An Empirical Analysis of Information Filtering Methods

The growth in the the number of news articles, blogs, images, and videos available on the Web is making if more challenging for people to find potentially useful information People have relied on search engines to satisfy their short-term needs, such as finding the telephone number for a restaurant; however, these systems have not been designed to support long-term needs, such as the research interests of academics. One approach to supporting long-term needs is to use an Information Filtering system to select potentially useful information from the vast amount being produced everyday. The similarities between Information Retrieval systems and Information Filtering systems are well-established. They have prompted the use of retrieval models and methods in filtering systems, which has had some success but has been criticised as a limiting factor due to the unique challenges of document filtering. A significant difference between these systems is the use case: a filtering system is intended to push information to the user over a period of time, whereas a retrieval system is intended for the user to pull information to themselves for immediate use. The main challenge that needs to be addressed by a filtering system is the transient nature of the information published on the Web and the drifting nature of information needs. These factors lead to an uncertain interplay between the components comprising a filtering system and this thesis presents an empirical analysis of how the main system components affect performance. The analysis explores the role of each system component independently and in conjunction with other components. The main contribution of this thesis is a deeper understanding of how different components affect performance and the interplay between these components. The outcome of this thesis intends to act as a guide for both practitioners and researchers interested in overcoming some of the challenges of building filtering systems.