Browse Prior Art Database

A Method and System for Performing Sponsored Search Using a Temporal Click Model

IP.com Disclosure Number: IPCOM000199670D
Publication Date: 2010-Sep-14
Document File: 6 page(s) / 196K

Publishing Venue

The IP.com Prior Art Database

Related People

Eren Manavoglu: INVENTOR [+3]

Abstract

Disclosed is a method and system for performing sponsored search using a Temporal Click Model. The temporal click model is used to estimate positional bias, externalities and unbiased user perceived ad quality from user click logs.

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 19% of the total text.

A Method and System for Performing Sponsored Search Using a Temporal Click Model

Abstract

Disclosed is a method and system for performing sponsored search using a Temporal Click Model.  The temporal click model is used to estimate positional bias, externalities and unbiased user perceived ad quality from user click logs. 

Description

Disclosed is a method and system for performing sponsored search using a Temporal Click Model.  The temporal click model is used to estimate positional bias, externalities and unbiased user perceived ad quality from user click logs.  This model is based on a strict Bayesian frame work.  Closed-form representations of the ad quality and user behavior posteriors can be derived using this framework, making it scalable and computationally efficient to handle the challenges imposed by the voluminous click logs.

The method initially enables a user to start a query session by submitting a query to a search engine.  The search engine retrieves the ads that match the user query and finds the ranking that optimizes the objective function and presents them to the user in different slots on the results page, alongside organic web results.  The set of ads presented to the user can be represented as an ad impression sequence A =<a1 … aD>.  Here, ai is the ad presented in position i and D is the total number of slots.  An ad ai in this sequence is shown at a higher position (i.e. ranked higher) than ad aj if i < j.  The clicked ads may be represented as a sequence, a sequence of click events ordered by their time of click: C =<c1 …. cT>.  Here T< D and ci corresponds to the ad in position i.  This sequence is called as a temporal click sequence.

For an ad impression sequence A =<a1a2>, there are five possible click sequences: <>, <a1>, <a1a2>, <a2> and <a2a1>.  Multiple clicks on the same ad in the same position are discarded in the Temporal Click Model.  Thereafter, to have simplified notation and have sequences of equal lengths, the symbol o to indicate a no-ad-clicked action is used.  Thus, the five possible click sequences are : <oo>, <a1o>, <a1a2>, <a2o> and <a2a1>.  These sequences are referred as click sequences type 1..5.

The temporal click model may be described as a generative process as illustrated in figure.  The user submits a query and checks an ad impression sequence A =<a1a2>.  The user may either examine or ignore the ads.  This behavior of the user is considered as a probabilistic event.  The probabilistic event is represented by a binary random variable E.  The probability of examining ads, denoted as P (E = 1lA), is set to a global model parameter γ, i.e., P(E = 1lA) = γ

Figure

If the user decides not to ignore the ads, both of the ads are examined as the position rationality hypothesis states, and picks one.  The perceived quality of ads a1 and a2 are denoted by Ra1 and Ra2, respectively, where Rai Є [0..1].  The temporal click model suggests that the quality of a2 must be g...