# US 2025/0168431 A1 · Machine Learning Systems and Methods for Predicting End-User Consumption of Future Multimedia Transmissions

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

- Number: US 2025/0168431 A1
- Document ID: US20250168431A1
- Status: Published application
- Kind: A1
- Application number: 19/033,038
- Filed: 2025-01-21
- Published: 2025-05-22
- Earliest claimed: 2022-09-27
- Assignees: Gracenote, Inc.
- Applicants: Gracenote, Inc.
- Inventors: Scott John Sereday; Cathy Zhuang; Shufang Ban; Oana Monica Dan
- Web page: https://patentgrep.com/patent/US20250168431A1

## CPC

- H04N 21/251: {Learning process for intelligent management, e.g. learning user preferences for recommending movies (details of learning user preferences for the retrieval of video data in a video database G06F16/739; computer systems using learning methods G06N3/08)}
- H04N 21/252: {Processing of multiple end-users' preferences to derive collaborative data}
- H04N 21/25883: {being end-user demographical data, e.g. age, family status or address (arrangements for identifying locations of users in broadcast systems H04H60/52)}
- H04N 21/25891: {being end-user preferences (retrieval of video data in a video database based on user preferences G06F16/739; arrangements for recognizing users' preferences H04H60/46; user profiles in network data switching protocols H04L67/306; processing of user preferences or user profiles in wireless networks H04W8/18)}
- H04N 21/44204: {Monitoring of content usage, e.g. the number of times a movie has been viewed, copied or the amount which has been watched (monitoring of user activities for profile generation for accessing a video database G06F16/739; protecting generic digital content where the protection is independent of the precise nature of the content G06F21/10; arrangements for monitoring the use made of the broadcast services in broadcast systems H04H60/31)}
- H04N 21/812: {involving advertisement data (advertising per se G06Q30/02)}
- H04N: PICTORIAL COMMUNICATION, e.g. TELEVISION

## Abstract

Methods and systems for prediction audience ratings are disclosed. A database of television (TV) viewing data may include program records for a multiplicity of existing TV programs. A system may receive a training plurality of program records from the TV viewing data, and for each program record a most similar TV program based on content characteristics may be identified. A synthetic program record may be constructed by merging features of each record and its most similar record. Audience performance metrics may be omitted from synthetic records. An aggregate of the training plurality of program records and the synthetic program records may be used to train a machine-learning (ML) model to predict audience performance metrics of the new or hypothetical TV programs not yet available for viewing and/or not yet transmitted or streamed.

## Claims

### Claim 1

A system comprising:
a database comprising program records each identifying a television (TV) program and including historical presentation-logistics (PL) features, content-descriptor (CD) features, and historical viewer-rating (VR) metrics, wherein the historical PL features comprise information identifying a content-delivery platform that previously sourced the TV program for end-user viewing and specifying a delivery mode used to deliver the TV program, and a release-schedule drop pattern (RSDP) that was used by the content-delivery platform for viewing availability and/or delivery, wherein the CD features characterize the TV program, and wherein the historical VR metrics comprise, for the historical PL features, statistical quantification of viewing performance of the TV program among one or more audience categories;
one or more processors; and
memory storing instructions that, when executed by the one or more processors, cause the system to carry out operations including:
for each of the program records, identifying from among the program records a most similar TV program based on a quantitative comparison of the CD features of the program record with those of the other program records, wherein the most similar TV program is different from the program of the program record;
for each of the program records, creating a synthetic program record comprising the historical PL features from the program record and the CD features of the most similar TV program;
by applying the program records and the synthetic program records as input and the historical VR metrics of the program records as ground-truths, training a machine-learning (ML) model to predict audience performance metrics of the TV programs of the program records; and
configuring the ML model for predicting audience performance metrics of one or more runtime program records respectively associated with hypothetical TV programs not yet available for viewing and/or not yet transmitted.

### Claim 2 (depends on 1)

The system of claim 1, wherein the operations further include:
receiving a runtime simulated program record comprising an identifier of a planned and/or proposed new TV program, the CD features from the program records, and planned and/or proposed PL features;
applying the ML model to the runtime simulated program record to predict VR metrics for the new TV program; and
providing the predicted VR metrics for the new TV program as output to a user interface.

### Claim 3 (depends on 2)

The system of claim 2, wherein the operations further include, prior to receiving the runtime simulated program record:
at the user interface, receiving user input comprising data corresponding to: (i) the identifier of the new TV program, (ii) an identifier of an existing TV program in the database, and (iii) prospective PL data comprising a planned and/or proposed content-delivery platform and a planned and/or proposed delivery mode and RSDP for projected use by the planned and/or proposed content-delivery platform for the TV program;
identifying a runtime most similar TV program to the existing TV program based on a comparison of CD features of the existing TV program with those of the other TV programs in the database;
creating the runtime simulated program record comprising the planned and/or proposed PL features derived from the prospective PL data; and
aggregating the CD features from the runtime most similar TV program into the runtime simulated program record.

### Claim 4 (depends on 3)

The system of claim 3, wherein the historical PL features further comprise historically-applied ratings-prediction conditioning data, including viewer demographics and viewing timeline windows within which historical VR metrics were determined,
and wherein the prospective PL data further comprise prospective ratings-prediction conditioning data, including prospective viewer demographics and viewing timeline windows within which VR metrics are to be predicted by the trained ML model.

### Claim 5 (depends on 2)

The system of claim 2, wherein the operations further include: prior to training the ML model, determining a priority subset of total features of all available features of each program record that are priority features for use in model predictions,
wherein training the ML model further comprises using the priority subset to cause training of the ML model to use only the priority features of the program records and the synthetic program records,
and wherein applying the trained ML model to the runtime simulated program record comprises using the priority subset to cause the trained ML model to be applied only to the priority features of the simulated program record.

### Claim 6 (depends on 3)

The system of claim 3, wherein the delivery mode of each TV program of the TV programs is at least one of: linear, wherein availability for viewing corresponds to broadcast transmission of the any TV program, or streaming, wherein availability for viewing corresponds to available for streaming upon viewer request,
and wherein the RSDP for the TV program specifies at least one of: a drop date when the TV program became or will become available for viewing, a number of episodes of the TV program that became or will become available for viewing at the drop date, or a schedule of availability of episodes of the TV program beginning at the drop date.

### Claim 7 (depends on 1)

The system of claim 1, wherein the operations further include:
receiving a runtime modified program record comprising an existing program record having historical PL features replaced with runtime PL features provided as user input via a user interface;
applying the ML model to the runtime modified program record to predict VR metrics for the TV program identified in the existing program record; and
providing the predicted VR metrics for the TV program identified in the existing program record as output to a user interface.

### Claim 8 (depends on 1)

The system of claim 1, wherein the ML model comprises a natural gradient boost (NG-boost) algorithm, wherein training the ML model comprises generating model objects,
and wherein configuring the trained ML model for predicting audience performance metrics of the one or more runtime program records comprises providing the model objects for application to the one or more runtime program records.

### Claim 9 (depends on 1)

The system of claim 1, wherein the operations further include aggregating the program records and the synthetic program records into an input data table, wherein each row comprises program records, and each column corresponds to a feature of the program records, and wherein the synthetic records are appended as rows at the end of the input data table.

### Claim 10 (depends on 1)

The system of claim 1, wherein the statistical quantification of viewing performance of the TV program among one or more audience categories comprises total number of minutes of the TV program watched, aggregated for each of one or more viewer demographics.

### Claim 11

A method carried out by a computing system having access to a database comprising program records each identifying a television (TV) program and including historical presentation-logistics (PL) features, content-descriptor (CD) features, and historical viewer-rating (VR) metrics, wherein the historical PL features comprise information identifying a content-delivery platform that previously sourced the TV program for end-user viewing and specifying a delivery mode used to deliver the TV program, and a release-schedule drop pattern (RSDP) that was used by the content-delivery platform for viewing availability and/or delivery, wherein the CD features characterize the TV program, and wherein the historical VR metrics comprise, for the historical PL features, statistical quantification of viewing performance of the TV program among one or more audience categories,
wherein the method comprises:
for each of the program records, identifying from among the program records a most similar TV program based on a quantitative comparison of the CD features of the program record with those of the other program records, wherein the most similar TV program is different from the program of the program record;
for each of the program records, creating a synthetic program record comprising the historical PL features from the program record and the CD features of the most similar TV program;
by applying the program records and the synthetic program records as input and the historical VR metrics of the program records as ground-truths, training a machine-learning (ML) model to predict audience performance metrics of the TV programs of the program records; and
configuring the ML model for predicting audience performance metrics of one or more runtime program records respectively associated with hypothetical TV programs not yet available for viewing and/or not yet transmitted.

### Claim 12 (depends on 11)

The method of claim 11, further comprising:
receiving a runtime simulated program record comprising an identifier of a planned and/or proposed new TV program, the CD features from the program records, and planned and/or proposed PL features;
applying the ML model to the runtime simulated program record to predict VR metrics for the new TV program; and
providing the predicted VR metrics for the new TV program as output to a user interface.

### Claim 13 (depends on 12)

The method of claim 12, further comprising, prior to receiving a runtime simulated program record:
at the user interface, receiving user input comprising data corresponding to: (i) the identifier of the new TV program, (ii) an identifier of an existing TV program in the database, and (iii) prospective PL data comprising a planned and/or proposed content-delivery platform and a planned and/or proposed delivery mode and RSDP for projected use by the planned and/or proposed content-delivery platform for the TV program;
identifying a runtime most similar TV program to the existing TV program based on a comparison of CD features of the existing TV program with those of the other TV programs in the database;
creating the runtime simulated program record comprising the planned and/or proposed PL features derived from the prospective PL data; and
aggregating the CD features from the runtime most similar TV program into the runtime simulated program record.

### Claim 14 (depends on 13)

The method of claim 13, wherein the historical PL features further comprise historically-applied ratings-prediction conditioning data, including viewer demographics and viewing timeline windows within which historical VR metrics were determined,
and wherein the prospective PL data further comprise prospective ratings-prediction conditioning data, including prospective viewer demographics and viewing timeline windows within which VR metrics are to be predicted by the trained ML model.

### Claim 15 (depends on 13)

The method of claim 13, wherein the delivery mode of each TV program of the TV programs is at least one of: linear, wherein availability for viewing corresponds to broadcast transmission of the any TV program, or streaming, wherein availability for viewing corresponds to available for streaming upon viewer request,
and wherein the RSDP for the TV program specifies at least one of: a drop date when the TV program became or will become available for viewing, a number of episodes of the TV program that became or will become available for viewing at the drop date, or a schedule of availability of episodes of the TV program beginning at the drop date.

### Claim 16 (depends on 11)

The method of claim 11, further comprising:
receiving a runtime modified program record comprising an existing program record having historical PL features replaced with runtime PL features provided as user input via a user interface;
applying the ML model to the runtime modified program record to predict VR metrics for the TV program identified in the existing program record; and
providing the predicted VR metrics for the TV program identified in the existing program record as output to a user interface.

### Claim 17 (depends on 11)

The method of claim 11, wherein the ML model comprises a natural gradient boost (NG-boost) algorithm, wherein training the ML model comprises generating model objects,
and wherein configuring the trained ML model for predicting audience performance metrics of the one or more runtime program records comprises providing the model objects for application to the one or more runtime program records.

### Claim 18 (depends on 11)

The method of claim 11, further comprising aggregating the program records and the synthetic program records into an input data table, wherein each row comprises program records, and each column corresponds to a feature of the program records, and wherein the synthetic records are appended as rows at the end of the input data table.

### Claim 19 (depends on 11)

The method of claim 11, wherein the statistical quantification of viewing performance of the TV program among one or more audience categories comprises total number of minutes of the TV program watched, aggregated for each of one or more viewer demographics.

### Claim 20

A non-transitory computer-readable medium having instructions stored thereon that, when carried out by one or more processors of a computing system comprising a database comprising program records each identifying a television (TV) program and including historical presentation-logistics (PL) features, content-descriptor (CD) features, and historical viewer-rating (VR) metrics, wherein the historical PL features comprise information identifying a content-delivery platform that previously sourced the TV program for end-user viewing and specifying a delivery mode used to deliver the TV program, and a release-schedule drop pattern (RSDP) that was used by the content-delivery platform for viewing availability and/or delivery, wherein the CD features characterize the TV program, and wherein the historical VR metrics comprise, for the historical PL features, statistical quantification of viewing performance of the TV program among one or more audience categories, cause the computing system to carry out operations including:
for each of the program records, identifying from among the program records a most similar TV program based on a quantitative comparison of the CD features of the program record with those of the other program records, wherein the most similar TV program is different from the program of the program record;
for each of the program records, creating a synthetic program record comprising the historical PL features from the program record and the CD features of the most similar TV program;
by applying the program records and the synthetic program records as input and the historical VR metrics of the program records as ground-truths, training a machine-learning (ML) model to predict audience performance metrics of the TV programs of the program records; and
configuring the ML model for predicting audience performance metrics of one or more runtime program records respectively associated with hypothetical TV programs not yet available for viewing and/or not yet transmitted.
