Bruce J. McKenzie

What do web users do? An empirical analysis of web use

This paper provides an empirical characterization of user actions at the web browser. The study is based on an analysis of 4 months of logged client-side data that describes user actions with recent versions of Netscape Navigator. In particular, the logged data allow us to determine the title, URL and time of each page visit, how often they visited each page, how long they spent at each page, the growth and content of bookmark collections, as well as a variety of other aspects of user interaction with the web. The results update and extend prior empirical characterizations of web use. Among the results we show that web page revisitation is a much more prevalent activity than previously reported (approximately 81% of pages have been previously visited by the user), that most pages are visited for a surprisingly short period of time, that users maintain large (and possibly overwhelming) bookmark collections, and that there is a marked lack of commonality in the pages visited by different users. These results have implications for a wide range of web-based tools including the interface features provided by web browsers, the design of caching proxy servers, and the design of efficient web sites.

3D or not 3D?: evaluating the effect of the third dimension in a document management system

Several recent research systems have provided interactive three-dimensional (3D) visualisations for supporting everyday work such as file and document management. But what improvements do these 3D interfaces offer over their traditional 2D counterparts? This paper describes the comparative evaluation of two document management systems that differ only in the number of dimensions used for displaying and interacting with the data. The 3D system is heavily based on Robertson et al.s Data Mountain, which supports users in storing, organising and retrieving “thumbnail” representations of documents such as bookmarked Web-pages. Results show that our subjects were faster at storing and retrieving pages in the display when using the 2D interface, but not significantly so. As expected, retrieval times significantly increased as the number of thumbnails increased. Despite the lack of significant differences between the 2D and 3D interfaces, subjective assessments showed a significant preference for the 3D interface.

An empirical analysis of web page revisitation

There is a surprising lack of empirical research into user interaction with the web. This paper reports the results of an analysis of four months of logged data describing web use. The results update and extend earlier studies carried out in 1994 and 1995. We found that web page revisitation is a much more prevalent activity than previously reported (approximately 80% of pages have been previously visited by the user), that most pages are visited for a surprisingly short period of time, and that users maintain large (and possibly overwhelming) bookmark collections.

An Evaluation of Cone Trees

Cone Trees are an appealing interactive 3D visualisation technique for hierarchical data structures. They were originally intended to maximise effective use of available screen space and to better exploit the abilities of the human perceptual system. Prior work has focused on the fidelity of the visualisation rather than providing empirical user studies. This paper describes the design, implementation and evaluation of a lowfidelity animated and rapidly interactive 3D cone tree system. Results of the evaluation show that our subjects were slower at locating data using cone trees than when using a ‘normal’ tree browser, and that their performance deteriorated rapidly as the branching factor of the data-structure increased. Qualitative results, however, indicate that the subjects were enthusiastic about the cone tree visualisation and that they felt it provided a better ‘feel’ for the structure of the information space.

Selecting a hashing algorithm

Hashing is so commonly used in computing that one might expect hash functions to be well understood, and that choosing a suitable function should not be difficult. The results of investigations into the performance of some widely used hashing algorithms are presented and it is shown that some of these algorithms are far from optimal. Recommendations are made for choosing a hashing algorithm and measuring its performance.

Error repair in shift-reduce parsers

Local error repair of strings during CFG parsing requires the insertion and deletion of symbols in the region of a syntax error to produce a string that is error free. Rather than precalculating tables at parser generation time to assist in finding such repairs, this article shows how such repairs can be found during shift-reduce parsing by using the parsing tables themselves. This results in a substantial space saving over methods that require precalculated tables. Furthermore, the article shows how the method can be integrated with lookahead to avoid finding repairs that immediately result in further syntax errors. The article presents the results of experiments on a version of the LALR(1)-based parser generator Bison to which the algorithm was added.

Compression of sparse matrices by arithmetic coding

The compression of matrices where the majority of the entries are a fixed constant (most typically zero), usually referred to as sparse matrices, has received much attention. We evaluate the performance of existing methods, and consider how arithmetic coding can be applied to the problem to achieve better compression. The result is a method that gives better compression than existing methods, and still allows constant-time access to individual elements if required. Although for concreteness we express our method in terms of two-dimensional matrices where the majority of the values are zero, it is equally applicable to matrices of any number of dimensions and where the fixed known constant is any value. We assume that the number of dimensions and their ranges are known, but will not assume that any information is available externally regarding the number of non-zero entries.

Compression of sparse matrices by blocked Rice coding

This correspondence considers the compression of matrices where the majority of the entries are a fixed constant (most typically zero), usually referred to as sparse matrices. We show that using Golomb or Rice encoding requires significantly less space than previous approaches. Furthermore, compared to arithmetic coding, the space requirements are only slightly increased but access is ten times faster for both Golomb and Rice encoding. By blocking the data, the access time can be kept constant as only a single block needs to be decoded to access any element. Although such blocking increases the space overheads, this is marginal until the block sizes become so small that only a few nonzero values will be found in a block. We provide formulas giving the space overhead of blocked Rice encoding and validate these empirically.

LR parsing of CFGs with restrictions

A method for extending the LR parsing method to enable it to deal with context‐free grammars containing embedded restrictions is presented. Such restrictions are usually dealt with in LR‐based parsers by executing semantic code outside the context of the LR method. By including such restrictions within the LR method itself, potential shift‐reduce and reduce‐reduce conflicts can be resolved and provide greater control over the language accepted. The proposed method can be incorporated easily into existing LR‐based parser‐generating systems.