Citation indexes are valuable tools for research, in part because they provide a means with which to measure the relative impact of articles in a collection of scientific literature. In retrieval systems for citation indexes, recent work has demonstrated the benefit of using ranking metrics based on measures of impact. While this approach is e#ective in identifying a few of the most important contributions to an area, many documents ranked highly in response to queries are irrelevant to the topic of interest. The problem here is that with such techniques Boolean methods are used to identify candidates for retrieval, even though such methods are poor determiners of relevance. As a solution to this problem, we present an indexing technique that pulls together measures of relevance and significance in a single retrieval metric. This approach, which we call Reference Directed Indexing (RDI) is based on a comparison of the terms authors use in reference to documents. Initial retrieval experiments with RDI indicate that it retrieves documents on par with significance-based techniques in terms of impact, and comparable to traditional vector-space approaches with regard to relevance.

The rapid increase of World Wide Web users and the development of services with high bandwidth requirements have caused the substantial increase of response times for users on the Internet. Web latency would be significantly reduced, if browser, proxy or Web server software could make predictions about the pages that a user is most likely to request next, while the user is viewing the current page, and prefetch their content.

This paper provides a synopsis of a server-driven link prefetching mechanism that we have designed and implemented for the Mozilla web browser, a popular Open Source web browser. The mechanism depends on the origin server or an intermediate proxy server determining the best set of documents for the browser to prefetch. The browser follows prefetch directives provided by the server, either embedded in an HTML document using the <LINK> tag or specified via Link HTTP response headers. The browser determines when best to prefetch the specified URLs based on its own heuristics. In this paper, we describe the mechanism and discuss some of the practical issues that impacted its design and implementation.

User-perceived retrieval latencies in the World Wide Web can be improved by pre-loading a local cache with resources likely to be accessed. A user requesting content that can be served by the cache is able to avoid the delays inherent in the Web, such as congested networks and slow servers. The difficulty, then, is to determine what content to prefetch into the cache. This work explores machine learning algorithms for user sequence prediction, both in general and specifically for sequences of Web requests. We also consider information retrieval techniques to allow the use of the content of Web pages to help predict future requests. Although history-based mechanisms can provide strong performance in predicting future requests, performance can be improved by including predictions from additional sources. While past researchers have used a variety of techniques for evaluating caching algorithms and systems, most of those methods were not applicable to the evaluation of prefetching algorithms or systems. Therefore, two new mechanisms for evaluation are introduced. The first is a detailed trace-based simulator, built from scratch,

As the World Wide Web matures, it makes leaps forward in both size and complexity. In this expanding environment, the needs and interests of individual users become buried under the sheer weight of possible viewing choices. To counter this, there has been a rise in research in adaptive websites, a combination of data mining, machine learning, user modelling, Human Computer Interaction (HCI), optimization theory and graph theory which seeks to sift through the tides of possible pages to provide users with a high-quality stream of information. This paper provides a description of adaptive website research, including the goals aimed at, the challenges discovered and the approaches to solutions.

Summary. Most requests on the Web are made on behalf of human users, and like other human-computer interactions, the actions of the user can be characterized by identifiable regularities. Much of these patterns of activity, both within a user, and between users, can be identified and exploited by intelligent mechanisms for learning Web request patterns. Our focus is on Markov-based probabilistic techniques, both for their predictive power and their popularity in Web modeling and other domains. Although history-based mechanisms can provide strong performance in predicting future requests, performance can be improved by including predictions from additional sources. In this chapter we review the common approaches to learning and predicting Web request patterns. We provide a consistent description of various algorithms (often independently proposed), and compare performance of those techniques on the same data sets. We also discuss concerns for accurate and realistic evaluation of these techniques. 1

We study the performance of multiuser document prefetching in a two-tier heterogeneous wireless system. Mobility-aware prefetching was previously introduced to enhance the experience of a mobile user roaming between heterogeneous wireless access networks. However, an undesirable effect of multiple prefetching users is the potential for system instability due to the racing behavior between the document access delay and the user prefetching quantity. This phenomenon is particularly acute in the heterogeneous environment. We investigate into alleviating the system traffic load through prefetch thresholding, accounting for server queuing prioritization. We propose a novel analysis framework to evaluate the performance of the thresholding approach. Numerical and simulation results show that the proposed analysis is accurate for a wide variety of access, service, and mobility patterns. We further demonstrate that stability can be maintained even under heavy usage, providing both the same scalability as a non-prefetching system and the performance gain associated with prefetching.

Analysis of web site usage data involves two significant challenges: firstly the volume of data, arising from the growth of the web, and secondly, the structural complexity of web sites. In this paper we apply Data Mining and Information Visualization techniques to the web domain in order to benefit from the power of both human visual perception and computing; we term this Visual Web Mining. In response to the two challenges, we propose a generic framework, where we apply Data Mining techniques to large web data sets and use Information Visualization methods on the results. The goal is to correlate the outcomes of mining Web Usage Logs and the extracted Web Structure, by visually superimposing the results. We propose several new information visualization diagrams and analyze their utility and elaborate on the architecture of a prototype implementation.

Traditional information retrieval systems have ignored the potential improvement in precision provided by personalization. We present a study of the behavior and evaluation of personalized information retrieval systems. We describe the construction of a collection of user web browsing data for application in retrieval evaluation. Several novel techniques for personalizing retrieval are presented and evaluated. Although performance is mixed, results point to the need to develop other algorithms within this evaluation framework.

This paper presents the design and implementation of a Web prefetching module in Mozilla, an open-source and cross-platform browser. We have incorporated two kinds of predictors: a historybased predictor and a content-based predictor. These two predictors can analyze a user’s behavior and the contents of recent HTML pages to predict likely next links; thus, they provide personalized predictions which are then utilized to determine which resources should be prefetched. 1