Browse Prior Art Database

Method and System for Building Audiences for Display Campaigns

IP.com Disclosure Number: IPCOM000237802D
Publication Date: 2014-Jul-14
Document File: 4 page(s) / 68K

Publishing Venue

The IP.com Prior Art Database

Related People

Ling Huang: INVENTOR [+2]

Abstract

A method and system is disclosed for building audiences for display campaigns using Collaborate Filtering (CF). The method and system uses a machine learning algorithm that is based on CF incorporating a massive category hierarchy. The method and system is implemented in a grid computing platform of Apache Hadoop*.

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 50% of the total text.

Method and System for Building Audiences for Display Campaigns

Abstract

A method and system is disclosed for building audiences for display campaigns using Collaborate Filtering (CF).  The method and system uses a machine learning algorithm that is based on CF incorporating a massive category hierarchy.  The method and system is implemented in a grid computing platform of Apache Hadoop*.

Description

Disclosed is a method and system for building audiences for display campaigns using Collaborate Filtering (CF).  The method and system uses a machine learning algorithm that is based on CF incorporating a massive category hierarchy.  The method and system is implemented in a grid computing platform of Apache Hadoop*.

In accordance with the method and system, a user response model is formulated as a supervised learning problem.  There are three key components for an advertisement request in a display advertising server: user, advertiser and publisher.  The user views the advertisement that is identified and stored as browser cookie in the server.  The advertiser is charged with executing an advertising campaign and the publisher indicates an online website inventory.  The advertisement request is denoted as a triplet:

The triplet is for an advertisement  and an advertiser. denotes all target users  and  includes all page  in available inventory.

In a scenario, if a Cost per Action (CPA) campaign is monitored, then Conversion Ratio (CVR) is used to monitor CPA campaign.  The CVR for a campaign is a proportion of targeted population who convert within a short period after displaying an advertisement.  After providing a successful conversion, a response class label is defined as , where 1 denotes a converted event and -1 denotes a non-converted event.  Thereafter, a set of impressions and responses are collected and divided as a training set and a test set, wherein

is denoted as the training set and is denoted as the test set.

The system collects and stores vast volume of temporal user events related to advertiser, and publisher and third party data while growing in size and scope.  An event corresponds to a user activity such as impression, click, and conversion.  Additionally, events are associated with metadata including advertiser hierarchy, publisher hierarchy and third party hierarchy.  An Open Segment Manager (OSM) collects, stores and integrates large scale event data using a massive category hierarchy.  All metadata across numerous data sources are stored in the OSM which are grouped by a hierarchical tree structure.

The method and system utilizes Collaborate Filtering (CF) algorithms to relate two different entities, users and items.  Additionally, a neighborhood based CF is used to focus on relationships between items and between users.  In order to implement the CF algorithm, identifier features are chosen that uses all binary indicators that take value of 1 when present and 0 otherwise.  A user by taxonomy membership...