# US 12,602,414 B2 · System and for to determine diverse and relevant passages of documents

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

- Number: US 12,602,414 B2
- Document ID: US12602414B2
- Status: Granted
- Kind: B2
- Application number: 18/191,717
- Filed: 2023-03-28
- Published: 2023-10-12
- Granted: 2026-04-14
- Earliest claimed: 2022-03-29
- Assignees: Open Text Corporation
- Applicants: Open Text Inc.
- Inventors: Jeremy Garner Pickens
- Web page: https://patentgrep.com/patent/US12602414B2

## CPC

- G06F 16/3334: {Selection or weighting of terms from queries, including natural language queries}
- G06F 16/24578: {using ranking}
- G06F 16/34: Browsing; Visualisation therefor (browsing or visualisation for clustering or classification G06F16/358)
- G06F 16/435: Filtering based on additional data, e.g. user or group profiles
- G06F 40/205: Parsing
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

Systems and methods for selecting multiple, diverse passages of a document based on such a relevancy model are disclosed.

## Claims

### Claim 1

A system for content relevance search comprising:
a processor;
a non-transitory computer readable medium, comprising instructions for:
receiving content and a relevance model related to the content, the relevance model comprising:
a plurality of features; and
a relevance value weight for each feature;
processing the content to generate a plurality of search results based on the plurality of features in the relevance model, each of the search results comprising one or more of the features in the relevance model;
generating an initial score for each of the search results using the relevance model, the initial score for each of the search results being calculated by:
determining a sum of the relevance value weights of the one or more features comprised in the search result;
multiplying the sum by a number of unique features within the search result, thereby generating a multiplied sum; and
dividing the multiplied sum by a positional spread value, the positional spread value determined based on a distance between a first position of a first feature in the search result and a second position of a second feature in the search result, wherein the number of unique features comprise the first and second features;
based on the generation of the initial score for each search result, selecting an initial one of the search results;
modifying the relevance value weight of each feature comprised in the initial one of the search results, by reducing the relevance value weight of each feature comprised in the initial one of the search results;
generating a subsequent score for each search result other than the initial one of the search results, wherein the subsequent score for each search result of the remaining search results is generated using the relevance model and based on:
the one or more features comprised in each remaining search result;
the number of unique features within each remaining search result;
the positional spread of features within each remaining search result; and
the relevance value weights of the one of more features comprised in each remaining search result, the relevance value weights being sourced from the relevance model, wherein the relevance model incorporates the relevance value weight of each feature that was comprised in the initial one of the search results; and
based on the generation of the subsequent score for each remaining search result, selecting a subsequent search result from all of the remaining search results, the subsequent search result and the initial search result characterized by diverse and relevant ones of the features.

### Claim 2 (depends on 1)

The system of claim 1, wherein the content comprises a text document, an image file, an audio file, or an audio-visual file, wherein:
when the content comprises a text document, the plurality of search results comprises passages of text and the features of the relevance model comprise text terms; and
when the content comprises an image file, the plurality of search results comprises a grouping of pixels of the image file and the features of the relevance model comprise image samples;
when the content comprises an audio file, the plurality of search results comprises audio snippets and the features of the relevance model comprise audio samples; and
when the content comprises an audio-visual file, the plurality of search results comprises audio-visual snippets and the features of the relevance model comprise audio-visual samples.

### Claim 3 (depends on 2)

The system of claim 2, wherein when the content comprises a text document, the initial score for each of the search results is further based on the positional spread value of each passage of text computed based on the difference between a position of the last character in each passage relative to a reference text position in the text document and a position of the first character in each passage relative to the reference text in the text document.

### Claim 4 (depends on 1)

The system of claim 1, wherein content comprises a text document and the plurality of search results comprises passages of text and the features of the relevance model comprise text terms, and the search results are computed based on parsing the text document to find a number of the text terms within the text document.

### Claim 5 (depends on 4)

The system of claim 4, wherein the text terms in the initial search result and the text terms in the subsequent search result are different thereby producing diverse relevant text terms between the initial search result and subsequent search result.

### Claim 6 (depends on 1)

The system of claim 1, wherein modifying the relevance value weight of each of the features comprised in the initial search result comprises setting the relevance value weight of each of the features comprised in the initial search result to zero.

### Claim 7 (depends on 1)

The system of claim 1, further comprising:
iterating the processing of the search results comprising:
generating an iterative score for each of the search results based on the one or more features of each of the search results and the relevance value weight of the one of more features of each of the search results;
based on generating of the iterative score for each of the search results, selecting a next iterative search result;
modifying the relevance value weight of each of the features comprised in the next iterative search result, the modifying of the relevance value weight reducing the relevance of each respective feature comprised in the next iterative search result;
repeating, a predefined number of iterations, the iterating the processing of the plurality of search results.

### Claim 8

A method for content relevance search, comprising:
receiving content and a relevance model related to the content, the relevance model comprising:
a plurality of features; and
a relevance value weight for each feature;
processing the content to generate a plurality of search results based on the plurality of features in the relevance model, each of the search results comprising one or more of the features in the relevance model;
generating an initial score for each of the search results using the relevance model, the initial score for each of the search results being calculated by:
determining a sum of the relevance value weights of the one or more features comprised in the search result;
multiplying the sum by a number of unique features within the search result, thereby generating a multiplied sum; and
dividing the multiplied sum by a positional spread value, the positional spread value determined based on a distance between a first position of a first feature in the search result and a second position of a second feature in the search result, wherein the number of unique features comprise the first and second features;
based on the generation of the initial score for each search result, selecting an initial one of the search results;
modifying the relevance value weight of each feature comprised in the initial one of the search results, by the relevance value weight of each feature comprised in the initial one of the search results;
generating a subsequent score for each search result other than the initial one of the search results, wherein the subsequent score for each search result of the remaining search results is generated using the relevance mode and is based on:
the one or more features comprised in each remaining search result;
the number of unique features within each remaining search result;
the positional spread of features within each remaining search result; and
the relevance value weights of the one of more features comprised in each remaining search result, the relevance value weights being sourced from the relevance model, wherein the relevance model incorporates the relevance value weight of each feature that was comprised in the initial one of the search results; and
based on the generation of the subsequent score for each remaining search result, selecting a subsequent search result from all of the remaining search results, the subsequent search result and the initial search result characterized by diverse and relevant ones of the features.

### Claim 9 (depends on 8)

The method of claim 8, wherein the content comprises a text document, an image file, an audio file, or an audio-visual file, wherein:
when the content comprises a text document, the plurality of search results comprises passages of text and the features of the relevance model comprise text terms; and
when the content comprises an image file, the plurality of search results comprises a grouping of pixels of the image file and the features of the relevance model comprise image samples;
when the content comprises an audio file, the plurality of search results comprises audio snippets and the features of the relevance model comprise audio samples; and
when the content comprises an audio-visual file, the plurality of search results comprises audio-visual snippets and the features of the relevance model comprise audio-visual samples.

### Claim 10 (depends on 9)

The method of claim 9, wherein when the content comprises a text document, the initial score for each of the search results is further based on the positional spread value of each passage of text computed based on the difference between a position of the last character in each passage relative to a reference text position in the text document and a position of the first character in each passage relative to the reference text in the text document.

### Claim 11 (depends on 8)

The method of claim 8, wherein content comprises a text document and the plurality of search results comprises passages of text and the features of the relevance model comprise text terms, and the search results are computed based on parsing the text document to find a number of the text terms within the text document.

### Claim 12 (depends on 11)

The method of claim 11, wherein the text terms in the initial search result and the text terms in the subsequent search result are different thereby producing diverse relevant text terms between the initial search result and subsequent search result.

### Claim 13 (depends on 8)

The method of claim 8, wherein modifying the relevance value weight of each of the features comprised in the initial search result comprises setting the relevance value weight of each of the features comprised in the initial search result to zero.

### Claim 14 (depends on 8)

The method of claim 8, further comprising:
iterating the processing of the search results comprising:
generating an iterative score for each of the search results based on the one or more features of each of the search results and the relevance value weight of the one of more features of each of the search results;
based on generating of the iterative score for each of the search results, selecting a next iterative search result;
modifying the relevance value weight of each of the features comprised in the next iterative search result, the modifying of the relevance value weight reducing the relevance of each respective feature comprised in the next iterative search result;
repeating, a predefined number of iterations, the iterating the processing of the plurality of search results.

### Claim 15

A non-transitory computer readable medium, comprising:
receiving content and a relevance model related to the content, the relevance model comprising:
a plurality of features; and
a relevance value weight for each feature;
processing the content to generate a plurality of search results based on the plurality of features in the relevance model, each of the search results comprising one or more of the features in the relevance model;
generating an initial score for each of the search results using the relevance model, the initial score for each of the search results being calculated by:
determining a sum of the relevance value weights of the one or more features comprised in the search result;
multiplying the sum by a number of unique features within the search result, thereby generating a multiplied sum; and
dividing the multiplied sum by a positional spread value, the positional spread value determined based on a distance between a first position of a first feature in the search result and a second position of a second feature in the search result, wherein the number of unique features comprise the first and second features;
based on the generation of the initial score for each search result, selecting an initial one of the search results;
modifying the relevance value weight of each feature comprised in the initial one of the search results, by reducing the relevance value weight of each feature comprised in the initial one of the search results;
generating a subsequent score for each search result other than the initial one of the search results, wherein the subsequent score for each search result of the remaining search results is generated using the relevance mode and is based on:
the one or more features comprised in each remaining search result;
the number of unique features within each remaining search result;
the positional spread of features within each remaining search result; and
the relevance value weights of the one of more features comprised in each remaining search result, the relevance value weights being sourced from the relevance model, wherein the relevance model incorporates the relevance value weight of each feature that was comprised in the initial one of the search results; and
based on the generation of the subsequent score for each remaining search result, selecting a subsequent search result from all of the remaining search results, the subsequent search result and the initial search result characterized by diverse and relevant ones of the features.

### Claim 16 (depends on 15)

The non-transitory computer readable medium of claim 15, wherein the content comprises a text document, an image file, an audio file, or an audio-visual file, wherein:
when the content comprises a text document, the plurality of search results comprises passages of text and the features of the relevance model comprise text terms; and
when the content comprises an image file, the plurality of search results comprises a grouping of pixels of the image file and the features of the relevance model comprise image samples;
when the content comprises an audio file, the plurality of search results comprises audio snippets and the features of the relevance model comprise audio samples; and
when the content comprises an audio-visual file, the plurality of search results comprises audio-visual snippets and the features of the relevance model comprise audio-visual samples.

### Claim 17 (depends on 16)

The non-transitory computer readable medium of claim 16, wherein when the content comprises a text document, the initial score for each of the search results is further based on the positional spread value of each passage of text computed based on the difference between a position of the last character in each passage relative to a reference text position in the text document and a position of the first character in each passage relative to the reference text in the text document.

### Claim 18 (depends on 15)

The non-transitory computer readable medium of claim 15, wherein content comprises a text document and the plurality of search results comprises passages of text and the features of the relevance model comprise text terms, and the search results are computed based on parsing the text document to find a number of the text terms within the text document.

### Claim 19 (depends on 18)

The non-transitory computer readable medium of claim 18, wherein the text terms in the initial search result and the text terms in the subsequent search result are different thereby producing diverse relevant text terms between the initial search result and subsequent search result.

### Claim 20 (depends on 15)

The non-transitory computer readable medium of claim 15, wherein modifying the relevance value weight of each of the features comprised in the initial search result comprises setting the relevance value weight of each of the features comprised in the initial search result to zero.

### Claim 21 (depends on 15)

The non-transitory computer readable medium of claim 15, further comprising:
iterating the processing of the search results comprising:
generating an iterative score for each of the search results based on the one or more features of each of the search results and the relevance value weight of the one of more features of each of the search results;
based on generating of the iterative score for each of the search results, selecting a next iterative search result;
modifying the relevance value weight of each of the features comprised in the next iterative search result, the modifying of the relevance value weight reducing the relevance of each respective feature comprised in the next iterative search result;
repeating, a predefined number of iterations, the iterating the processing of the plurality of search results.