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

System and Method for Improving a User's Demographic Predictions by Using Both First Party and Third Party Data

IP.com Disclosure Number: IPCOM000246610D
Publication Date: 2016-Jun-21
Document File: 5 page(s) / 202K

Publishing Venue

The IP.com Prior Art Database

Related People

Datong Chen: INVENTOR [+2]

Abstract

Disclosed is a method and system for improving a user's demographic predictions by using both first party and third party data. The method and system leverages two information sources for predicting the user's age and gender namely an aggregated third party feedback in batch level and a social site declared demography for certain users. The two information sources are used to learn two types of knowledge, namely knowledge of predicting demography of the users in a user feature space and a cumulative knowledge for each individual user through long-term third party feedback.

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 40% of the total text.

System and Method for Improving a User’s Demographic Predictions by Using Both First Party and Third Party Data

Abstract

Disclosed is a method and system for improving a user’s demographic predictions by using both first party and third party data.  The method and system leverages two information sources for predicting the user’s age and gender namely an aggregated third party feedback in batch level and a social site declared demography for certain users.  The two information sources are used to learn two types of knowledge, namely knowledge of predicting demography of the users in a user feature space and a cumulative knowledge for each individual user through long-term third party feedback.

Description

A method and system is disclosed for improving a user’s demographic predictions by using both first party and third party data.  The method and system leverages two information sources for predicting the user’s age and gender namely an aggregated third party feedback in batch level and a social site declared demography for certain users.  The two information sources are used to learn two types of knowledge, namely knowledge of predicting demography of the users in a user feature space and a cumulative knowledge for each individual user through long-term third party feedback.

The method and system provides a standard machine learning model to learn two types of information collected from both the first party and the third party users.

The first type of knowledge for predicting demography of the users in a user feature space is learnt by extracting the user features from user behaviors on web, mobile device and advertisements.  Subsequently, the user feature space can be modeled to better leverage the data across multiple users with same features.  Further, the statistical machine learning model also addresses the prediction problem for new users or users without any third party feedback.

Similarly, the second type of knowledge is the cumulative knowledge for each individual user through long-term third party feedback or the first party declaration.  Here, the feature based model does not give very accurate prediction if the learning model does not have strong predictive signals.  Therefore, the modeling at individual level is a good compensation in the second type of knowledge by collecting the third party feedback from multiple batches for a same user.

In accordance with a scenario, the method and system is explained by considering a first or single batch of users for predicting demography of users by using both the first party and the third party data. 

For first N impressions, let us assume pd for d = 1,…..D =14 is the ground truth proportion from the third party with Ʃd pd = 1.  Here the ground truth can be, but not limited to an age and gender of a user.  Subsequently, let Y denote the users having the first party data.  Therefore, the long-likelihood of the objective function can be:

where L3 is the third party...