Browse Prior Art Database

Method for Learning Click-Through Rates (CTR) for an Advertisement (Ad) Based on Its Intrinsic Qualities and a Position Bias Associated with an Ad Slot

IP.com Disclosure Number: IPCOM000198734D
Publication Date: 2010-Aug-13
Document File: 6 page(s) / 254K

Publishing Venue

The IP.com Prior Art Database

Related People

Tamas Sarlos: INVENTOR [+4]

Abstract

Disclosed is a method for learning click-through rates (CTR) for an advertisement (ad) based on its intrinsic qualities and a position bias associated with an ad slot where the ad is placed. The method estimates CTR for an ad irrespective of prior knowledge of position bias for ad slots.

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 52% of the total text.

Method for Learning Click-Through Rates (CTR) for an Advertisement (Ad) Based on Its Intrinsic Qualities and a Position Bias Associated with an Ad Slot

Abstract

Disclosed is a method for learning click-through rates (CTR) for an advertisement (ad) based on its intrinsic qualities and a position bias associated with an ad slot where the ad is placed.  The method estimates CTR for an ad irrespective of prior knowledge of position bias for ad slots.

Description

A method is disclosed for learning click-through rates (CTR) for an advertisement (ad) based on its intrinsic qualities and a position bias associated with an ad slot where the ad is placed.  The method estimates CTR for an ad irrespective of prior knowledge of position bias for ad slots.

In accordance with the method disclosed herein, a fixed search keyword is considered. Accordingly, A set of  of advertisers bid for the keyword, where a bid of advertiser . There are  positions (or slots) of

varying quality available to the search engine, where the quality of position  is given by  is a probability that a user inspects position ).  The ad slots are numbered in an order of non-decreasing quality, i.e..  The CTR  is a probability that a user clicks on ad i conditioned on being inspected.  Every time a user searches for the keyword, the search engine selects  advertisers from the set A and matches each selected ad with exactly one slot.  The expected revenue from an ad  assigned to position  is  and the total expected revenue is the sum of this quantity over all the selected ads.

A goal of a search engine is to maximize revenue for a keyword.  Accordingly, regret is required to be minimized.  The regret is defined as follows:

Let  denote list of the  advertisers in decreasing order of expected revenue conditioned on being inspected, i.e. .  Consider any (possibly randomized) online ad scheduling policy, POL, which selects an ordered list of ads  to display at time .  The list is ordered such that if  ad  is matched to slot  at time .  Then, the expected regret  of POL for a duration  is defined as the loss in revenue of POL compared to the optimal set of  ads till time :

The method employs the regret equation as defined above to estimate CTRs for ads by taking into considering intrinsic qualities of the ads and the position bias for ad slots.

In an embodiment, when the position biases for the ad slots are known, the method estimate CTRs in the following manner.  It is evident that some revenue is lost each time some ad  is shown in a lower position than it deserves (or not shown at all), or, equivalently, some ad is shown at a higher positio...