Generating recommendations based on descriptors in a multi-dimensional search space

Patent metadata

Field	Value
Publication number	US-11922476-B2
Application number	US-202117365272-A
Country	US
Kind code	B2
Filing date	Jul 1, 2021
Priority date	Jul 1, 2021
Publication date	Mar 5, 2024
Grant date	Mar 5, 2024

Abstract

Aspects described herein may relate to methods, systems, and apparatuses that provide new capabilities for recommending purchases to a user based on a new descriptor referred to as “purchase embeddings.” The purchase embedding may include a tuple in a new multi-dimensional search space for describing and tracking purchases of goods and services. Recommendations may be based on a distance between tuples in the search space, which provides a measure of similarity between items represented by the tuples.

First claim

We claim:

1. A computer implemented method for characterizing items using tuples in a multi-dimensional search space, the method comprising:
retrieving, from one or more databases, a training set comprising a plurality of known tuples each paired with associated parameters;
applying a machine learning algorithm to the training set to generate a classifier model;
storing the classifier model in the one or more databases;
acquiring, by a server via a computer network, transaction records from remote computing devices, wherein the transaction records identify purchases of the items;
extracting, from the transaction records, purchase parameters characterizing the purchases of the items, wherein the purchase parameters are variables of the classifier model;
generating, by the server using the classifier model and based on the purchase parameters, computed the tuples respectively representing the items in the multi-dimensional search space, wherein a measure of similarity between two of the items is indicated by a distance in the multi-dimensional search space between two of the computed tuples respectively representing the two of the items;
compiling item records, each including a different one of the computed tuples; and
storing the item records in one or more databases.

2. The computer implemented method of claim 1 , wherein the distance in the multi-dimensional search space is an inner product between the two of the computed tuples.

3. The computer implemented method of claim 1 , further comprising:
receiving a search request comprising one or more search parameters;
determining, based on the one or more search parameters, a search tuple representing the search request in the multi-dimensional search space;
calculating one or more distances in the multi-dimensional search space between the search tuple representing the search request and one or more of the computed tuples representing the items;
identifying, based on the calculating, a cluster in the multi-dimensional search space comprising the search tuple representing the search request and a subset of the one or more of the computed tuples representing the items;
generating a search result based on the subset; and
transmitting, from the server via the computer network, the search result to a user computing device.

4. The computer implemented method of claim 3 , further comprising:
determining, from among the item records, a first subset of item records associated a particular purchaser;
determining, from the first subset of item records, a user preference; and
generating the search request based on the user preference.

5. The computer implemented method of claim 3 , further comprising:
receiving purchase history parameters associated with the user computing device;
filtering, based on the purchase history parameters, the item records stored in the one or more databases; and
determining, based on the filtering, the one or more of the computed tuples representing the items for calculating the one or more distances in the multi-dimensional search space.

6. The computer implemented method of claim 1 , wherein compiling the item records comprises:
identifying, from the purchase parameters, a subset of the purchase parameters characterizing purchasers respectively of the items;
associating, with each of the computed tuples, one purchase parameter of the subset of the purchase parameters; and
including, in each of the item records, the one purchase parameter of the subset of the purchase parameters respectively associated with the computed tuple in that item record.

7. The computer implemented method of claim 1 , further comprising:
including, in one of the item records, one of the purchase parameters that indicates:
a geographic region where an item represented by the computed tuple in the one of the item records is available for purchase, or
a purchaser preference related to the item represented by the computed tuple in the one of the item records.

8. The computer implemented method of claim 1 , wherein the transaction records comprise:
a purchase receipt acquired from a purchaser computing device,
a vendor sales record from a vendor computing device, or
a payment record from a payment service computing device.

9. The computer implemented method of claim 1 , wherein the purchase parameters comprise:
a parameter characterizing one of the items, or
characterizing transaction information for one of the purchases.

10. The computer implemented method of claim 1 , further comprising:
identifying, from the purchase parameters, information associated with a purchaser of one of the items;
querying, based on the information associated with the purchaser, the one or more databases for an auxiliary record;
receiving, based on querying, the auxiliary record, wherein the auxiliary record comprises one or more auxiliary parameters characterizing the purchaser; and
including one of the one or more auxiliary parameters in one of the item records comprising the computed tuple representing the one of the items.

11. A server comprising:
at least one computer processor; and
computer memory comprising computer-executable instructions that when executed by the at least one computer processor, cause the server to:
query one or more databases for item records associated with a user, wherein each of the item records associated with the user comprises:
a parameter associated with the user, and
a tuple of a plurality of tuples,

wherein the plurality of tuples respectively represent, in a multi-dimensional search space, a plurality of purchasable items, and wherein a measure of similarity between any two items of the plurality of purchasable items is indicated by a distance between two tuples of the plurality of tuples respectively representing the two items;
receive, in response to the query of the one or more databases, the item records associated with the user;
extract a first subset of tuples from the item records associated with the user;
generate a search tuple based on the first subset of tuples;
calculate distances in the multi-dimensional search space between the search tuple and other tuples respectively in other item records in the one or more databases;
receive a geographic location of a user computing device associated with the user;
identify, based on the distances that were calculated and the geographic location, a second subset of tuples of the plurality of tuples within a predetermined distance to the search tuple;
derive, from the second subset of tuples, vendor locations where items represented by the second subset of tuples are available; and
generate a search result based on the second subset of tuples, wherein the search result includes the vendor locations; and
transmit the search result to a user computing device.

12. The server of claim 11 , wherein the distance between the two tuples of the plurality of tuples respectively representing the two items is:
an inner product,
a Euclidean distance,
a squared Euclidean distance,
a Manhattan distance,
a Minkowski distance,
a Chebyshev distance,
an edit distance, or
a Levenshtein distance.

13. The server of claim 11 , wherein to generate the search tuple based on the first subset of tuples, the computer-executable instructions, when executed by the at least one computer processor, cause the server to:
identify a cluster in the multi-dimensional search space that includes the first subset of tuples; and
select the search tuple from within the cluster.

14. The server of claim 13 , wherein the first subset of tuples define vertices of the cluster, and w

Assignees

• Capital One Services Llc

Inventors

• Key Kate
• Pham Vincent
• Goodsitt Jeremy
• Walters Austin
• Rafferty Galen
• Truong Anh

Classifications

• G06Q30/0625 Primary

  by formulating product or service queries, e.g. using keywords or predefined options · CPC title