FINDING AND REFINDING WEB PAGES IN CONTEXT
A Tree-based Model of Web History
David Briffa and Chris Staff
Department of Intelligent Computer Systems, University of Malta, Tal-Qroqq, Msida MSD 2080, Malta
Keywords: Web history navigation, Revisiting web pages in context, Automatic query generation, Global
reconnaissance, Tree-views, User modelling.
Abstract: A modern challenge for the World Wide Web (Web) is not of just finding information without getting ‘lost’
in hyperspace, but also re-finding it efficiently. Web Nav is an integrated navigation system that combines
both history and page recommendations into one context based tool. Web Nav’s framework signifies a
paradigm shift in the viewing of history from a linear stack-based system to a hierarchal tree-based system.
Web Nav was evaluated qualitatively and quantitatively, analysing 13 users’ activity over a seven day
period. The results are mixed but there is sufficient evidence to suggest tree-based views of history can be
beneficial: to allow users to revisit web pages in context; to show user sessions as trees; and to
automatically generate queries based on session contexts to recommend web pages.
1 INTRODUCTION
In this paper, we tackle two major Web Navigation
problems: finding and re-finding information.
Modern search engines generally do not take the
session context into account (i.e., a group of web
pages related to some task being performed by the
user). Two different users searching for the term
‘jaguar’ will be presented with the same results set,
even though in the session contexts, one user has
been visiting web pages related to jaguar the animal
and the other has been visiting pages related to
Jaguar the car. We also tackle the problem of re-
finding information. Between 58% and 81% of page
visits are page re-visits (c.f. section 2.3), so
mechanisms for organizing and easily accessing
already found information are important.
In our opinion, representations of history should
preserve the contextual structure in which web pages
are visited. Context can also help to automatically
construct queries to find more relevant information.
Web Nav (Briffa, 2010) incorporates context and
global reconnaissance into a tree-based model of
history.
2 FINDING INFORMATION
Finding information on the Web can be seen as a
combination of searching and browsing (Herder,
2006). Searching involves submitting a query to a
search engine, and browsing involves navigating
between pages using hyperlinks (Herder, 2004). One
of the main problems on the Web is the ‘Lost in
Hyperspace’ problem, where the user starts
browsing, and finds herself ‘lost’ in terms of where
she has been and where she intends to go.
Adaptation tools can give a sense of direction such
as through page recommendations or direct guidance
(Brusilovsky, 2001).
In a Web Browser, searching is catered for
through search engines and embedded widgets that
quickly return results for a query. Search can be
improved by building user models from visited
pages. PowerScout (Lieberman, Fry, & Weitzman,
2001) builds a user model from the pages visited and
constructs a query that is submitted to a third party
search engine, returning page recommendations.
A Web user must constantly choose between
browsing links or initiating a search for pages. This
is directly related to Information Foraging Theory
(Pirolli & Card, 1999). According to this theory,
users try to maximize the ratio of energy spent and
information gain. An automated approach may use
both neighbouring pages and a search engine to
determine which strategy will be more successful.
FollowMyLink (Briffa, 2009) integrates search
and browsing by turning user selected text on a Web
page into hyperlinks on-the-fly. The user is taken to
the top ranking page in the results set following a
426
Briffa D. and Staff C..
FINDING AND REFINDING WEB PAGES IN CONTEXT - A Tree-based Model of Web History.
DOI: 10.5220/0003401004260429
In Proceedings of the 7th International Conference on Web Information Systems and Technologies (WEBIST-2011), pages 426-429
ISBN: 978-989-8425-51-5
Copyright
c
 2011 SCITEPRESS (Science and Technology Publications, Lda.)

query generated from contextual information.
FollowMyLink maintains separate user models as a
user browses, using page relations and browsing
behaviour to decide which user model to update.
When a user selects text on a Web page and invokes
FollowMyLink, a query is automatically generated
from the user model which is updated after each link
traversal. Y!Q (Kraft, Maghoul, & Chang, 2005)
uses only the context supplied by text surrounding a
user selection. Y!Q may take the user to a page the
user has visited recently, whereas FollowMyLink
takes the user to a previously unseen relevant page.
2.1 Re-finding Information
Apart from searching and browsing, a third element
to navigation is backtracking. This involves going to
a previously visited page. A browser supports a
number of tools for both short term and long term
backtracking, such as the Back/Forward buttons,
history lists, bookmarks, and the History window.
Backtracking in Web navigation is important,
considering the number of page visits on the Web
that are actually re-visits. (Tauscher & Greenberg,
1996) give a figure of 58% in 1996, while
(Cockburn & Mckenzie, 2000) calculated a value of
81% in 2000. Herder estimates that 74% of page
visits are revisits (Herder, 2006). Reasons for
revisiting pages include: checking if information has
changed; authoring a page; exploring it further; or
the page is on a path to another revisited page
(Tauscher & Greenberg, 1996).
Of particular interest are the latter two cases.
First, the need to explore pages further may result in
a page being bookmarked. Although bookmarking is
efficient with regards to space and organization, any
context gathered from previous pages is lost from
one session to another, and hence when returning
there is no context or user model available for the
page in question. Second, the mention of a path
raises questions on whether bookmarking one page
is enough, or whether pages the user followed on the
way to the bookmarked page should also be
persisted. Empirical evidence shows that the path is
not only important, but should also be a factor in
history mechanisms (Teevan, 2004), with
‘waypoints’ such as page titles and descriptions to be
considered as supplementary metadata to the path
(Capra & Perez-Quinones, 2003). Tauscher and
Greenberg provide a comprehensive analysis of the
types of history mechanisms available, and focus
particularly on the stack based history list as applied
to the Back and Forward buttons. A drawback of this
approach is that the resulting history list does not
contain all visited pages, due to how all pages above
the stack pointer are removed during backtracking
(Cockburn & Jones, 2000). Cockburn and Jones also
suggest that users themselves have a skewed
understanding of how the history list works.
Tauscher and Greenberg refer to context sensitive
subspace history lists, which suggest the grouping of
pages in history into a subspace based on context.
Other systems may give a view of history in the
form of maps or views. WebView (Cockburn, A.,
Greenberg, S., McKenzie, B., Jasonsmith, M., &
Kaasten, S, 1999) and WebNet (Cockburn & Jones,
2000) generate overviews of the users browsing
path. As the views are not persisted to memory, they
are not useful for long-term backtracking. CZWeb
(Fisher, B., Agelidis, M., Dill, J., Tan, P., Collaud,
G., & Jones, C., 1997) uses a fish-eye view.
Figure 1: Tree Structure for a hypothetical path.
3 WEB NAV
We built a Mozilla Firefox Extension for the
persistence and use of context-based paths. As
Firefox uses a stack-based paradigm for its tab
history, we designed a separate framework so Web
Nav can store its own copy of navigation history as a
tree-based structure in an SQLite database. The
overall design rationale is to use this framework as a
basis for providing both context-based history
information and context-based page
recommendations so that users are never ‘lost’ or
unable to backtrack to a specific page.
3.1 Web Nav History
Web Nav tracks visited pages in a Context Model
through tab-based storage using the Mozilla Session
Store API. We identify the corresponding node of a
page by checking the Session Store. Paths are built
by making connections between nodes to reflect the
FINDING AND REFINDING WEB PAGES IN CONTEXT - A Tree-based Model of Web History
427

type of traversal performed. Figure 1 shows a typical
tree structure, where node A is the current page.
The tree shows the page where a link was clicked
to visit page A and other branches in the path. Paths
are acyclic. An instance of page B is added as a
child to page A, even though it already exists as the
root. A page may be seen in multiple contexts, and
in multiple instances within the same context.
We distinguish between two movement forms:
browsing - where a new node is created, and moving
- where we simply move to a different node in the
tree. This can be equated to backtracking in the
conventional sense. Each action resulting in a new
page-load saves the appropriate information in
Session Store so that the subsequent page-load event
handler may process the new page correctly and
create the necessary links in the database. If the
movement is a backtrack, then the nodeID of the
target node to move to is stored in Session Store.
Otherwise the nodeID of the parent node is stored so
that it may be used to create the appropriate
connection to any newly created node.
3.2 Interests and Recommendations
Web Nav saves page interests and uses them to
provide recommendations. Each ‘node’ saved in a
context can have associated NodeInterests that are
compiled through an indexing procedure. As in
FollowMyLink (Briffa, 2009), we collect the
relevant text from the corresponding page, and after
stemming and a modified TF.IDF calculation we
create a set of weighted keywords representing the
most relevant terms in the page. Each set of
NodeInterests is also used to update a set of
ModelInterests for the respective model. We save
NodeInterests and ModelInterests for the entire
model because we use two recommendation
algorithms. ModelInterests are used for the
algorithm that creates a query based on terms from
the entire context (tree) (Briffa, 2009). NodeInterests
are used in the algorithm that generates a query
based on the current branch, effectively creating an
on-the-fly merge of the nodes in the current path.
We use this algorithm to provide recommendations
that are localized to the path, rather than generalized
to the entire tree.
3.3 Adaptations
To support visual adaptations resulting from the
approach described in sections 3.1 and 3.2, we
implemented both a tab-based interface and a global
interface. For tab-based history and
recommendations we implemented the Web Nav
Popup view (figure 2).
Figure 2: Web Nav Popup View showing options at the far
left, history information on the left and page
recommendations on the right.
The view is split into two sections. The left side
shows the history information for the tab graphically
as either a tree or a linearly ordered view. In the tree
view, the parent and children of the current page are
shown. The user can navigate through the tree to see
the entire history for the tab’s context. In the linear
view a user may choose to show all the nodes in a
context sorted by some criteria, such as recency or
frequency.
Figure 3: Web Nav Manager showing a list of contexts at
the top left, with a preview below. A map of the selected
context is shown to the right, where nodes are coloured
based on frequency of visits.
A user can return to a previously visited context
using the Web Nav Manager (figure 3). The user can
see all the persisted contexts, as well as a map view
of all the paths saved. Users can jump back into a
session, with all context information saved.
4 EVALUATION
We conducted a preliminary evaluation for the gene-
WEBIST 2011 - 7th International Conference on Web Information Systems and Technologies
428

ral usefulness of Web Nav and the appeal of the
paradigm involved in two stages. 14 volunteers used
the system in their own home for approximately 7
days, and submitted a qualitative questionnaire about
their experience with Web Nav.
The first stage of evaluation used empirical data
from action logs and database entities to determine
which users exhibited low browsing and low
backtracking behaviour. One user’s log file was
corrupt and another four users were removed as they
did not use the browser enough to yield any
meaningful data. The second stage concerned a
deeper analysis into the empirical data. It showed
that the ‘Up’ and ‘Down’ buttons were used less
than the regular ‘Back’ and ‘Forward’ buttons.
Qualitative data suggests that the reason might be
habitual. Surprisingly, although the usage data was
poor, qualitative preference showed that 38.5% of all
13 volunteers (excluding the one with a corrupt log
file) preferred Up/Down. The Web Nav popup view
was used regularly throughout the evaluation period,
and the Web Nav bookmarking feature was used
occasionally. However, only a few users actually
used recommendations, and even less re-visited
recommended pages, possibly due to a construction
bug found after evaluation had commenced,
especially since 84.6% of users indicated interest in
having recommendations provided and 69.2% of
users said the recommendations generated were
‘somewhat relevant’. Qualitative data suggested that
both recommendation methods were equally
preferred. The Web Nav Manager, while not used as
often as the Web Nav popup view, still showed
promise, especially since in many cases the opening
of the Web Nav Manager resulted in the
backtracking to a node in a previous context session.
The overall experience of users seems to have
been positive, with 100% of users agreeing that the
concept of paths shown as trees is useful/interesting.
Moreover, 69.3% of users expressed interest in
possibly using Web Nav in the future.
5 CONCLUSIONS
We have shown an approach to a tree-based
navigation system that has yielded fairly promising
results in light of the overwhelming shift required to
change to a new paradigm. Its main contribution lies
in its underlying framework for the persistence of
tree-based contexts.
REFERENCES
Briffa, D., 2009. FollowMyLink: An Implementation of
User Directed Browsing. (Unpublished report), Dept.
of Intelligent Computer Systems, University of Malta.
Briffa, D., 2010, Web Nav: A Context Based Navigation
Assistant for the Web. (Unpublished Report), Dept. of
Intelligent Computer Systems, University of Malta.
Brusilovsky , P., 2001. Adaptive Hypermedia. User
Modeling and User-Adapted Interaction, 11, 87-110,
Kluwer Academic Publishers.
Capra, R. G., and Perez-Quinones, M. A., 2003. Re-
Finding Found Things: An Exploratory Study of How
Users Re-Find Information. Technical Report,
Virginia Tech.
Cockburn, and Jones., 2000. Which way now? Analysing
and easing inadequacies in WWW navigation.
International Journal of Human-Computer Studies,
45, 105-129.
Cockburn, A., and Mckenzie, B., 2000. What Do Web
Users Do? An Empirical Analysis of Web Use.
International Journal of Human-Computer Studies,
903-922.
Cockburn, A., Greenberg, S., McKenzie, B., Jasonsmith,
M., and Kaasten, S., 1999. WebView: A Graphical
Aid for Revisiting Web Pages. Proceedings of
OZCHI'99 Australian Conference on Human
Computer Interaction.
Fisher, B., Agelidis, M., Dill, J., Tan, P., Collaud, G., and
Jones, C., 1997. CZWeb: Fish-eye views for
visualizing the World Wide Web. M. J. Smith, G.
Salvendy & R. J. Koubek Design of Computing
Systems: Social and Ergonomic Considerations, 2,
719--722.
Herder, E., 2006. Forward, Back and Home Again -
Analyzing User Behavior on the Web. (Doctoral
dissertation), University of Twente. Amsterdam: F&N
Boekservice.
Herder, E., 2004. Sniffing Around For Providing
Navigation Assistance. Proc. of Workshop on
Adaptivity and User Modeling in Interactive Systems.
Berlin.
Kraft, R., Maghoul, F., and Chang, C. C., 2005. Y!Q:
contextual search at the point of inspiration.
Proceedings of the 14th ACM international conference
on Information and knowledge management (CIKM
'05). ACM, New York, NY, USA, 816-823.
Lieberman, H., Fry, C., and Weitzman, L., 2001.
Exploring the Web with Reconnaissance Agents.
Communications of the ACM, 44(8), 69-75.
Pirolli, P., and Card, K. S., 1999. Information Foraging.
Psychological Review, 106, 643-675.
Tauscher, L., and Greenberg, S., 1996. Design Guidelines
for Effective WWW History Mechanisms. Workshop
on Designing for the Web: Empirical Studies.
Microsoft Corporation, Redmond, WA.
Teevan, J., 2004. How people re-find Information when
the Web changes. Massachusetts Institute of
Technology Computer Science and Artificial
Intelligence Laboratory.
FINDING AND REFINDING WEB PAGES IN CONTEXT - A Tree-based Model of Web History
429