# US 12,602,443 B2 · Semantic-aware next best action recommendation

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

- Number: US 12,602,443 B2
- Document ID: US12602443B2
- Status: Granted
- Kind: B2
- Application number: 18/978,333
- Filed: 2024-12-12
- Published: 2025-04-24
- Granted: 2026-04-14
- Earliest claimed: 2023-09-08
- Assignees: Microsoft Technology Licensing, LLC
- Applicants: Microsoft Technology Licensing, LLC
- Inventors: Guillaume Didier Jean-Marc Dufour; Yang Chen; Lukasz Janusz Karolewski
- Web page: https://patentgrep.com/patent/US12602443B2

## CPC

- G06F 16/335: Filtering based on additional data, e.g. user or group profiles (filtering in web context G06F16/9535, G06F16/9536)
- G06F 16/9535: Search customisation based on user profiles and personalisation
- G06F 16/90332: {Natural language query formulation or dialogue systems}
- G06F 16/9536: Search customisation based on social or collaborative filtering
- G06F 40/30: Semantic analysis
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

In an example embodiment, an embedding model is used to generate an embedding of a natural language searching goal specified by a user, the embedding representing user intent of the user. Playbooks in a database of playbooks are also run through the embedding model to generate an embedding for each playbook indicative of a meaning of each playbook. A semantic relationship score can then be computed for each combination of the natural language search goal and a playbook, using the embeddings. These semantic relationship scores can then be passed into a ranking machine learning model, along with measured success rates for the playbooks, to generate a ranking of the playbooks. Based on this ranking, a set of filters and action corresponding to at least one of the playbooks may then be recommended to the user.

## Claims

### Claim 1

A method, comprising:
receiving a natural language search objective via an interface;
generating a search intent embedding based on the natural language search objective using an embedding model;
generating a semantic relationship score representative of a semantic similarity between the search intent embedding and a playbook of a set of playbooks;
selecting a playbook from the set of playbooks based on the semantic relationship score, the playbook comprising a set of filters;
converting the search intent embedding to a set of search criteria using the set of filters;
identifying a set of content using the search criteria; and
presenting the set of content at the user interface.

### Claim 2 (depends on 1)

The method of claim 1, wherein the embedding model is a generative artificial intelligence (GAI) model.

### Claim 3 (depends on 1)

The method of claim 1, comprising:
retrieving an action from the playbook; and
presenting the action at the user interface.

### Claim 4 (depends on 1)

The method of claim 1, comprising:
generating a playbook embedding for each playbook in the set of playbooks;
generating a semantic relationship score for each combination of the search intent embedding and the playbook embedding;
ranking the set of playbooks based on the semantic relationship scores; and
selecting the playbook from the set of playbooks with the highest rank.

### Claim 5 (depends on 4)

The method of claim 4, comprising ranking the set of playbooks based on the semantic relationship scores using a multi-tower neural network machine learning model.

### Claim 6 (depends on 4)

The method of claim 4, comprising ranking the set of playbooks based on the semantic relationship scores and a success rate for each of the playbooks.

### Claim 7 (depends on 1)

The method of claim 1, comprising:
searching for content using the search criteria to form a set of search results; and
identifying the set of content from the set of search results.

### Claim 8

An apparatus, comprising:
circuitry; and
memory storing instructions that when executed by the circuitry causes the circuitry to:
receive a natural language search objective via an interface;
generate a search intent embedding based on the natural language search objective using an embedding model;
generate a semantic relationship score representative of a semantic similarity between the search intent embedding and a playbook of a set of playbooks;
select a playbook from the set of playbooks based on the semantic relationship score, the playbook comprising a set of filters;
convert the search intent embedding to a set of search criteria using the set of filters;
identify a set of content using the search criteria; and
present the set of content at the user interface.

### Claim 9 (depends on 8)

The apparatus of claim 8, wherein the embedding model is a generative artificial intelligence (GAI) model.

### Claim 10 (depends on 8)

The apparatus of claim 8, the circuitry to:
retrieve an action from the playbook; and
present the action at the user interface.

### Claim 11 (depends on 8)

The apparatus of claim 8, the circuitry to:
generate a playbook embedding for each playbook in the set of playbooks;
generate a semantic relationship score for each combination of the search intent embedding and the playbook embedding;
rank the set of playbooks based on the semantic relationship scores; and
select the playbook from the set of playbooks with the highest rank.

### Claim 12 (depends on 11)

The apparatus of claim 11, the circuitry to rank the set of playbooks based on the semantic relationship scores using a multi-tower neural network machine learning model.

### Claim 13 (depends on 11)

The apparatus of claim 11, the circuitry to rank the set of playbooks based on the semantic relationship scores and a success rate for each of the playbooks.

### Claim 14 (depends on 8)

The apparatus of claim 8, the circuitry to:
search for content using the search criteria to form a set of search results; and
identify the set of content from the set of search results.

### Claim 15

A non-transitory machine readable medium comprising instructions that when executed by circuitry causes the circuitry to:
receive a natural language search objective via an interface;
generate a search intent embedding based on the natural language search objective using an embedding model;
generate a semantic relationship score representative of a semantic similarity between the search intent embedding and a playbook of a set of playbooks;
select a playbook from the set of playbooks based on the semantic relationship score, the playbook comprising a set of filters;
convert the search intent embedding to a set of search criteria using the set of filters;
identify a set of content using the search criteria; and
present the set of content at the user interface.

### Claim 16 (depends on 15)

The machine readable medium of claim 15, wherein the embedding model is a generative artificial intelligence (GAI) model.

### Claim 17 (depends on 15)

The machine readable medium of claim 15, comprising instructions that when executed by circuitry causes the circuitry to:
retrieve an action from the playbook; and
present the action at the user interface.

### Claim 18 (depends on 15)

The machine readable medium of claim 15, comprising instructions that when executed by circuitry causes the circuitry to:
generate a playbook embedding for each playbook in the set of playbooks;
generate a semantic relationship score for each combination of the search intent embedding and the playbook embedding;
rank the set of playbooks based on the semantic relationship scores; and
select the playbook from the set of playbooks with the highest rank.

### Claim 19 (depends on 18)

The machine readable medium of claim 18, comprising instructions that when executed by circuitry causes the circuitry to rank the set of playbooks based on the semantic relationship scores using a multi-tower neural network machine learning model.

### Claim 20 (depends on 18)

The machine readable medium of claim 18, comprising instructions that when executed by circuitry causes the circuitry to rank the set of playbooks based on the semantic relationship scores and a success rate for each of the playbooks.