# US 12,770,173 B2 · Item embeddings for machine learning systems

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 12,770,173 B2
- Document ID: US12770173B2
- Status: Granted
- Kind: B2
- Application number: 18/055,524
- Filed: 2022-11-15
- Published: 2024-05-16
- Granted: 2026-10-20
- Earliest claimed: 2022-11-15
- Assignees: Paypal, Inc.
- Applicants: PayPal, Inc.
- Inventors: Yarden Raiskin; Yuval Yaron
- Web page: https://patentgrep.com/patent/US12770173B2

## CPC

- G06Q 30/0631: {Recommending goods or services}
- G06Q 30/0629: {by pre-processing results, e.g. ranking or ordering results}
- G06Q: INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR

## Abstract

Techniques are disclosed relating to item representations. A computer system may access information identifying a first set of items and generate a representation of the first set of items that positions them in an embedding space. The computer system may send a request to another computer system for information pertaining to an item selected from the first set of items and receive correlation information that identifies recorded user behavior indicative of correlations between a second set of items and the selected item. The computer system may update the representation based on the correlation information, such that at least one of the first set of items that is included in the second set is moved closer to the selected item in the embedding space and at least one of the first set of items that is not included in the second set is moved farther away from the selected item.

## Claims

### Claim 1

A method, comprising:
accessing, by a first computer system, item information that identifies a first set of items;
performing, by the first computer system, an initial embedding operation using a machine learning model and categorical data about the first set of items to generate an initial representation of the first set of items that positions the first set of items in an embedding space as embeddings;
sending, by the first computer system and to a second computer system that processes user search queries, a request for information pertaining to an item selected from the first set of items;
receiving, by the first computer system and from the second computer system, correlation information identifying recorded user behavior indicative of correlations between a second set of items and the selected item, wherein the recorded user behavior is based on the user search queries;
based on the correlation information, the first computer system generating an updated version of the initial representation, wherein the generating includes performing multiple iterations in which different ones of the second set of items are selected for processing, and wherein a given iteration of the multiple iterations includes:
selecting a first item from the second set of items;
selecting a second item from the first set of items that is not included in the second set of items;
decreasing a distance between an embedding of the first item and an embedding of the selected item in the embedding space;
increasing a distance between an embedding of the second item and the embedding of the selected item in the embedding space; and
generating, by the first computer system based on the updated version, a recommendation based on two or more items of the first set of items being indicated as similar based on embeddings of the two or more items being positioned within a proximity threshold in the embedding space.

### Claim 2 (depends on 1)

The method of claim 1, wherein the embedding space is a vector space, and wherein the machine learning model embeds the first set of items into the vector space as vectors embeddings.

### Claim 3 (depends on 1)

The method of claim 1, wherein the recorded user behavior corresponds to web searches that are performed by a set of users with respect to the second set of items and the selected item.

### Claim 4 (depends on 1)

The method of claim 1, wherein the second set of items includes at least one item that is not included in the first set of items.

### Claim 5 (depends on 1)

The method of claim 1, wherein the second computer system implements a search engine that derives the recorded user behavior based on user interactions with the search engine.

### Claim 6 (depends on 1)

The method of claim 1, further comprising:
implementing, by the first computer system, one or more services that pertain to the first set of items, wherein the categorical data is stored by the first computer system as a part of providing the one or more services.

### Claim 7 (depends on 1)

The method of claim 1, wherein the generating of the updated version of the initial representation includes performing multiple iterations in which embeddings of a third set of items and an embedding of a second selected item selected from the first set of items are moved closer in the embedding space and embeddings of ones of the first set of items not included in the third set and the embedding of the second selected item are moved farther away from each other in the embedding space.

### Claim 8 (depends on 1)

The method of claim 1, wherein the correlation information identifies a particular item that is not included in the first set of items, and wherein the generating of the updated version of the initial representation includes embedding the particular item into the embedding space.

### Claim 9

A non-transitory computer-readable medium having program instructions stored thereon that are executable to cause a first computer system to perform operations comprising:
accessing item information that identifies a set of items;
generating, based on categorical data about the set of items, an initial representation of the set of items that positions the set of items in an embedding space as embeddings;
obtaining correlation information for at least two items of the set of items, wherein the correlation information is based on recorded user behavior and identifies, for a particular one of the at least two items, one or more similar items to that particular item, and wherein the recorded user behavior is based on a monitoring of a set of user interactions that includes search queries issued by a set of users in connection with the particular item and the one or more similar items; and
generating, based on the correlation information, an updated version of the initial representation that decreases a distance between an embedding of at least one of the set of items that is included in the one or more similar items and an embedding of the particular item in the embedding space and increases a distance between an embedding of at least one of the set of items that is not included in the one or more similar items and the embedding of the particular item in the embedding space; and
generating, based on the updated version, a recommendation based on two or more items of the set of items being indicated as similar based on embeddings of the two or more items being positioned within a proximity threshold in the embedding space.

### Claim 10 (depends on 9)

The medium of claim 9, wherein the generating of the initial representation is performed using a machine learning model that embeds the set of items in the embedding space based on the categorical data.

### Claim 11 (depends on 9)

The medium of claim 9, wherein the correlation information identifies an item that is not included in the set of items, and wherein the generating of the updated version of the initial representation includes embedding the item into the embedding space.

### Claim 12 (depends on 9)

The medium of claim 9, wherein the recorded user behavior is based on Global Positioning System (GPS) data collected by a second computer system for particular users.

### Claim 13

A system, comprising:
at least one processor; and
a memory having program instructions stored thereon that are executable by the at least one processor to perform operations comprising:
accessing item information that identifies a first set of items;
performing an initial embedding of the first set of items into a vector space as a set of embeddings based on categorical data collected about the first set of items;
sending a request to a different system for information relating to an item selected from the first set of items;
receiving, from the different system, correlation information that identifies recorded user behavior that is indicative of correlations between a second set of items and the selected item; and
based on the correlation information, modifying a particular one of the set of embeddings that corresponds to the selected item such that the particular embedding is moved closer in the vector space to ones of the set of embeddings that correspond to the second set of item and farther away from ones of the set of embeddings that correspond to those ones of the first set of items that are not included in the second set of items; and
producing a recommendation based on two or more items of the first set of items being indicated as similar based on embeddings of the two or more items being positioned within a proximity threshold in the vector space.

### Claim 14 (depends on 13)

The system of claim 13, wherein the initial embedding of the first set of items is performed using a machine learning model that produces the set of embeddings based on categorical data collected about the first set of items.

### Claim 15 (depends on 13)

The system of claim 13, wherein the operations further comprise:
embedding, into the vector space, at least one item of the second set of items that is not included in the first set of items.

### Claim 16 (depends on 13)

The system of claim 13, wherein the first set of items are positioned randomly in the vector space as a part of performing the initial embedding.
