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 Scoring Bid Term Relevance in Sponsored Search

IP.com Disclosure Number: IPCOM000240223D
Publication Date: 2015-Jan-14
Document File: 4 page(s) / 465K

Publishing Venue

The IP.com Prior Art Database

Related People

Fabrizio Silvestri: INVENTOR [+2]

Abstract

A method and system is disclosed for scoring bid term relevance in sponsored search. Given a bid term and a landing page , defined as a URL and the relative HyperText Markup Language (HTML) content of the URL, the method and system aims at computing a function measuring the relevance of for . The method and system utilizes machine learning techniques to learn r from a ground truth consisting of bid terms, landing pages, and relevance judgments triples.

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 36% of the total text.

Method and System for Scoring Bid Term Relevance in Sponsored Search

Abstract

A method and system is disclosed for scoring bid term relevance in sponsored search.  Given a bid term  and a landing page , defined as a URL and the relative HyperText Markup Language (HTML) content of the URL, the method and system aims at computing a function  measuring the relevance of  for .  The method and system utilizes machine learning techniques to learn r from a ground truth consisting of bid terms, landing pages, and relevance judgments triples.

Description

Generally, problems might arise while solving partial or exact match queries in a sponsored search.  For example, if “Advertisement A” is associated with a "bad" keyword and a user submits a query matching that keyword, then “Advertisement A” might be returned and displayed even if “Advertisement A” is not related with the submitted query.  If users are often exposed to advertisements not corresponding to users' interests, the users can get annoyed which can decrease the overall revenue of a content publisher.

Disclosed is a method and system for scoring bid term relevance in sponsored search.  Given a pair (k,p), where k is a bid term that an advertiser has associated with a landing page p, the problem of evaluating r(k,p), the relevance of bid terms for a given advertisement is studied.  The method and system utilizes machine learning techniques to learn r from a ground truth consisting of bid terms, landing pages, and relevance judgments triples.  The method and system captures relationships between bid terms and associated landing pages so that the learned r minimizes for all the (k,p) pairs an error function measuring the difference between the predicted and the editorially associated labels.

The efficiency of the method and system is evaluated and demonstrated using typical relevance and accuracy metrics on the editorially judged dataset.  Various feature combinations are evaluated in order to gain insights on the set of features that better describes the dataset.  In all the tested cases, very high performance values are obtained for both a binary classification task (Area Under Receiver Operating Characteristic (ROC) Curve (AUC) up to 0.93) and a graded relevance tasks (AUC up to 0.95).

Consider that there are two landing pages and bid term relevance problems.  Given a bid term  and a landing page , defined as a URL and the relative HyperText Markup Language (HTML) content of the URL, the method and system aims at computing a function  measuring the relevance of  for .  Two different definitions of relevance are considered such as binary relevance and graded relevance.  In binary relevance model (BRel), the function for optimizing is selected among the functions of the form  meaning that  is (+1) or is not (-1) relevant to .  In the graded relevance model (GRel), the function is chosen among the set of functions of the form  corresponding to the labels bad, fa...