Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

Method and System for Estimating a Quality versus Reach Curve for a Model-based Advertisement Campaign

IP.com Disclosure Number: IPCOM000237442D
Publication Date: 2014-Jun-18
Document File: 5 page(s) / 201K

Publishing Venue

The IP.com Prior Art Database

Related People

Mihajlo Grbovic: INVENTOR [+2]

Abstract

A method is disclosed for estimating a quality versus reach curve for a model-based advertisement campaign by leveraging on existing advertisement campaigns that are already running and potentially similar advertisement campaigns that are being booked. The method includes allowing an advertiser to observe the estimated quality of the advertisement campaign.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 40% of the total text.

Method and System for Estimating a Quality versus Reach Curve for a Model-based Advertisement Campaign

Abstract

A method is disclosed for estimating a quality versus reach curve for a model-based advertisement campaign by leveraging on existing advertisement campaigns that are already running and potentially similar advertisement campaigns that are being booked.  The method includes allowing an advertiser to observe the estimated quality of the advertisement campaign. 

Description

Disclosed is a method for estimating a quality versus reach curve for a model-based advertisement campaign by leveraging on existing advertisement campaigns that are already running and potentially similar advertisement campaigns that are being booked.  The method includes allowing an advertiser to observe the estimated quality of the advertisement campaign.

 

In accordance with the method, objective of the advertiser is identified.  For example, type of an advertisement campaign, category of the advertisement campaign, region in which the advertisement campaign will be run and average number of users that are being targeted for the advertisement campaign are identified.  Focusing on a subset of users helps attain a high response rate for the advertisement campaign while reducing various costs to both advertisers and publishers. 

The method involves identifying user actions and profile of the user wherein the profile of the user discloses gender, age and online activity information of the user.

It is assumed that an advertiser provides a list of positive users wherein the positive users purchased a product in the last month.  This is used to train the predictive model. 

The predictive model is trained for J existing campaigns,  = , where  is a probability of user action.  Given model predictions  the classification predictions are made by thresholding, as  where threshold is usually set to 0.5, . However, θ may be chosen anywhere between 0 and 1 to ensure a desired False Positive Rate (FPR) or True Positive Rate (TPR).  By sliding  a Receiver Operating Characteristic (ROC) curve may be formed.  Alternatively,  may be chosen to achieve a desired reach. 

To generalize,  may be any model capable of producing probability scores. A J seedlist sets for existing campaigns

where all examples are positives for a specific campaign and the query seedlist set  with  examples.  The  is

The J predictive models for existing campaigns, denoted as  are assumed to be already available.  Each model  may be used to score and sort all the users in descending order.  This results in a vector  where  =  and .  This may be used to plot a quality vs reach curve.

The output is  for the query seedlist and the output assists in estimating the probability.  The output is estimated automatically without training predictive model . 

Thereafter,  is estimated as an average of existing ,

.

A weighted average of existing  may be estimated based on campaign type...