# US 2021/0241157 A1 · Methods, systems and apparatus to improve multi-demographic modeling efficiency

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

- Number: US 2021/0241157 A1
- Document ID: US20210241157A1
- Status: Published application
- Kind: A1
- Application number: 17/171,670
- Filed: 2021-02-09
- Published: 2021-08-05
- Earliest claimed: 2016-11-23
- Assignees: The Nielsen Company (US), LLC
- Applicants: The Nielsen Company (US), LLC
- Inventors: Michael Sheppard; Ludo Daemen; Xiaoqi Cui
- Web page: https://patentgrep.com/patent/US20210241157A1

## CPC

- G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks
- G06Q 30/0201: Market modelling; Market analysis; Collecting market data
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- 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

Methods, apparatus, systems and articles of manufacture are disclosed to improve multi-demographic modeling efficiency. An example apparatus includes a feature set aggregator to segregate training data based on feature sets of interest, and to identify households that participate in at least one of the feature sets of interest, a class enumerator to reduce multi-demographic model iterations by enumerating demographic combinations for the identified households, the enumerated demographic combinations including a single identifier to represent a combination of two or more demographic categories, and a modeling engine to generate training coefficients associated with respective ones of the enumerated demographic combinations.

## Claims

### Claim 1

(canceled)

### Claim at position 2

-20. (canceled)

### Claim 21

An apparatus to reduce computational resources for a multi-demographic training model, comprising:
memory; and
at least one processor to execute computer readable instructions to at least:
reduce multi-demographic model iterations by enumerating demographic combinations for households that access at least one of a plurality of features represented in training data, the enumerated demographic combinations including an identifier, the identifier to represent a combination of two or more demographic categories and to include cartesian representations corresponding to occurrences or non-occurrences of a demographic category of interest, the demographic category of interest to be selected from the two or more demographic categories; and
generate probability values for ones of the enumerated demographic combinations based on training coefficients, the probability values to represent likelihoods that the ones of the enumerated demographic combinations accessed the at least one of the plurality of features, the training coefficients corresponding to the ones of the enumerated demographic combinations, the ones of the enumerated demographic combinations associated with (a) a training model and (b) a third-party model.

### Claim 22 (depends on 21)

The apparatus as defined in claim 21, wherein the at least one processor is to execute the computer readable instructions to identify ones of the households that access the at least one of the features of interest.

### Claim 23 (depends on 22)

The apparatus as defined in claim 22, wherein the at least one processor is to execute the computer readable instructions to associate demographic combinations with the ones of the households.

### Claim 24 (depends on 21)

The apparatus as defined in claim 21, wherein the at least one processor is to execute the computer readable instructions to generate the training coefficients based on a multinomial logistic regression with: (a) the training data, and (b) the ones of the enumerated demographic combinations.

### Claim 25 (depends on 21)

The apparatus as defined in claim 21, wherein the at least one processor is to execute the computer readable instructions to revert the probability values for respective ones of the enumerated demographic combinations to second probability values associated with individual demographic components.

### Claim 26 (depends on 21)

The apparatus as defined in claim 21, wherein the at least one processor is to execute the computer readable instructions to determine probability value differences between the training model and the third-party model.

### Claim 27 (depends on 26)

The apparatus as defined in claim 26, wherein the at least one processor is to execute the computer readable instructions to determine a trust metric for the third-party model based on a comparison of the probability value differences to a threshold, the trust metric to indicate the third-party model is reliable.

### Claim 28

A tangible computer-readable storage medium comprising instructions that, when executed, cause a processor to, at least:
reduce multi-demographic model iterations by enumerating demographic combinations for households that access at least one of a plurality of features represented in training data, the enumerated demographic combinations including an identifier, the identifier to represent a combination of two or more demographic categories and to include cartesian representations corresponding to occurrences or non-occurrences of a demographic category of interest, the demographic category of interest to be selected from the two or more demographic categories; and
generate probability values for ones of the enumerated demographic combinations based on training coefficients, the probability values to represent likelihoods that the ones of the enumerated demographic combinations accessed the at least one of the plurality of features, the training coefficients corresponding to the ones of the enumerated demographic combinations, the ones of the enumerated demographic combinations associated with (a) a training model and (b) a third-party model.

### Claim 29 (depends on 28)

The computer-readable medium as defined in claim 28, wherein the instructions are to cause the processor to identify ones of the households that access the at least one of the features of interest.

### Claim 30 (depends on 29)

The computer-readable medium as defined in claim 29, wherein the instructions are to cause the processor to associate demographic combinations with the ones of the households.

### Claim 31 (depends on 28)

The computer-readable medium as defined in claim 28, wherein the instructions are to cause the processor to generate the training coefficients based on a multinomial logistic regression with: (a) the training data, and (b) the ones of the enumerated demographic combinations.

### Claim 32 (depends on 28)

The computer-readable medium as defined in claim 28, wherein the instructions are to cause the processor to revert the probability values for respective ones of the enumerated demographic combinations to second probability values associated with individual demographic components.

### Claim 33 (depends on 28)

The computer-readable medium as defined in claim 28, wherein the instructions are to cause the processor to determine probability value differences between the training model and the third-party model.

### Claim 34 (depends on 33)

The computer-readable medium as defined in claim 33, wherein the instructions are to cause the processor to determine a trust metric for the third-party model based on a comparison of the probability value differences to a threshold, the trust metric to indicate the third-party model is reliable.

### Claim 35

An apparatus to reduce computational resources for a multi-demographic training model, comprising:
means for reducing multi-demographic model iterations by enumerating demographic combinations for households that access at least one of a plurality of features represented in training data, the enumerated demographic combinations including an identifier, the identifier to represent a combination of two or more demographic categories and to include cartesian representations corresponding to occurrences or non-occurrences of a demographic category of interest, the demographic category of interest to be selected from the two or more demographic categories; and
means for modeling to generate probability values for ones of the enumerated demographic combinations based on training coefficients, the probability values to represent likelihoods that the ones of the enumerated demographic combinations accessed the at least one of the plurality of features, the training coefficients corresponding to the ones of the enumerated demographic combinations, the ones of the enumerated demographic combinations associated with (a) a training model and (b) a third-party model.

### Claim 36 (depends on 35)

The apparatus as defined in claim 35, further including means for aggregating to identify ones of the households that access the at least one of the features of interest.

### Claim 37 (depends on 36)

The apparatus as defined in claim 36, wherein the modeling means is to generate the training coefficients based on a multinomial logistic regression with: (a) the training data, and (b) the ones of the enumerated demographic combinations.

### Claim 38 (depends on 35)

The apparatus as defined in claim 35, further including means for reverting the probability values for respective ones of the enumerated demographic combinations to second probability values associated with individual demographic components.

### Claim 39 (depends on 38)

The apparatus as defined in claim 38, wherein the reverting means is to determine probability value differences between the training model and the third-party model.

### Claim 40 (depends on 39)

The apparatus as defined in claim 39, wherein the reverting means is to determine a trust metric for the third-party model based on a comparison of the probability value differences to a threshold, the trust metric to indicate the third-party model is reliable.
