SWAMI: A Multiagent, Active Representation of a User’s Browsing Interests

SWAMI: A Multiagent, Active Representation of a User’s Browsing Interests

Mark Kilfoil (University of New Brunswick, Canada) and Ali Ghorbani (University of New Brunswick, Canada)
DOI: 10.4018/978-1-60960-523-0.ch011
OnDemand PDF Download:
$30.00
List Price: $37.50

Abstract

The rapid growth of the World Wide Web has complicated the process of Web browsing by providing an overwhelming wealth of choices for the end user. To alleviate this burden, intelligent tools can do much of the drudge-work of looking ahead, searching and performing a preliminary evaluation of the end pages on the user’s behalf, anticipating the user’s needs and providing the user with more information with which to make fewer, more informed decisions. However, to accomplish this task, the tools need some form of representation of the interests of the user. This article describes the SWAMI system: SWAMI stands for Searching the Web with Agents having Mobility and Intelligence. SWAMI is a prototype that uses a multi-agent system to represent the interests of a user dynamically, and take advantage of the active nature of agents to provide a platform for look-ahead evaluation, page searching, and link swapping. The collection of agents is organized hierarchically according to the apparent interests of the user, which are discovered on-the-fly through multi-stage clustering. Results from initial testing show that such a system is able to follow the multiple changing interests of a user accurately, and that it is capable of acting fruitfully on these interests to provide a user with useful navigational suggestions.
Chapter Preview
Top

The Web is a relatively new phenomenon, and has elevated certain problems to a critical level. In this section, the two most prominent problems of Web navigation and Web personalization are discussed, and a short summary of current solutions is presented.

Web Navigation

Because of its large size, dynamic nature and inconsistent structure, the Web is difficult to navigate. “Traditional,” direct navigation approaches depend on an evaluation of the relevance of the currently viewed page as the best indicator of the value of pages pointed to by the current page. This approach relies upon the benevolence of the creator of the link (Kleinberg, 1999), and the hope that by following a series of related links the user will end up at another cluster of useful pages. This “hope” is described as the “small world” phenomenon, which suggests that a highly complex but interacting system will, over time, evolve paths of a limited number of hops between any two related pages.

When the user has discovered a page of lessening interest to them than a previous page, they return backward to an appropriately interesting (although already viewed) page and go forward from a link on that page (if there is one) until all links from that page have been exhausted, retreating back up another level. This navigation strategy is implicitly promoted by the linear nature of Web navigation tools, such as the “back” button of a Web browser.

The traditional strategy closely resembles a depth-first graph search, where leaf nodes are represented by pages of less interest. Effectively, however, the user must go “one page too far” in such a scheme, and travel deeper and deeper distances from the original page they were browsing into possibly uninteresting areas. Due to the highly connective nature of the Web, this suggests that the user will spend more time in distant pages than in pages more closely connected to the original. This, intuitively, is the opposite of the desired result, as pages directly connected to the current page are most likely to be the most relevant pages to it (Lieberman, 1995).

Complete Chapter List

Search this Book:
Reset