based on the user identifier, the one or more machine learning models, and predetermined categories to produce grouped inputs;
monitor performance of the one or more machine learning models based on feedback signals and measurements from prior target values to detect model performance drift;
update the one or more machine learning models by retraining with updated inputs and deploying an updated model instance responsive to the detected model performance drift;
input the grouped inputs to the updated model instance; and
obtain the target value as an output from the updated model instance to display the target value to the requestor.
15 . The non-transitory computer-readable storage medium of claim 14 , wherein:
the set of inputs comprise heterogeneous data obtained from different sources.
16 . The non-transitory computer-readable storage medium of claim 14 , wherein:
the set of inputs are assimilated and modified for apportioning and recombining based on one or more of:
format,
syntax, or p 1 language.
17 . The non-transitory computer-readable storage medium of claim 14 , wherein the requestor is:
a user of a service provider that hosts the computer system, or
a client of the user.
18 . The non-transitory computer-readable storage medium of claim 14 , wherein:
the executable instructions that cause the computer system to obtain the request include instructions that cause the computer system to receive the request via at least one application programming interface (API) call from a user interface; and
the executable instructions further include instructions that further cause the computer system to provide, in response to the at least one API call, the target value to the user interface for display.
19 . The non-transitory computer-readable storage medium of claim 14 , wherein the target value is generated in real-time in response to obtaining the request.
20 . The non-transitory computer-readable storage medium of claim 14 , wherein the executable instructions further include instructions that further cause the computer system to:
select one or more additional machine learning models that:
use the plurality of inputs to group items of an item list into item clusters; and
output the item clusters to the updated model instance; and
cause the updated model instance:
receive at least one of the item clusters; and
generate an estimated value for the item cluster.