Latent user models for personalized ranking

Patent metadata

Field	Value
Publication number	US-9477757-B1
Application number	US-201213517767-A
Country	US
Kind code	B1
Filing date	Jun 14, 2012
Priority date	Jun 14, 2012
Publication date	Oct 25, 2016
Grant date	Oct 25, 2016

Abstract

A method includes generating a ranking model and a baseline mixing weight for each latent user category from a plurality of latent user categories based on a community preference dataset and one or more latent variables that relate the users from the community of users to the latent user categories. The method also includes generating a personalized mixing weight for each latent user category for a specified user based on an individual preference dataset, the ranking models for the latent user category, and one or more latent variables that relate the specified user to the latent user categories. The method also includes adjusting the personalized mixing weight for each latent user category for the specified user based on the baseline mixing weights, and generating ranking output for at least some objects from the plurality of objects using the personalized mixing weights and the ranking models.

First claim

What is claimed is:

1. A method comprising:
accessing a community preference dataset representing preferences of a community of users regarding a plurality of objects, wherein the community preference dataset includes a plurality of user-specified paired comparisons regarding the plurality of objects;
identifying a first user category in the community of users;
generating a ranking model and a baseline mixing weight for the first user category based on the community preference dataset;
accessing an individual preference dataset representing preferences of a first user of the first user category regarding at least one of the plurality of objects;
generating a personalized mixing weight for the first user based on the individual preference dataset and the ranking model for the first user category;
adjusting the personalized mixing weight for the first user based on a weighted average of the baseline mixing weight for the first user category and the personalized mixing weight for the first user, wherein the weighting of the weighted average is based on a size of the individual preference dataset; and
ranking at least two of the plurality of objects based on the adjusted personalized mixing weight for the first user and the ranking model for the first user category.

2. The method of claim 1 , wherein the adjusting of the personalized mixing weight comprises applying a smoothing algorithm.

3. The method of claim 1 , wherein the generating of the ranking model for the first user category comprises maximizing a log-likelihood:
log Π i=1 N p ( x i ,u i ),

wherein x i is at least one data element from the community preference dataset, u i is at least one user of the first user category, p(x i , u i ) is a probability that x i agrees with u i , and N is a number of data elements in the community preference dataset.

4. An apparatus, comprising:
one or more processors; and
one or more memory devices for storing program instructions used by the one or more processors, wherein the program instructions, when executed by the one or more processors, cause the one or more processors to:
access a community preference dataset representing preferences of a community of users regarding a plurality of objects, wherein the community preference dataset includes a plurality of user-specified paired comparisons regarding the plurality of objects,
identify a first user category in the community of users;
generate a ranking model and a baseline mixing weight for the first user category based on the community preference dataset,
access an individual preference dataset representing preferences of a first user of the first user category regarding at least one of the plurality of objects,
generate a personalized mixing weight for the first user based on the individual preference dataset and the ranking model for the first user category;
adjust the personalized mixing weight for the first user based on a weighted average of the baseline mixing weight for the first user category and the personalized mixing weight for the first user, wherein the weighting of the weighted average is based on a size of the individual preference dataset, and
rank at least two of the plurality of objects based on the adjusted personalized mixing weight for the first user and the ranking model for the first user category.

5. The apparatus of claim 4 , wherein the adjusting of the personalized mixing weight comprises applying a smoothing algorithm.

6. The apparatus of claim 4 , wherein the generating of the ranking model for the first user category comprises maximizing a log-likelihood:
log Π i=1 N p ( x i ,u i ),

wherein x i is at least one data element from the community preference dataset, u i is at least one user of the first user category, p(x i , u i ) is a probability that x i agrees with u i and N is a number of data elements in the community preference dataset.

Assignees

• Ning Huazhong
• Li Zhen
• Aradhye Hrishikesh
• Google Inc

Inventors

• Ning Huazhong
• Li Zhen
• Aradhye Hrishikesh

Classifications

• G06F16/951 Primary

  Indexing; Web crawling techniques · CPC title

• G06N20/00 Primary

  Machine learning · CPC title

• G06F17/30864 Primary

  Physics · mapped topic