Granted Patent B2

US 12,591,802 · App. 17/491,240 · Granted Mar 31, 2026

Inventors: Jingxian Liao (San Jose, CA); Wei Wang (San Jose, CA); Zhicheng Xue (Union City, CA)

Assignee: Intuit Inc.

G06N20/00G06F18/2148G06F18/24137

Abstract

Claims (102)

1 . A method comprising:

obtaining a first cluster of a first plurality of clusters, wherein the first cluster comprises a first subset of a plurality of reference entities;

calculating a first plurality of distances between a first plurality of features of a target entity and the first plurality of features of the first subset of the plurality of reference entities;

selecting, from the first subset, a first plurality of peer entities that are closest, within the first subset, to the target entity according to the first plurality of distances, each peer entity in the first plurality of peer entities being labeled with a corresponding value of a metric of the peer entity;

training, after selecting, a machine learning model specifically for the target entity, wherein the machine learning model is trained to estimate, as output, the corresponding value of the metric of the peer entity using, as input, values of the first plurality of features from only the first plurality of peer entities;

generating, by applying the machine learning model to the first plurality of features of the target entity, and after training the machine learning model specifically for the target entity, a first estimated value of the metric for the target entity; and

presenting the first estimated value of the metric.

2 . The method of claim 1 , wherein the first plurality of clusters comprises a plurality of centroids, and wherein the target entity and the plurality of reference entities are comprised by a plurality of entities, the method further comprising:

generating the first plurality of features for the plurality of entities;

dividing, using the first plurality of features, the plurality of reference entities into the first plurality of clusters;

calculating a second plurality of distances between the first plurality of features of the target entity and the plurality of centroids; and

selecting, using the second plurality of distances, the first cluster.

3 . The method of claim 2 , wherein the first plurality of features comprises a plurality of transaction aggregations and a plurality of operational features, the method further comprising:

aggregating the plurality of transaction aggregations and the plurality of operational features to obtain plurality of aggregated features,

wherein the dividing is based on the plurality of aggregated features.

4 . The method of claim 3 , wherein the plurality of entities is associated with a plurality of transactions, the method further comprising:

dividing the plurality of transactions into a plurality of categories;

selecting a subset of the plurality of categories that are relevant to the metric; and

generating the plurality of transaction aggregations by aggregating the plurality of transactions in the subset of the plurality of categories.

5 . The method of claim 2 ,

wherein the first plurality of features comprises a plurality of entity attributes, and

wherein the dividing is based on a subset of the plurality of entity attributes.

6 . The method of claim 2 , wherein the first plurality of features is generated for a first time window, the method further comprising:

generating a second plurality of features for the plurality of entities for a second time window;

dividing, using the second plurality of features, the plurality of reference entities into a second plurality of clusters, wherein the second plurality of clusters comprises a second cluster comprising a second subset of the plurality of reference entities;

calculating a third plurality of distances between a second plurality of features of the target entity and the second plurality of features of the second subset;

selecting, from the second subset and using the third plurality of distances, a second plurality of peer entities;

generating, by applying the machine learning model to the second plurality of features of the target entity, a second estimated value of the metric; and

combining the first estimated value and the second estimated value.

7 . The method of claim 6 , further comprising:

further training the machine learning model using second training data comprising a plurality of values of the second plurality of features for the second plurality of peer entities labeled with a value of the metric.

8 . The method of claim 1 , further comprising:

receiving a target entity request, wherein the first cluster is obtained in response to receiving the target entity request; and

recommending a response to the target entity request using the first estimated value of the metric.

9 . A system comprising:

a computer processor;

a repository configured to store:

a first plurality of clusters comprising a first cluster comprising a first subset of a plurality of reference entities; and

a metric estimation manager executing on the computer processor and configured to:

calculate a first plurality of distances between a first plurality of features of a target entity and the first plurality of features of the first subset,

select, from the first subset, a first plurality of peer entities that are closest, within the first subset, to the target entity according to the first plurality of distances, each peer entity in the first plurality of peer entities being labeled with a corresponding value of a metric of the peer entity,

train, after selecting, a machine learning model specifically for the target entity, wherein the machine learning model is trained to estimate, as output, the corresponding value of the metric of the peer entity using, as input, values of the first plurality of features labeled from only the first plurality of peer entities,

generate, by applying the machine learning model to the first plurality of features of the target entity, and after training the machine learning model specifically for the target entity, a first estimated value of the metric for the target entity, and

present the first estimated value of the metric.

10 . The system of claim 9 , further comprising a feature generator executing on the computer processor,

wherein the first plurality of clusters comprises a plurality of centroids,

wherein the target entity and the plurality of reference entities are comprised by a plurality of entities,

wherein the feature generator is configured to generate the first plurality of features for the plurality of entities, and

wherein the metric estimation manager is further configured to:

divide, using the first plurality of features for the plurality of entities, the plurality of reference entities into the first plurality of clusters,

calculate a second plurality of distances between the first plurality of features of the target entity and the plurality of centroids, and

select, using the second plurality of distances, the first cluster.

11 . The system of claim 10 , wherein the first plurality of features comprises a plurality of transaction aggregations and a plurality of operational features, and wherein the feature generator is further configured to:

aggregate the plurality of transaction aggregations and the plurality of operational features to obtain a plurality of aggregated features, wherein the plurality of entities is divided into the first plurality of clusters based on the plurality of aggregated features.

12 . The system of claim 11 , further comprising a transaction categorizer executing on the computer processor,

wherein the plurality of entities is associated with a plurality of transactions,

wherein the transaction categorizer is configured to:

divide the plurality of transactions into a plurality of categories, and

select a subset of the plurality of categories that are relevant to the metric, and

wherein the metric estimation manager is further configured to generate the plurality of transaction aggregations by aggregating transactions in the subset of the categories.

13 . The system of claim 10 ,

wherein the first plurality of features comprises a plurality of entity attributes, and

wherein the plurality of entities is divided into the first plurality of clusters based on a subset of the plurality of entity attributes.

14 . The system of claim 10 ,

wherein the first plurality of features is generated for a first time window,

wherein the feature generator is further configured to generate a second plurality of features for the plurality of entities for a second time window, and

wherein the metric estimation manager is further configured to:

divide, using the second plurality of features, the plurality of reference entities into a second plurality of clusters, wherein the second plurality of clusters comprises a second cluster comprising a second subset of the plurality of reference entities;

calculate a third plurality of distances between a second plurality of features of the target entity and the second plurality of features of the second subset;

select, from the second subset and using the third plurality of distances, a second plurality of peer entities;

generate, by applying the machine learning model to the second plurality of features of the target entity, a second estimated value of the metric; and

combine the first estimated value and the second estimated value.

15 . The system of claim 14 , wherein the machine learning model is further trained using second training data comprising a plurality of values of the second plurality of features for the second plurality of peer entities labeled with a value of the metric.

16 . The system of claim 9 , wherein the metric estimation manager is further configured to:

receive a target entity request, wherein the first cluster is obtained in response to receiving the target entity request; and

recommend a response to the target entity request using the first estimated value of the metric.

17 . A method comprising:

sending a target entity request to a metric estimation manager configured to perform:

obtaining a first cluster of a first plurality of clusters, wherein the first cluster comprises a first subset of a plurality of reference entities,

calculating a first plurality of distances between a first plurality of features of a target entity and the first plurality of features of the first subset of the plurality of reference entities,

selecting, from the first subset, a first plurality of peer entities that are closest, within the first subset, to the target entity according to the first plurality of distances, each peer entity in the first plurality of peer entities being labeled with a corresponding value of a metric of the peer entity,

training, after selecting, a machine learning model specifically for the target entity, wherein the machine learning model is trained to estimate, as output, the corresponding value of the metric of the peer entity using, as input, values of the first plurality of features from only the first plurality of peer entities;

generating, by applying the machine learning model to the first plurality of features of the target entity, and after training the machine learning model specifically for the target entity, a first estimated value of the metric for the target entity, and

transmitting the first estimated value of the metric;

receiving, from the metric estimation manager, the first estimated value of the metric; and

presenting the first estimated value of the metric.

18 . The method of claim 17 , wherein the first plurality of clusters comprises a plurality of centroids, wherein the target entity and the plurality of reference entities are comprised by a plurality of entities, and wherein the metric estimation manager is further configured to perform:

generating a first plurality of transaction aggregations and a first plurality of operational features for the plurality of entities;

aggregating the first plurality of transaction aggregations and the first plurality of operational features to obtain a first plurality of aggregated features;

dividing, using the first plurality of aggregated features, the plurality of reference entities into the first plurality of clusters;

calculating a second plurality of distances between the first plurality of features of the target entity and the plurality of centroids; and

selecting, using the second plurality of distances, the first cluster.

19 . The method of claim 18 , wherein the first plurality of aggregated features is generated for a first time window, and wherein the metric estimation manager is further configured to perform:

generating a second plurality of transaction aggregations and a second plurality of operational features for the plurality of entities;

aggregating the second plurality of transaction aggregations and the second plurality of operational features to obtain a second plurality of aggregated features;

dividing, using the second plurality of aggregated features, the plurality of reference entities into a second plurality of clusters, wherein the second plurality of clusters comprises a second cluster comprising a second subset of the plurality of reference entities;

calculating a third plurality of distances between the second plurality of aggregated features of the target entity and the second plurality of aggregated features of the second subset;

selecting, from the second subset and using the third plurality of distances, a second plurality of peer entities;

generating, by applying the machine learning model to a second plurality of features of the target entity, a second estimated value of the metric; and

combining the first estimated value and the second estimated value.

20 . The method of claim 19 , wherein the metric estimation manager is further configured to perform:

further training the machine learning model using second training data comprising values of the second plurality of features for the second plurality of peer entities labeled with a value of the metric.

Assignments (1)

Continuity (1)

References Cited (26)