# US 11,113,707 B1 · Artificial intelligence identification of high-value audiences for marketing campaigns

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

- Number: US 11,113,707 B1
- Document ID: US11113707B1
- Status: Granted
- Kind: B1
- Application number: 17/156,000
- Filed: 2021-01-22
- Granted: 2021-09-07
- Earliest claimed: 2021-01-22
- Assignees: Isolation Network, Inc.
- Applicants: Isolation Network, Inc.
- Inventors: Shantanu K Sharma; Daniel Cownden; James Parks; Jia Wen Tian; Keivan Monfared; Yu-Chien Huang
- Web page: https://patentgrep.com/patent/US11113707B1

## CPC

- G06Q 30/0205: {based on location or geographical consideration}
- G06N 5/04: Inference or reasoning models
- G06N 20/00: Machine learning
- G06Q 30/0206: {Price or cost determination based on market factors}
- 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
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A marketing analytics pipeline that receives transaction-level trend reporting of media IP assets distributed on digital service providers (DSPs), the marketing analytics pipeline including a marketing action analytics hub configured to receive the growth and re-engagement opportunities from an opportunity detection analytics hub or from a marketing platform directly and to identify high growth potential audiences using predictive models of engagement; and a marketing platform configured to interact with the marketing action analytics hub and to create marketing campaigns based around the high growth potential audiences and suggest marketing actions to media IP asset managers and owners to deliver to the high growth potential audiences on appropriate marketing channels via targeted marketing campaigns. The marketing action analytics hub uses a geodemographic and interest-based targeting method of generating descriptions of high growth potential audiences.

## Claims

### Claim 1

A method for authorizing execution of targeted marketing campaigns for streaming media intellectual property (IP) assets, the method comprising:
streaming, by media asset consumers, of media IP assets distributed to digital service providers (DSPs) via a media asset distributor, for which the consumers have obtained streaming access through subscription or agreement to other terms of use from the DSPs;
documenting and compiling the streaming interactions of the consumer into a transaction-level trend data report, wherein items in the report comprise specific media IP assets;
implementing a marketing analytics pipeline that:
receives the transaction-level trend data report,
compiles the streaming interactions of individual consumers from the transaction-level trend data report into specific geodemographic groups including geographic regions, genders, ages and DSP subscription status, and without identification of individual consumers,
creates a ranked list of media IP assets and corresponding marketing actions,
retrieves links on the DSPs for the media IP assets in the ranked list of media IP assets, the links being retrieved by use of APIs of the DSP and/or a smartlink API,
creates for each media IP asset in the ranked list of media IP assets, using the smartlink API, one or more landing pages comprising the associated links on the DSPs, and
creates marketing campaigns for media IP asset managers/owners including the one or more landing pages comprising the links on the DSPs for each of the one or more media IP assets in the ranked list;
presenting the ranked list of media IP assets and corresponding marketing actions and marketing campaigns to the media IP asset managers/owners in a graphical user interface (GUI), wherein the GUI includes a media IP asset and a corresponding listing of marketing campaigns, and wherein the GUI enables the media IP asset managers/owners to manage funds, set territory specific budgets and store promotional advertisement material for the marketing campaigns;
receiving modifications made via the GUI by the media IP asset managers/owners; and
receiving authorization to execute the marketing campaigns from the media IP asset managers/owners,
wherein the marketing analytics pipeline comprises:
a marketing action analytics hub configured to receive the growth and re-engagement opportunities from an opportunity detection analytics hub or from a marketing platform directly and to identify high growth potential audiences using predictive models of engagement; and
a marketing platform configured to interact with the marketing action analytics hub and to create targeted marketing campaigns based around the high growth potential audiences, the marketing platform further being configured to suggest specific marketing actions to the media IP asset managers/owners, wherein the marketing actions are designed for delivery to the high growth potential audiences on appropriate marketing channels via the targeted marketing campaigns;
wherein the marketing action analytics hub is comprised of an offline model generation component and an online audience generation component and uses a targeting method based on the specific geodemographic group data compiled from the streaming interactions of individual consumers and the interests of these groups for generating descriptions of high growth potential audiences,
wherein the offline model generation component is configured to use the compiled streaming interactions of geodemographic groups contained in a consumption database to fit statistical models based on consumption patterns of an entirety of distributed media IP assets on a regular cadence by clustering media IP assets into categories and employing an alternating least squares framework to generate a specific matrix factorization for each category by approximating an engagement matrix E as a product of two latent factor matrices Eest:=XYτ, where Yτ is the matrix transpose of Y;
wherein the choice of the rows and columns of E are informed by the truncation of media IP assets and geodemographic groups based on consumption thresholds, and the values of the entries of E are evaluations of a scalar valued multivariable function ƒ on vectors in a consumption tensor whose dimensions correspond to media IP assets, geodemographic groups, and streaming channels, X is an m×k matrix, Y is an n×k matrix and Eest is itself an approximation of the best rank k≤min(m,n) approximation of E that minimizes a cost function
cost
⁢
⁢
(
X
,
Y
)
:=
∑
(
i
,
j
)
∈
S
train
⁢
(
E
ij
-
X
i
⁢
Y
j
⊤
)
2
+
λ
(
∑
i
∈
I
⁢

X
i

2
+
∑
j
∈
J
⁢

Y
j

2
)
over all possible choices of X and Y such that Strain and Stest form a partition (disjoint nonempty subsets that cover the whole set) of indices of E, Xi and Yj respectively denote the i-th and j-th rows of the matrices X and Y and I and J are truncated sets of media IP assets and geodemographic groups respectively;
wherein λ and k are two hyperparameters for each category determined by a cross validation scheme that minimizes the sum of squared errors SSE(k, λ):=Σ(i,j)∈Stest(Eij(k,λ)−Eij)2 using an adaptive grid search where Eij(k,λ) is the (i, j)-entry in the matrix Eest(k,λ) denoting the estimate obtained from the alternating least squares component for a given hyperparameter pair (k, λ), where the chosen k and λ are then used to train the model on the combined train and test dataset, i.e. Strain is now I×J, to minimize the cost(X,Y) to fit Eest=XYτ;
wherein ∥Xi∥ and ∥Yj∥ are the l2 norms of Xi and Yj respectively, where the analytic expressions that minimize the cost function for Xi and Yj for a fixed Y and a fixed X respectively are approximated by iteratively applying Xi=(ED(i)Y((YτY+λ·Id)−1)τ)i and Yj=(EτD(j)X((XτX+λ·Id)−1)τ)j, where Id is the appropriately sized identity matrix, D(i) is an n×n diagonal matrix with the j-th diagonal element one if (i, j)∈Strain and zero otherwise, and similarly for the m×m matrix D(j).

### Claim 2 (depends on 1)

The method according to claim 1, wherein the online audience generation component is comprised of a high growth potential audience identification component configured to receive both: (1) the compiled streaming interactions of geodemographic groups contained in the consumption database, the media IP assets and the marketing campaign goals from the opportunity detection analytics hub or directly from the media IP asset managers/owners through the marketing platform, and (2) the trained statistical models from an online model selection component or an online model update component to produce a subset of high growth potential geodemographic groups chosen according to a goal-weighted predicted engagement score calculated by assessing both the engagement predicted by the statistical model in each geodemographic group as well as the alignment with the received marketing campaign goal for each geodemographic group where alignment is determined using both the model prediction of engagement and the observed engagement of each geodemographic group, for the purpose of ensuring consistency between the targeting priorities as indicated by the campaign goal and as represented by the chosen high growth potential geodemographic groups.

### Claim 3 (depends on 2)

The method according to claim 2, wherein the high growth potential audience identification component is configured to receive both: (1) the compiled streaming interactions of geodemographic groups contained in the consumption database, the media IP assets and the marketing campaign goals from the opportunity detection analytics hub or directly from the media IP asset managers/owners through the marketing platform, and (2) the trained statistical models containing media IP asset category dependent matrices E and Ê and a set of geodemographic groups J from the online model selection component or the online model update component, to produce a subset of high growth potential geodemographic groups J*(i0)⊆J, where geodemographic groups j∈J*(i0) for a media IP asset i0 are chosen according to the goal-weighted predicted engagement score, wherein the goal-weighted predicted engagement score Êi0j is calculated by applying marketing campaign goal weights to three broad classes of geodemographic groups defined by {j∈J|Δ(i0, j)»μi0+σi0} associated with marketing campaign goals of audience expansion, {j∈J|Δ(i0, j)«μi0−σi0} associated with marketing campaign goals of growth trend acceleration and {j∈J|μi0−σi0«Δ(i0, j)«μi0+σi0} associated with marketing campaign goals of sustaining audience engagement, where Δ(i0, j):=Êi0j−Ei0j, and μi0 and σi0 are the mean and the standard deviation of {Δ(i0, j)|j∈J} respectively.

### Claim 4 (depends on 3)

The method according to claim 3, wherein in the online audience generation component, the online model update component is configured to include a media IP asset i0 that is not present in its respective category dependent set of media IP assets I selected in the online model selection component by appending a row Ei0 to the matrix E obtained from applying the function ƒ on the tensor vectors corresponding to the consumption of the media IP assets i0 in the geodemographic groups J and then iterating the alternating least squares framework where the initialization of the optimization step is informed by the matrices X and Y of the category dependent offline trained model to generate an adapted trained online statistical model Ê.

### Claim 5 (depends on 3)

The method according to claim 3, wherein the online audience generation component is further comprised of a metadata interest identification component configured to send a collection of metadata that most strongly inform the observed predicted errors of a set J*(i0) for a media IP asset i0 as additional territory specific advertising interests or keywords in the initial advertising criteria on appropriate third-party advertising platforms by collecting the metadata of the media IP assets that provide the largest contribution to the sum of the inner product of the i0-th row of Ω and the j-th column of E where Ω=X(XτX+λ·Id)−1Xτ·Id.

### Claim 6 (depends on 3)

The method according to claim 3, wherein the online audience generation component is further comprised of a marketing campaign budget optimization component that creates a third-party advertising platform description with optimal estimated reach by continually interacting with third-party advertising platforms through APIs and adding additional geodemographic groups j prioritized according to Êi0j* to the DSP specific targeting descriptions and then converting these DSP specific descriptions to an appropriate third-party advertising platform description until either the optimal estimated reach is obtained for each territory and budget as determined by the campaign budget optimization model or until there are no longer any high growth potential geodemographic groups available, in which case the budget is reduced according to maximum territorial budget limits based on estimated third-party advertising platform reach determined by a historical campaign analytics database trained offline.

### Claim 7 (depends on 1)

The method according to claim 1, wherein the marketing analytics pipeline receives multiple transaction-level trend data reports corresponding respectively to multiple DSPs.

### Claim 8 (depends on 7)

The method according to claim 7, wherein the multiple transaction-level trend data reports are received at a regular cadence.