Methods for viewer modeling and bidding in an online advertising campaign

Patent metadata

Field	Value
Publication number	US-11010794-B2
Application number	US-201414295811-A
Country	US
Kind code	B2
Filing date	Jun 4, 2014
Priority date	Mar 13, 2013
Publication date	May 18, 2021
Grant date	May 18, 2021

Abstract

Systems and methods are disclosed for employing supervised machine learning methods with activities and attributes of viewers with truth as input, to produce models that are utilized in determining probabilities that an unknown viewer belongs to one or more demographic segment categories. Using these models for processing viewer behavior, over a period of time a database of known categorized viewers is established, each categorized viewer having a probability of belonging to one or more segment categories. These probabilities are then used in bidding for online advertisements in response to impression opportunities offered in online media auctions. The probabilities are also used in predicting on-target impressions and GRPs (Gross Rating Points) in advance of online advertising media campaigns, and pricing those campaigns to advertiser/clients. Strategies are also disclosed for fulfilling a campaign when an available inventory of known categorized viewers is not adequate to fulfill a campaign in a required runtime.

First claim

What is claimed is:

1. A computerized method comprising:
establishing a database of polarization profiles for a plurality of media properties, wherein the polarization profiles of the plurality of media properties comprise a plurality of polarization ratings corresponding to the plurality of media properties and reflecting a plurality of probabilities that the plurality of media properties comprise polarized viewers;
generating a truth set of viewers from online interactions of a plurality of computing devices by:
assigning a default probability that a truth viewer of the truth set of viewers belongs to a categorization segment;
identifying, from online behavior of the truth viewer, a media property of the plurality of media properties; and
determining a ground-truth categorization for the truth viewer based on a probability that the truth viewer belongs to a categorization segment, wherein the probability is based on combining the polarization ratings of the media property and the default probability;

generating signatures for the truth set of viewers by identifying online interactions of the truth set of viewers;
training a machine learning categorization model utilizing the signatures by predicting one or more categorization segments for the truth set of viewers and comparing the one or more predicted categorization segments to the ground-truth categorization;
identifying an impression opportunity from a remote server for a specific viewer in an online media auction;
applying the machine learning categorization model to a signature corresponding to the specific viewer to determine one or more categorization segments for the specific viewer; and
providing, based on the determined one or more categorization segments for the specific viewer, a bid in response to the impression opportunity for the specific viewer in the online media auction.

2. The computerized method of claim 1 , wherein generating the bid further comprises:
utilizing the machine learning categorization model to determine probabilities that the specific viewer belongs to the one or more categorization segments for the specific viewer; and
comparing the probabilities with a pre-determined probability threshold value.

3. The computerized method of claim 1 , wherein establishing the database of polarization profiles for the plurality of media properties further comprises:
determining a gain factor for a polarized viewer group and for a media property of the plurality of media properties based on activity of a least frequent viewer class relative to a number of known members of the viewer group; and
wherein a polarization rating of the plurality of polarization ratings for the polarized viewer group and for the media property is based at least in part on the determined gain factor.

4. The computerized method of claim 3 , wherein the polarization rating is further based on a product of:
the determined gain factor; and
a ratio of known viewers from the polarized viewer group for a media property and a total number of viewers for the media property.

5. The computerized method of claim 2 , further comprising:
providing the bid in response to determining that the specific viewer is a member of a categorization segment targeted by an online media campaign;
determining that the online media campaign is on target to under-fulfill the media campaign; and
automatically lowering the probability threshold value such that a larger number of impression opportunities are bid for.

6. The computerized method of claim 5 , further comprising:
determining that a campaign is on target to under-fulfill the media campaign;
determining that lowering the probability threshold value for the determination to bid has not brought campaign progress in line with fulfilling the media campaign; and
bidding on impression opportunities for unknown viewers visiting Media Properties having polarization characteristics consistent with demographic characteristics targeted by the campaign.

7. The computerized method of claim 2 , further comprising:
providing the bid in response to determining that the specific viewer is a member of a categorization segment targeted by an online media campaign; determining that the campaign is on target to over-fulfill the media campaign; and
automatically raising the probability threshold value such that a smaller number of impression opportunities are bid for.

8. The computerized method of claim 5 , wherein generating the bid further comprises setting a first probability threshold value for a first categorization segment and a second probability threshold value for a second categorization segment.

9. The computerized method of claim 8 , further comprising utilizing a user interface to set the first probability threshold value and the second probability threshold value.

10. The computerized method of claim 1 , wherein generating the signatures for the truth set of viewers further comprises generating a compilation of media properties accessed by devices of the truth set of viewers.

11. The computerized method of claim 8 wherein if the campaign is on target to over-fulfill the media campaign, the first and second probability threshold values are automatically raised such that a smaller number of impression opportunities per time are bid for.

12. The computerized method of claim 8 wherein if the campaign is on target to under-fulfill the media campaign, the first and second probability threshold values are automatically lowered such that a larger number of impression opportunities per time are bid for.

13. A system comprising:
at least one processor, and
at least one non-transitory computer readable storage medium storing instructions thereon that, when executed by the at least one processor, cause the system to:
establish a database of polarization profiles for a plurality of media properties, wherein the polarization profiles of the plurality of media properties comprise a plurality of polarization ratings corresponding to the plurality of media properties and reflecting a plurality of probabilities that the plurality of media properties comprise polarized viewers;
generate a truth set of viewers from online interactions of a plurality of computing devices by:
assigning a default probability that a truth viewer of the truth set of viewers belongs to a categorization segment;
identifying, from online behavior of the truth viewer, a media property of the plurality of media properties; and
determining a ground-truth categorization for the truth viewer based on a probability that the truth viewer belongs to a categorization segment, wherein the probability is based on combining the polarization ratings of the media property and the default probability;

generate signatures for the truth set of viewers by identifying online interactions of the truth set of viewers;
train a machine learning categorization model utilizing the signatures by predicting one or more categorization segments for the truth set of viewers and comparing the predicted one or more categorization segments to the ground-truth categorization;
identify an impression opportunity from a remote server for a specific viewer in an online media auction;
apply the machine learning categorization model to a signature corresponding to the specific viewer to determine one or more categorization segments for the specific viewer; and
provide, based on the determined one or more categorization segments for the specific viewer, a bid in response to the impression opportunity for the specific viewer in the online media auction.

14. The system of claim 13 , further comprising instru

Assignees

• Adobe Inc

Inventors

• Trenkle John M
• Hughes John
• Rose Adam
• Thakkar Kevin
• Lopatecki Jason

Classifications

• G06Q10/067

  Enterprise or organisation modelling · CPC title

• G06Q10/06375

  Prediction of business process outcome or impact based on a proposed change · CPC title

• G06Q30/0242

  Determining effectiveness of advertisements · CPC title

• G06Q30/0275 Primary

  Auctions · CPC title

• G06Q30/0204

  Market segmentation · CPC title