IT&SOCIETY, VOLUME 1, ISSUE 3, WINTER 2003, PP. 159-183
`http://www.ITandSociety.org
`IMPROVING WEB PAGE REVISITATION:
`ANALYSIS, DESIGN AND EVALUATION
`
`
`
`ANDY COCKBURN
`SAUL GREENBERG
`STEVE JONES
`BRUCE MCKENZIE
`MICHAEL MOYLE
`ABSTRACT
`Several years of research suggest improvement is needed in how people return to
`their previously visited Web pages. Web page revisitation is one of the most frequent
`actions in computer use, so any interface improvements in this area can have a very large
`effect. Five categories of revisitation research are involved: 1) Characterizations of user
`behavior; 2) System models of navigation and their impact on the user’s understanding; 3)
`Interface methods for increasing the efficiency of the Back button; 4) Alternative system
`models for navigation; and 5) Alternative methods for presenting Web navigation histories.
`Revisitation is a dominant activity, with an average of 80% of page visits being to
`previously seen pages. The Back button is heavily used, but poorly understood.
`Three interface strategies for improving Web page revisitation are described: 1) A
`gesture-based mechanism for issuing the frequent Back and Forward commands addresses
`low-level interface issues; it is shown to be both popu lar and effective; 2) A ‘temporal’
`behavior for the Back and Forward buttons aims to overcome the problems associated with
`poor understanding of the current behavior of Back, strongly suggesting that revisitation
`can be improved by providing temporally ordered lists of previously visited pages; 3) Next-
`generation browsers could integrate the current tools for revisitation into a single utility,
`thus allowing simple visualization methods to aid users in identifying miniature target
`pages.
`______________________________
`Andy Cockburn is senior lecturer in the Dept. of Computer Science, University of Canterbury, Christchurch,
`NZ, wher e he directs the Canterbury Human-
`Computer Interaction and Multi- Media Lab.
`andy@cosc.canterbury.ac.nz.
`Saul Greenberg is professor in the Dept. of Computer Science at the University of Calgary, Canada, where he
`directs the Grouplab Laboratory for HCI and CSCW.
`Steve Jones is a senior lecturer in the Dept. of Computer Science at the University of Waikato, Hamilton, NZ.
`Bruce McKenzie is a software engineer at the Jade Development Centre in Christchurch, NZ.
`Michael Moyle is a software engineer at Allied Telesyn Research in Christchurch, New Zealand.
`
`
`
`
`© 2003 Stanford University
`
`APPLE 1023
`
`1
`
`

`

`160
`WEBPAGE REVISITATION COCKBURN, GREENBERG, JONES, MCKENZIE & MOYLE
`
`Every day, millions of people worldwide have problems trying to
`return to previously visited Web pages. These problems often amount to little
`more than the minor annoyance of finding that a page has “disappeared’” from
`those accessible with the Back button. Sometimes, however, extensive
`searching is necessary to return to a page;
` f o r example, users forget to
`bookmark a “valuable’” page, and they cannot remember how it was originally
`retrieved. Although the impact of these problems on each individual may be
`small, it is clear that easing these problems can yield enormous benefits when
`multiplied across millions of users, and billions of page accesses.
`The problems of revisiting Web pages have been examined since 1994,
`along with designs and evaluations of several systems aimed at improving
`Web-page revisitation. This article presents an integration and synthesis of
`this work, in the following order:
`1. User behavior. Recent results of a Web -use log-analysis show that
`revisiting pages is a dominant activity on the Web.
`2. System and user models of the current behavior of the Back button. An
`easy-to-repeat experiment demonstrates that many users
`misunderstand the rudimentary behavior of the main interface tool for
`revisitation—the Back button. Despite this misunderstanding, why is
`Back heavily used?
`. The efficiency
`3. Improving the efficiency of the Back command
`limitations of the interface mechanisms used to issue the Back
`command are described, also with an evaluation of a gesture-based
`shortcut (similar to the scheme recently provided in the Opera Web
`browser).
`4. Improving understanding and efficiency of the Back model. An
`alternative ‘temporal’ behavior for the Back and Forward buttons is
`described, with a presentation of the results of its evaluation.
`5. Improving the presentation of revisitation tools. The implications of
`the earlier findings demo nstrate how the next-generation of Web
`browsers could integrate and enhance the diverse tools for revisitation
`that are available in current browsers (Back/Forward, bookmarks, and
`history lists). This work is ongoing, and preliminary results are
`encouraging.
`
`
`USER BEHAVIOR : WHAT D O W EB USERS DO?
`
`Considering that Web browsers are among the most widely used
`computer applications, there has been only modest research into how they
`are used. This section briefly summarizes prior analyses of browsing behavior
`and then describes results from recent studies of what Web users do as they
`navigate the Web.
`
`IT&SOCIETY, Vol. 1, Issue 3, Winter 2003 http://www.ITandSociety.org
`
`
`2
`
`

`

`161
`WEBPAGE REVISITATION COCKBURN, GREENBERG, JONES, MCKENZIE & MOYLE
`
`Prior analyses: Some researchers detail the demographics of Internet
`users (e.g., gender percentages, age, occupations, educational attainment,
`etc.), as well as the demographics of their technologies (e.g., people’s
`connection speed to the Internet and browser selection). The most well-
`known of these is the 1994-1998 biannual WWW Surveys
`(http://www.cc.gatech.edu/user_surveys), developed by the Graphics,
`Visualization and Usability Center at the Georgia Institute of Technology
`(Kehoe and Pitkow 1996). Recently, sociologists have begun studying Internet
`use, with a natural emphasis on how Web use changes social structures,
`rather than focusing on the efficacy of the user interfaces used to navigate the
`Web (DiMaggio et al. 2001).
`Other researchers have analyzed the tasks that people do as they
`navigate. For example, Byrne et al. (1999) videotaped eight people as they
`used their browser over the course of their day and codified user behaviors.
`From this, they developed a task-based taxonomy of browsing, including the
`six, general Web tasks:
`1. Use information—a series of activities in which people use information
`gathered from the Web,
`2. Locate on page—search for particular information on a page,
`3. Go to—the act of trying to get the browser to display a particular URL,
`4. Provide information —sending information to the browser (e.g.,
`authentication, addresses, search terms),
`5. Configure browser—changing the configuration of the browser itself,
`and
`6. React to environment—supplying information or dealing with a
`problem on demand of the browser.
`The authors then proceed to sub-divide these general tasks into more
`specific ones, and to codify how often they occur. While these results reflect a
`total of only a few hours of use by a few people, it provides insight into the
`actual things that people do.
`Researchers have analyzed traces (or logs) of users’ actions to reveal
`statistics of use. Some use server-side logs. Since most Web servers collect
`data indicating when a particular page has been accessed and by what IP
`address, this is an easy data source to mine. For example, it is relatively
`simple to analyze logs to expose the frequency of page hits on a Web site.
`However, server-side logs are limited in that they often do not distinguish
`well between different users; they collect no data on actual browser use, and
`they are often missing crucial data. Pirolli, Pitkow, and Rao (1996) and Chi,
`Pirolli and Pitkow
`(2000) discuss the problems of extracting meaningful
`information from server-side logs. Another option is to specially equip the
`browser so that it logs the users’ actions. The advantage of this “client-side
`
`IT&SOCIETY, Vol. 1, Issue 3, Winter 2003 http://www.ITandSociety.org
`
`
`3
`
`

`

`162
`WEBPAGE REVISITATION COCKBURN, GREENBERG, JONES, MCKENZIE & MOYLE
`
`logging” is that it can record the exact history of the user’s actions with his or
`her particular browser.
`Perhaps the most well-known, client-side log analyses of Web use are
`by Catledge and Pitkow (1995) and Tauscher and Greenberg (1997). These
`studies instrumented the then-popular XMosaic browser to record the pages
`that users visited and the interface mechanisms used to access them. The
`participants in both studies were primarily staff, faculty and students in
`university computing departments. Catledge and Pitkow logged three weeks
`of use by 107 users in 1995, whereas Tauscher and Greenberg analyzed five to
`six weeks of use by 23 users in 1995. Catledge and Pitkow revealed that the
`dominant user interface techniques for visiting pages were clicking on
`hypertext anchors (52%) and on the Back button (41%). Navigating to pages
`by typing the URL, clicking Forward, or selecting from ‘Bookmarks’ were all
`lightly used, accounting for about 2% each. Tauscher and Greenberg
`confirmed that link selection and clicking Back are the dominant navigation
`mechanisms, accounting for approximately 50% and 30% of navigation acts.
`Tauscher and Greenberg also analyzed the recurrence rate of page
`visits: “the probability that any URL visited is a repeat of a previous visit,
`expressed as a percentage.” They found that the recurrence rate for the
`subjects participating in their study was 58%, and a reanalysis of the data
`from 55 of Catledge and Pitkow’s subjects produced a recurrence rate of 61%.
`This result shows that users had previously seen approximately 60% of pages
`visited. Tauscher and Greenberg’s analysis also reveals that users tend to
`revisit pages just visited a short while ago, access only a few pages
`frequently, browse in very small clusters of related pages, and generate only
`short sequences of repeated URL paths.
`In these early studies, people rarely used bookmarks (less than 2% of
`user actions). However, a later 1996 survey by Abrams et al. (1998) suggested
`that bookmark use was rising. Some 84% of his respondents had more than
`eleven bookmarks, indicating that people at least had the intention of
`returning to key pages. Indeed, Pitkow (1996) reported from a survey of 6619
`users that “organizing retrieved information” is one of the top three usability
`problems of using the Web, reported by 34% of participants.
`
`Our Recent Client-Side Log Analysis of Web Use: While all exemplary
`studies, the findings of these early studies may not reflect current use of the
`Web. Things have changed considerably since the mid–90s. Instead of early
`adopters and technologically savvy ‘elite’ users, people from virtually all
`demographic backgrounds use the Web (although not equally). Modern Web
`browsers have highly polished interfaces with features far beyond those
`supported by XMosaic. Many Web navigation aids such as search engines and
`Web directories are now a fundamental part of Web use, yet these tools were
`either in their infancy or did not exist at the time of the prior studies. The
`technology itself has changed—broadband connections now give nearly
`
`IT&SOCIETY, Vol. 1, Issue 3, Winter 2003 http://www.ITandSociety.org
`
`
`4
`
`

`

`163
`WEBPAGE REVISITATION COCKBURN, GREENBERG, JONES, MCKENZIE & MOYLE
`
`instantaneous response—a sharp contrast to the “World Wide Wait’” of the
`early days. The kinds of information accessible on the Web are also
`completely unlike those of the prior decade. Once an academic repository, the
`Web is now a massive commercial and populist arena.
`METHOD
`
`To update and extend the findings of these previous studies, the Web-
`browsing activities of 17 Computer Science staff and graduate students were
`analyzed over a 119-day period from early October 1999 to late January 20001.
`The use of computer scientists in these evaluation(s) introduces obvious risks
`of generalizing the r esults to other communities, but one suspects that
`computer scientists use the Web for purposes similar to those of most users.
`Both the Catledge and Pitkow and the Tauscher and Greenberg studies used
`computer scientists.
`The data were gathered through the history and bookmark files that
`Netscape Navigator (versions 4.5–4.7) maintains. Netscape Navigator was the
`browser used by the participants in their everyday work, and its user
`interface features are similar to those of other popular browsers. The
`browser’s history file keeps a list of the URLs the user has visited, the time of
`the user’s last and first visit, the number of visits and the title of each page.
`The bookmark file stores information about the user’s bookmarks and their
`organization into folders. Copies of these files were obtained through
`incremental backups that were automatically created every night at the
`University of Canterbury. To eliminate the chance that participants would
`modify their behavior due to their awareness that their actions were being
`logged, participants were asked for permission to retrieve their backup files
`after the terminating date of the study. That is, participants were asked to
`have their data mined retrospectively, instead of asking them permission to
`monitor their future Web uses.
`
`Pages visited per day: The participants made a total of 84,841 page
`visits, spread across 17,242 different URLs, averaging approximately 42 pages
`each day. The actual number of pages visited by each user per active day
`would be substantially higher than this, because the average includes
`weekends and the Christmas/New Year vacation period during which few
`participants would have used their browsers at work. Even with this
`underestimation, these results suggest that people visit almost double the
`number of pages now than in the mid-90s. That is, the mean of 42 pages per
`day compares to approximately 14 (Catledge and Pitkow 1995) to 21 pages per
`day (Tauscher and Greenberg 1997) in the earlier studies.
`
`The
`Percentage of pages that participants had previously seen:
`premise behind the Back button, history systems, and bookmarks is that
`people frequently navigate to pages that they have seen before. As already
`
`IT&SOCIETY, Vol. 1, Issue 3, Winter 2003 http://www.ITandSociety.org
`
`
`5
`
`

`

`164
`WEBPAGE REVISITATION COCKBURN, GREENBERG, JONES, MCKENZIE & MOYLE
`
`mentioned, previous studies have found that the average revisitation rate is
`between 58-61%. Has this figure changed? To analyze revisitation, the total
`pages visited were plotted over time by each participant (Figure 1a). The
`mean total number of page visits by each participant was 4991 (s.d.= 6106),
`ranging from 281 to 24,309. (Participant 15’s heavy Web-use statistics can be
`explained by his role as a Web-master.) The distinct pages visited over time
`by each participant were then plotted in Figure 1b. Here, the mean number of
`URLs visited was 1227 (s.d. = 1086), with a range from 74 to 4251. As the
`figures indicate, the total number of page visits is strongly correlated with
`the growth of the number of distinct URLs visited. Linear regression over all
`subjects gives a slope of 5.1 and an R-squared value of 0.8837 (F1,940=7140,
`p<0.0001). This slope reflects the revisitation rate for the subject pool: for
`each new URL added to the set of distinct URLs, four pages are revisited.
`This revisitation rate is confirmed using Tauscher and Greenberg’s formula
`for revisitation rate:
`
`Revisitation Rate:
`
`=
`
`R
`
`100
`
`*
`
`total
`
`_
`
`visit
`count
`total
`_
`URL
`_
`count
`visit
`total
`_
`_
`
`_
`
`count
`
`
`
`This calculates a revisitation rate of 81%, with individual participant
`revisitation rates ranging from 61% to 92%—substantially higher than the
`previously reported values of 58% and 61%.
`
`Use of bookmarks and other shortcuts to pages: One factor behind a
`high revisitation rate is that almost all participants had one or two pages that
`they visited far more often than any other. Participant 2, for instance, had
`visit counts of 4352, 384, 199, and 117 for his top four pages. In general,
`participants’ top three pages accounted for 20% of all page accesses.
`Given the high visit counts to particular pages, one would have
`thought these would be prime candidates for becoming shortcuts. Netscape
`Navigator supports a variety of shortcut techniques—including the
`configurable ‘Home’ button, bookmarks, and the “personal toolbar.’”
`Netscape’s history and bookmark files were analyzed to see whether people
`created shortcuts to their frequently visited pages. Although most users were
`found to have shortcuts to their top two pages, few had shortcuts for their
`third, fourth and fifth most frequently visited pages.
`
`Analysis of the bookmark file showed wide variation in bookmark use:
`participant 16 had none, while participants 2 and 8 had more than 500 each.
`The mean maximum size of the subjects’ bookmarks collection was 184, but as
`
`IT&SOCIETY, Vol. 1, Issue 3, Winter 2003 http://www.ITandSociety.org
`
`
`6
`
`-
`

`

`165
`WEBPAGE REVISITATION COCKBURN, GREENBERG, JONES, MCKENZIE & MOYLE
`
`F IGURE 1 : GROWTH IN T HE T OTAL N UMBER OF P AGES V ISITED (A) AND
`DISTINCT P AGES V ISITED (B) OVER T IME (MM / YY) FOR E ACH U SE
`
`(a) Total pages visited
`
`
`
`
`
`
`
`(b) Distinct pages visited.
`
`IT&SOCIETY, Vol. 1, Issue 3, Winter 2003 http://www.ITandSociety.org
`
`
`7
`
`

`

`166
`WEBPAGE REVISITATION COCKBURN, GREENBERG, JONES, MCKENZIE & MOYLE
`
`suggested above, the standard deviation was very high (s.d. = 166). Some
`people organized the bookmarks into folders, where the mean number of
`folders used to store bookmarks was 18
`(although again the standard
`deviation is very high—s.d. = 16). We saw that the rate of bookmark addition
`heavily outweighed the rate of deletion, with averages of 27.6 additions and
`3.7 deletions. This imbalance implies that users have, or will have, problems
`managing the size and organization of their bookmark collections over time.
`For example, bookmarks in Netscape are normally selected via a pop-up,
`cascading menu, the length of which depends on the number of “top level”
`items in the bookmark file. Participant 17 had 130 items in this top-level,
`which would produce a very cumbersome cascading menu.
`Figure 2 plots the number of items in the top-level bookmark structure
`for three of the subjects over time. The obvious steps in the figure show that
`participants periodically tried to reorganize their bookmarks to overcome the
`problem of the menu growing too long; this effect was also noted by Abrams et
`al. (1998). Rather than deleting items, subjects would typically relocate them
`to new folders (counted as a top-level item). It is worth noting that the
`interface mechanisms for managing bookmarks have improved since Netscape
`4.5–4.7, possibly easing bookmark management.
`The apparent reluctance to delete bookmarks is at odds with the
`relatively transient nature of Web sites and pages. Two months after
`collecting the bookmark data, scripts were run that attempted to access each
`page in the participants’ bookmark collections. Any page returning 404 “Not
`found”, 301 “Moved Permanently”, or 5xx (host unavailable) was deemed
`invalid. Approximately 25% of the pages were invalid. This indicates that
`over time, bookmark collections will become cluttered with useless items.
`
`Proportion of pages that do not have titles: Titles are used by Netscape
`and Microsoft Internet Explorer in a variety of ways, including labeling items
`on the Back pull-down menu, default-identification tags in the bookmark and
`history lists, and labeling the window-manager border. Missing, incorrect and
`inconsistent titles can frustrate the user’s ability to identify pages to which
`they wish to return (Cockburn and Greenberg 2000).
`Some 5% of the distinct URLs visited by the participants did not have
`an HTML “Title” tag associated with the page. Although there are alternative
`interface techniques that could be used to aid page identification, such as
`thumbnail images of the page as in Cockburn et al. (1999); Kaasten et al.
`(2002) and Robertson et al. (1998), it is likely that text titles will remain an
`important page-identification cue. There is little that any browser can do to
`ensure the presence and accuracy of the text titles, but Web
`-authoring
`software could promote careful consideration of page titles. Often page
`-
`titling facilities are ‘hidden’ under sub-menus or dialogue boxes, rather than
`prominently displayed for each newly authored page.
`
`
`IT&SOCIETY, Vol. 1, Issue 3, Winter 2003 http://www.ITandSociety.org
`
`
`8
`
`

`

`167
`WEBPAGE REVISITATION COCKBURN, GREENBERG, JONES, MCKENZIE & MOYLE
`
`F IGURE 2 : N UMBER OF URL S AND F OLDERS IN THE T O P-L EVEL B O O K M A R K F ILE OF T HREE U SERS
`SHOWING DRAMATIC ‘PRUNING’ OF THE B O O K M A R K STRUCTURE, P LOTTED A GAINST T IME (MM / YY)
`
`
`
`
`Summary: In summary, the main findings of this log-analysis study are
`as follows. Web-page revisitation is the dominant activity in Web browsing.
`On average, users have previously seen four out of every five pages visited.
`Most participants had one or two pages that they visited far more often than
`all other pages. Bookmark use was highly varied, with users either making
`light or heavy use of bookmark collections. Finally, there were two
`indications that users find bookmark management troublesome: first, 25% of
`the bookmarks did not refer to legitimate pages when one tried to access
`them; second, some of the heavy bookmark users appeared to be forced into
`‘shuffling’ bookmarks when their top-level menus became too long.
`
`SYSTEM AND U SER MODELS: THE B ACK B UTTON
`
`
`The experimental results summarized in the previous section show the
`importance of having efficient, e
`ffective mechanisms for returning to
`previously visited Web pages. The Back button is one of the main interface
`components for returning to Web pages, accounting for approximately 40% of
`navigational actions, according to Tauscher and Greenberg (1997), yet the
`easy-to-repeat experiment described below shows that many users
`misunderstand its behavior. Having described the experiment, it is possible
`to speculate why the Back button is heavily used despite the common
`misunderstanding.
`
`
`IT&SOCIETY, Vol. 1, Issue 3, Winter 2003 http://www.ITandSociety.org
`
`
`9
`
`

`

`168
`WEBPAGE REVISITATION COCKBURN, GREENBERG, JONES, MCKENZIE & MOYLE
`
`: A simple, mass-participation,
`Understanding of the Back button
`informal study that takes only minutes to conduct demonstrates how Web
`users misunderstand the behavior of the Back button.
`To run the informal study in a classroom, one can use a Web browser
`projected onto a large screen, telling the audience that they will be asked to
`predict whether a page can be returned to using only the Back button, not by
`using links or other history tools. Have the browser showing a home page
`that is well-known to the audience, such as the “Computer Science
`Homepage.” Then slowly and clearly use a link to visit a page off the home
`page—for example, “People in Computer Science.” Tell the audience that they
`will revisit this page shortly. Clearly demonstrate the use of the Back button
`to return to the homepage. Next, slowly and clearly use a different link to
`visit another page off the home page. With the browser displaying this final
`page, ask the students to write on a scrap of paper whether the first linked
`page (“People in Computer Science”) can be returned to using the Back
`button and only the Back button (no links or history tools). Ask them to write
`the number of times the Back button will have to be clicked if the page is
`accessible with Back. Finally, collect the responses. Using paper responses
`encourages participation, and protects the participants from possible
`embarrassment in a show of hands. The point is that stack-based behavior of
`Back means that the page is inaccessible with Back.
`A more elaborate version of this experiment was first conducted in
`1995, when the Web was still relatively new (Cockburn and Jones 1996). The
`eleven participants were all computer scientists who used the Web daily.
`Eight wrongly predicted that the page could be accessed with Back. Since
`then, we have repeated the experiment as described above many times in
`large Computer Science undergraduate classes. Approximately 50% of the
`students—all of whom use the Web in their studies—misunderstood whether
`pages were accessible with Back, with the most common error being to state
`that the “People in Computer Science” page can be returned to with two Back
`clicks.
`
`Stack-based behavior: All major commercial browsers use the same
`stack-based model for Web-page navigation. In this model, there are two
`ways—“load” and “revisit”—of displaying pages in the browser. Pages are
`loaded when the user clicks on a link, types a URL, selects a “Favorite’” page,
`and so on. The effect of “load” is to add the page to the top of a stack of visited
`pages. Pages are revisited with the Back and Forward buttons, which move
`downward and upward through the stack of visited pages. The menus
`associated with Back and Forward allow users to directly revisit pages on the
`stack. When the user loads a page after revisiting pages, the new page is
`added to the stack immediately above the current stack position, and all
`pages above that stack position are removed. This explains why the “People
`in Computer Science” page cannot be revisited in the previous example.
`
`IT&SOCIETY, Vol. 1, Issue 3, Winter 2003 http://www.ITandSociety.org
`
`
`10
`
`

`

`169
`WEBPAGE REVISITATION COCKBURN, GREENBERG, JONES, MCKENZIE & MOYLE
`
`•
`
`•
`
`Given the notation fi to indicate loading a page (e.g., clicking on a
`link), and the notation (cid:239) to indicate revisiting a page with Back, if the user
`navigates through pages afibfic, then the stack contains {a, b, c}, the
`underscore indicating the page being displayed in the browser. If the user
`then navigates back to a by pressing the Back button twice (c(cid:239)b(cid:239)a) then the
`stack contains {a, b, c}. If the user then clicks on a link to page h (afih), then
`b and c are removed from the stack, giving {a, h }.
`
`What is good about the Back button: The two primary limitations of
`Back are that many users misunderstand its operation and that not all
`recently seen pages can be revisited. Yet, despite these limitations, Back is
`heavily used. There are several factors contributing to Back’s
`success
`(Cockburn and Greenberg 2000):
`•
`
`It can allow rapid return to recently visited pages (assuming they are
`still on the stack).
`It is robust. People can use it, even with a naïve model of the way it
`works.
`k through pages
`It is cognitively undemanding. Users can backtrac
`using a simple “click until the desired page is recognized” strategy.
`It is ‘ready-to-hand’. Unlike interface features such as the ‘History list’,
`which must be explicitly popped up when needed, there is little
`overhead in accessing Back because it is on constant display.
`• It consumes minimal screen real estate.
`These beneficial properties have led to the Back button being one of the most
`heavily used interface components in existence.
`
`Summary: A large proportion of regular Web users misunderstand the
`stack-based behavior of the Back button. Despite this misunderstanding, they
`use it heavily, because it has many desirable properties. However, this
`conflict between heavy use and misunderstanding clearly results in
`suboptimal use.
`The remaining three sections describe the design and evaluation of
`three approaches for improving human performance with the Back button.
`The first approach investigates the efficiency of using a shortcut “gesture”
`action for issuing the Back command. The second evaluates the potential
`cognitive and motor benefits of changing the stack-based Back model to a
`complete temporal list, overcoming the problems of pages disappearing off
`the set accessible with Back. The third investigates the potential benefits of
`integrating the diverse schemes for Web revisitation currently used in
`browsers.
`
`
`•
`
`IT&SOCIETY, Vol. 1, Issue 3, Winter 2003 http://www.ITandSociety.org
`
`
`11
`
`

`

`170
`WEBPAGE REVISITATION COCKBURN, GREENBERG, JONES, MCKENZIE & MOYLE
`
`IMPROVING THE E FFICIENCY OF THE B ACK C OMMAND
`
`
`The mouse-driven cursor is the main input device for Web
`navigation—to visit a link the user points the cursor to the link, and to revisit
`a page, the user moves the cursor to the Back button. Fitts’ Law (Fitts 1954)
`predicts the time taken to move the cursor between targets (such as the Back
`button and the page links), but faster alternatives than mouse-pointing are
`available.
`In recognizing the importance of providing efficient mechanisms for
`issuing the Back command, both Microsoft Internet Explorer and Netscape
`Navigator provide keyboard shortcuts, with Internet Explorer using
`backspace or Alt+left-arrow, and Netscape Navigator using Alt+left-arrow.
`The main limitation of key-bindings for Web browsing is the overhead in
`homing the hands between the mouse (used for links) and the keyboard.
`Homing is necessary for the Alt+left-arrow keybinding, because many
`keyboards are arranged so that the Alt and left-arrow keys cannot be
`simultaneously pressed with one hand. Similarly, for right-handed users, the
`backspace key-binding is awkward because the user must either reach across
`the keyboard with his left hand, or take his right hand off the mouse. Another
`shortcut for Back, provided by both browsers, is the context menu that can be
`popped up by pressing the right-mouse button. Accessing the Back menu item
`incurs overheads in waiting for the menu to be posted and in the Fitts’ Law
`limitations of pointing to the menu item. Finally, specialized input devices
`such as the Microsoft Intellimouse Explorer
`(http://www.microsoft.com/hardware/mouse/ie_info.asp) mouse provide
`additional buttons for the Back and Forward commands.
`These shortcuts are available only to the relatively small proportion of
`users who buy the devices. From experience, studies and subjects’ comments,
`it appears that the Back command is seldom issued through any interface
`mechanism other than the Back button. The question, then, is how to improve
`the efficiency of issuing the Back command, without the low-level costs of
`target acquisition or of homing the hands between keyboard and mouse.
`Inspired by work on gesture-based marking menus (Callahan et al.
`1988), in 2001 a gesture-based mechanism for issuing the Back and Forward
`commands was designed, implemented and evaluated. Similar features were
`simultaneously released in the Opera (
`http://www.opera.com) commercial
`Web browser,
` and slightly later by the Mozilla Optimoz project
`(http://www.optimoz.org). Neither of these commercial implementations has
`apparently been formally evaluated.
`With gesture navigation, the user “flicks” the mouse, with the left
`button held down. A leftward flick navigates Back, a rightward flick Forward.
`To support gesture navigation within unaltered commercial browsers in this
`evaluation, a website was constructed where each page contained a
`Javascript program instructing the browser to navigate Back or Forward
`
`IT&SOCIETY, Vol. 1, Issue 3, Winter 2003 http://www.ITandSociety.org
`
`
`12
`
`

`

`171