Paper Reading #19: Personalized News Recommendation Based on Click Behavior

Reference Information:
Title: Personalized News Recommendation Based on Click Behavior
Authors: Jiahui Liu, Peter Dolan, Elin Rønby Pedersen
Presentation Venue: IUI’10, February 7–10, 2010, Hong Kong, China


Summary:

Google News logo from news.google.com

This paper discusses the need for news websites to provide easier access to news articles the user might be interested in. The problem with news sites such as Google News is the number of articles available for the user, there are a ton. This paper suggests using the users web history to track what articles they have read before and creates a profile for each user. This allows Google News to provide the user with news topics that they are interested in based on their history.


The framework used for this project considers a couple of ways to track recommended articles for users. Genuine articles are the articles that the pulled directly from the the user's web history results. It is also important to track local news event unique to the area the user is located. The example used is Spanish users read more about sports during the Euro Cup. After the genuine articles have been found, more articles are recommended if the users current interests match anything going on in local events. This is important because user's news interests change over time. The genuine articles represent the overall interests of the user from their entire history of the site. The current interests is reflected in the current events cross reference.


To finally determine if the article will appear on a user's recommended articles the framework does some calculations. A content-based recommendation score is calculated based on topics and the users previous clicks, this is done using a unique equation. A collaborative filtering score is then calculated using a different method. Multiplying these two together gives the overall score and the order the articles are ranked in.


To test the method vs the old method used Google just used their active users. A control group was set up using the new method discussed in this paper and an equal sized group was set up to use the old method. It was shown that more recommended articles were clicked rather than specific sections(such as sports) with the new method. It was also shown that users using the new method visited the site more frequently.

Discussion:
I like this idea a lot because it seems like it would make browsing news quicker and more enjoyable. The only problem I can think of that people may not like is the site using web-history to generate a recommended list. However, the site said it was only web-history from the user's actions on Google News so I don't think that is a problem. What I liked most about this idea was that there was already a method that news sites used but it is further being evolved into what looks like a  more successful and efficient method. One last thing I like is how they considered people want to view news quickly rather than browse through a ton of content and also how they linked it to getting more people to go to their site for business purposes because of this feature.

Just noting that when I went to the Google News and couldn't find a recommended tab after logging in and browsing a little. I noticed that they added a section based on my location but I didn't see any recommended tab.