Title: Personalized News Recommendation Based on Click Behavior
Author: Jiahui Liu, Peter Dolan, Elin Rønby Pedersen
Publisher: IUI '10, February 7-10, 2010 Hong Kong
Summary
Online news outlets have become a very popular way for people to access the news from millions of sources around the world. One challenge these news delivery organizations face is helping their users find interesting articles to read. The voluminous amount of articles available can be overwhelming to users.
Content-based recommendation is a response to this information overload. It plays a central role in recommendation systems. Some systems require users to manually create and update profiles, similar to Google News. Few users may be unwilling or too burdened to take on this extra step.
The researchers in this article have developed a way to automatically construct a recommendation system based on profiles learned from user activity in Google News. They first conducted a large-scale anonymous analysis of Google News users' click logs. Based on this analysis, they constructed a Bayesian framework for predicting users' current interests and trends. They combined the content-based recommendation mechanism which uses learned user profiles with an existing collaborative filtering mechanism to generate personalized recommendations.
Experiments on live traffic demonstrated that the hybrid method improved the quality of news recommendations and increased traffic to the web site.
Discussion
I thought this was interesting because it was something that we kind of talked about in class last week. This however is a little more about discovery of things that people like versus things that they may not know they liked. I use Google News a lot and there really only is a couple of sections that I actively follow. I could see myself using a system like this to recommend things I like.
Summary
Content-based recommendation is a response to this information overload. It plays a central role in recommendation systems. Some systems require users to manually create and update profiles, similar to Google News. Few users may be unwilling or too burdened to take on this extra step.
The researchers in this article have developed a way to automatically construct a recommendation system based on profiles learned from user activity in Google News. They first conducted a large-scale anonymous analysis of Google News users' click logs. Based on this analysis, they constructed a Bayesian framework for predicting users' current interests and trends. They combined the content-based recommendation mechanism which uses learned user profiles with an existing collaborative filtering mechanism to generate personalized recommendations.
Experiments on live traffic demonstrated that the hybrid method improved the quality of news recommendations and increased traffic to the web site.
Discussion
I thought this was interesting because it was something that we kind of talked about in class last week. This however is a little more about discovery of things that people like versus things that they may not know they liked. I use Google News a lot and there really only is a couple of sections that I actively follow. I could see myself using a system like this to recommend things I like.
I also couldn't help but wonder if Dr. Hammond had this in mind when we were discussing that earlier last week. This system sounds like it could be very useful to Google News users.
ReplyDelete