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 Providing Score Look-Alike Recommendations

IP.com Disclosure Number: IPCOM000246821D
Publication Date: 2016-Jul-04
Document File: 5 page(s) / 694K

Publishing Venue

The IP.com Prior Art Database

Related People

Qiang Ma: INVENTOR [+4]

Abstract

A method and system is disclosed for scoring users within an expanded audience in a way which relates directly to a business metric that an advertiser wants to optimize. The method and system utilizes three scoring models and incorporates the potential value of a user to an advertiser into the scoring models for significantly improving look-alike models.

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 31% of the total text.

Method and System for Providing Score Look-Alike Recommendations

Abstract

A method and system is disclosed for scoring users within an expanded audience in a way which relates directly to a business metric that an advertiser wants to optimize.  The method and system utilizes three scoring models and incorporates the potential value of a user to an advertiser into the scoring models for significantly improving look-alike models.

Description

Look-alike models are quickly revolutionizing the online programmatic advertising industry as efficient tools to expand the reach to a quality audience from a smaller customer set.  The traditional look-alike models mainly rely on the similarity assumption based on behavior similar to the behavior of a set of input users leading to higher conversion rate and are not optimized against any business metrics of the advertiser.  In other words, these models directly model the expanded user set as a set with a membership similarity measure and there is no direct way to estimate the potential value of a given user within the expanded user set for a particular advertiser.

Disclosed is a method and system for scoring users within an expanded audience in a way which relates directly to a business metric that an advertiser wants to optimize.  The method and system utilizes three scoring models and incorporates the potential value of a user to an advertiser into the scoring models for significantly improving look-alike models.

Look-alike models based on pairwise similarities rely on the assumption that targeting users with similar behavioral patterns to the input set of users will correlate with an advertiser’s goal.  However, these goals could vary significantly between advertisers operating in different contexts and hence the potential value of a user to an advertiser is to be explicitly incorporated in the scoring models.

The method and system utilizes a general pairwise similarity-based approach, called Scoring Look-Alike based on the Locally Sensitive Hashing (LSH) algorithm that incorporates information from the direct goal of the advertiser in a balanced way.  However, the Scoring Look-Alike approach was redesigned to take into account the specifics of the in-memory computational model and calculates the distributed signature matrix and other side information within the computational environment.  The side information (e.g. conversion related information) directly relates to the advertiser's goal leading to a more balanced model which incorporates that side information with the similarities to calculate the final scores measuring the user’s potential value for the advertiser.

The method and system utilizes three scoring models.  The first scoring model scores users purely based on similarity to an input seed set.  The second model ranks users based on the importance of various features pertaining to the users for predicting value of the users to the advertiser.  The third model combines the results...