Relational RSS Clustering Techniques

1. INTRODUCTION There has been an explosion in the amount of available news and current event information on the Internet. 24-hours news outlets and independent bloggers alike flood the wires with a constant stream of data. Though an increasing number of people rely on the Internet as their primary source of news and current events, it is becoming increasingly difficult for users to find what they are looking for. Real Simple Syndication (RSS) provided a means to bring the data directly to the user through easily parsable XML feeds. Of course this only changed how users looked for news, not how much news there was. RSS aggregators, likes Google News, were created to learn a user's preferences and only display RSS stories that a user would like. Google News, in particular, has found a great deal of success applying algorithms from the text clustering community toward RSS data. Users are able to choose a story and the search engine provides a list of related stories that the user (hopefully) finds interesting. The potential applications of RSS clustering, however, go well beyond simple aggregation. In the future, Semantic-Web technologies may be able to actually extract knowledge from a stream of RSS feeds, i.e. find similarities, detect patterns, and infer things that the best human analysts cannot detect. This type of application is well beyond the scope of this project. Instead, we focus on aspect of such an application: sorting large amounts of RSS data. We present here the results of applying standard clustering techniques to the RSS problem, and an analysis of how well these techniques work. In Section 2 we describe some of the unique features that separate RSS clustering from other text clustering applications. The methods and results of our application are presented in Section 3, along with an extensive analysis of these results. We finish with some suggestions to improve future applications.