# US 2022/0044139 A1 · Search system and corresponding method

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

- Number: US 2022/0044139 A1
- Document ID: US20220044139A1
- Status: Published application
- Kind: A1
- Application number: 17/462,339
- Filed: 2021-08-31
- Published: 2022-02-10
- Earliest claimed: 2012-12-27
- Assignees: Microsoft Technology Licensing, LLC
- Applicants: Microsoft Technology Licensing, LLC
- Inventors: Benjamin William Medlock; David Paul Williams; James Aley
- Web page: https://patentgrep.com/patent/US20220044139A1

## CPC

- G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks
- G06F 3/0484: for the control of specific functions or operations, e.g. selecting or manipulating an object, an image or a displayed text element, setting a parameter value or selecting a range
- G06F 16/3346: {using probabilistic model}
- G06F 16/907: Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F 16/951: Indexing; Web crawling techniques
- G06F 16/9538: Presentation of query results
- G06N 20/00: Machine learning
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

There is provided a search system comprising a statistical model trained on text associated with a piece of content. The text associated with the piece of content is drawn from a plurality of different data sources. The system is configured to receive text input and generate using the statistical model an estimate of the likelihood that the piece of content is relevant given the text input. A corresponding method is also provided.

## Claims

### Claim at position 1

-33. (canceled)

### Claim 34

A system comprising:
a processor; and
non-transitory memory including instructions, which when executed by the processor, cause the processor to:
receive text input;
identify a content type indicated by the text input using a classifier model, the content type identifying a subject matter of the text input representable in a plurality of representation formats;
in response to identifying the content type for the text input, select a single statistical model from among a set of statistical models associated with respective content types, based on the identified content type matching a content type of the single statistical model;
generate, with the selected statistical model, a relevance estimate for a content item in relation to the text input, wherein the selected statistical model is trained with text associated with the content item drawn from a plurality of different data sources and a relevance value; and
output the content item as a content suggestion when the relevance estimate of the content item exceeds a relevance threshold.

### Claim 35 (depends on 34)

The system of claim 34, wherein the statistical model is trained on one or more features extracted from the text associated with the content item, and wherein execution of the program instructions by the one or more processors causes the one or more processors to:
extract the one or more features from the text input; and
query the statistical model with each of the one or more features of the text input to generate the relevance estimate for the content item in relation to the text input.

### Claim 36 (depends on 34)

The system of claim 34, wherein execution of the program instructions by the one or more processors causes the one or more processors to:
generate a respective relevance estimate for each content item of a plurality of content items in relation to the text input using a statistical model, associated with each content item, selected from a plurality of statistical models, and trained on text associated with pieces of content items; and
rank the plurality of content items by the respective relevance estimate and determine one or more of the most relevant content items based on the relevance estimate.

### Claim 37 (depends on 36)

The system of claim 36, wherein execution of the program instructions by the one or more processors causes the one or more processors to output at least one representation of each of the one or more most relevant content items.

### Claim 38 (depends on 36)

The system of claim 36, wherein at least one content item is associated with a particular entity and wherein execution of the program instructions by the one or more processors causes the one or more processors to output at least one representation of one or more entities associated with the one or more most relevant content items.

### Claim 39 (depends on 36)

The system of claim 36, wherein execution of the program instructions by the one or more processors causes the one or more processors to:
classify, by content type, each of the plurality of statistical models trained on text;
classify the text input as a content type; and
determine a subset of the plurality of statistical models which are classified the same content type as the content type of the text input.

### Claim 40 (depends on 34)

The system of claim 34, wherein execution of the program instructions by the one or more processors causes the one or more processors to:
compare a portion of the text input to a first language model trained on natural language text;
compare the portion of the text input to a second language model trained on conversational text; and
classify the portion of the text input as natural language or conversational.

### Claim 41 (depends on 34)

The system of claim 34, wherein the plurality of representation formats include at least one of a text file, an image, a video clip, or a URL.

### Claim 42

A method for determining whether a content item is relevant to text input, the method comprising:
receiving text input;
identifying a content type indicated by the text input using a classifier model, the content type identifying a subject matter of the text input representable in a plurality of representation formats;
in response to identifying the content type for the text input, selecting a single statistical model from among a set of statistical models associated with respective content types based on the identified content type matching a content type of the single statistical model;
generating, with the selected statistical model, a relevance estimate for a content item in relation to the text input, wherein the selected statistical model is trained with text associated with the content item drawn from a plurality of different data sources and a relevance value; and
outputting the content item as a content suggestion when the relevance estimate of the content item exceeds a relevance threshold.

### Claim 43 (depends on 42)

The method of claim 42, further comprising:
extracting, using one or more processors, one or more features from the text input; and
querying, using one or more processors, the statistical model with each of the one or more feature of the text input to generate the relevance estimate for the content item in relation to the text input.

### Claim 44 (depends on 42)

The method of claim 42, further comprising:
comparing a portion of the text input to a first language model trained on natural language text;
comparing the portion of the text input to a second language model trained on conversational text; and
classifying the portion of the text input as natural language or conversational.

### Claim 45 (depends on 44)

The method of claim 44, further comprising discarding the portion of the text input if it is classified as conversational.

### Claim 46 (depends on 42)

The method of claim 42, further comprising:
classifying, by content type, each of a plurality of statistical models trained on text;
classifying the text input as a content type; and
determining a subset of the plurality of statistical models which are classified the same content type as the content type of the text input.

### Claim 47

A non-transitory machine-readable medium including instructions for determining whether a content item is relevant to text input, which when executed by a processor, cause the processor to:
receive the text input;
identify a content type indicated by the text input using a classifier model, the content type identifying a subject matter of the text input representable in a plurality of representation formats;
in response to identifying the content type for the text input, select a single statistical model from among a set of statistical models associated with respective content types, based on the identified content type matching a content type of the single statistical model;
generate, with the selected statistical model, a relevance estimate for the content item in relation to the text input, wherein the selected statistical model is trained with text associated with the content item drawn from a plurality of different data sources and a relevance value; and
output the content item as a content suggestion when the relevance estimate of the content item exceeds a relevance threshold.

### Claim 48 (depends on 47)

The non-transitory machine-readable medium of claim 47, wherein the text input is not input into the computer system by a user.

### Claim 49 (depends on 47)

The non-transitory machine-readable medium of claim 47, wherein execution of the program instructions by the one or more processors causes the one or more processors to:
receive non-textual evidence; and
generate using a statistical model that is trained on non-textual data the relevance estimate for the content item in relation to the non-textual evidence.

### Claim 50 (depends on 47)

The non-transitory machine-readable medium of claim 47, wherein execution of the program instructions by the one or more processors causes the one or more processors to:
classify, by content type, each of a plurality of statistical models trained on text;
classify the text input as a content type; and
determine a subset of the plurality of statistical models which are classified the same content type as the content type of the text input.

### Claim 51 (depends on 50)

The non-transitory machine-readable medium of claim 50, wherein the selected statistical model is trained on one or more features extracted from the text associated with the content item, wherein execution of the program instructions by the one or more processors causes the one or more processors to:
extract the one or more features from the text input, and
query each of the statistical models of the subset of statistical models with each of the one or more features of the text input to generate a relevance estimate for each content item associated with each statistical model of the subset of statistical models in relation to the text input.

### Claim 52 (depends on 47)

The non-transitory machine-readable medium of claim 47, wherein execution of the program instructions by the one or more processors causes the one or more processors to:
compare a portion of the text input to a first language model trained on natural language text;
compare the portion of the text input to a second language model trained on conversational text; and
classify the portion of the text input as natural language or conversational.

### Claim 53 (depends on 52)

The non-transitory machine-readable medium of claim 52, wherein execution of the program instructions by the one or more processors causes the one or more processors to discard the portion of the text input if it is classified as conversational.
