NU 2012-084

Inventors

Lawrence A. Birnbaum*

Kristian J. Hammond*

Shawn O'Banion

Short Description

A profiling system that recommends media content based on user interests.

Background

In recent years, the nature of news consumption has changed in many ways. This is evident when considering the expansive forms of news media (mainstream, hyperlocal, blogs, social media, etc.) and multiple consumption devices (web browsers, tablet PCs, mobile devices, etc.) that are available to readers. With these options, publishers are eager to serve their consumers. Some news organizations have switched to an identity-based model of journalism, one which puts understanding the consumer's wants and needs above all else. Unfortunately, the technologies that are currently utilized still have difficulty in understanding the true preferences of readers in a way that is unobtrusive and automatic.

Abstract

With the advent of social media platforms, news publishers are keen to supply users with the content they are interested in without filtering through extraneous content to find it. Northwestern inventors have developed an unobtrusive and automatic content-based approach for modeling user interests. Using Twitter as a data source, this system recommends content, specifically news stories, on the basis of user's interests as determined from their own social media postings (and potentially postings of the people they 'follow' or 'like'). The user modeling technique employed identifies high-level categories (e.g., sports, politics) and detailed topics (e.g., Chicago Bulls, Mitt Romney) in a user's Twitter conversation, providing a comprehensive view of an individual's interests for the purpose of generating news story recommendations. This approach mines existing data that serves as evidence of users' interests and preferences. Historical Twitter data is used to identify instances where users share links to news articles, which is treated as an indicator of interest in the corresponding stories and topics. In a preliminary study, the inventors demonstrate that their system is more than twice as likely to recommend the exact story that was shared by the user, than alternative methods that only recommended 'popular' stories. In addition, it is 4x more likely to recommend similar stories as the one shared by the user.

Applications

• Personalized news content
• Personalized recommendations for other online content e.g. blogs

Advantages

• Automatic learning of user preferences
• Unobtrusive learning of user preferences
• Direct targeting of user interests
• Detects related topics of conversation that may not be explicitly mentioned in a tweet

Publications

O'Banion S, Birnbaum L and Hammond K (2012). Social media-driven news personalization. Proceedings of the 4th ACM RecSys workshop on Recommender systems and the social web. Dublin, Ireland, ACM: 45-52.

IP Status

Copyright © 2012 by Northwestern University, All Rights Reserved.