3D Visualization of Hierarchical Clustered Web Search Results


Mehrnaz Sadat Akhavi 1, Mohammad Rahmati 1, Nerssi Nasiri Amini 2
1 Computer Engineering Department, Amirkabir University of Technology, Tehran, Iran
2 Computer Engineering Department, Iran University of Science and Technology, Tehran, Iran
{Akhavi, Rahmati}@aut.ac.ir, Nasiri_amini@comp.iust.ac.ir


Abstract

The immaturity of semantic search engines has
initiated researchers to apply various novel post-
processing techniques on traditional search engine
results among which clustering routines are most
conspicuous. While many of these routines are focused
on hierarchical clustering, little has been done toward
an effective visualization of such data. Due to the
richness of information observed in 3D in comparison
with 2D (because of its abundant visual cues), we have
proposed an intuitive 3D metaphor for the
visualization of hierarchical clustered results. Our
metaphor is based on fractal trees which are usually
constructed by linear recursive algorithms traversing
node-link hierarchies. The search results of different
hierarchical clustering algorithms shall be visualized
as either single-tree or forest metaphors which are
user’s customizable alternatives in our implemented
prototype. In the forest depiction, our approach is even
applicable to flat clustered results.

Keywords: Search results visualization, tree
visualization, hierarchical clustering, virtual reality,
human-computer interfaces.

1. Introduction

The information overload and the rapid growth of
resources on the World Wide Web, makes search
engines one of the most important and frequent
services used to find required information over the net.
However searching the web is sometimes frustrating,
not only because of the underlying information
retrieval technology but also because of the
contemporary web-based user-interfaces. The most
sate-of-the-art user-interfaces for the conventional
search engines typically return long lists of ordered
documents, ranked and displayed linearly in html
format (usually 10 results per each page). Users are
forced to sift through a great number of results which
are not often relevant to their interests. A recent study
[1] shows that 88% of users will try a new search if
they do not find what they seek in the first three pages.
Reviewing 30 results ordered by ranking mechanism
which often does not reflect the users' preferences,
makes them revise their search query or shift to another
search engine. Therefore applying post-retrieval
document visualization techniques seems to be helpful.
Clustering the search results and improving the current
textual user-interfaces into graphical are the main
approaches in this regard.
Search results clustering algorithms attempt to
group results together based on their similarities; thus
results relating to a certain topic will hopefully be
placed in distinct clusters. Clustering algorithms are
distinguished by their implementation as flat clustering
(only one-level partitioning of the data) or hierarchical
clustering. Hierarchical approaches result in a tree-like
construction where the clusters of closely related
documents are nested within bigger clusters containing
documents that are less similar. Organizing the huge
amount of results to nested clusters, allows users to
refine the search, starting from very general topics and
moving towards the more detailed ones. Hierarchical
clustering is so popular that it has been extensively
employed in most industrial systems (e.g. Vivisimo,
Mooter, IBoogie and Clusty), open source solutions
(e.g. Carrot2) and research prototypes (e.g. Credo or
SnakeT) [2].
Presentation interface is another important post-
processing approach toward representing the search
results in a more effective and intuitive manner. While
a great deal of effort has been made to enhance the
underlying post-processing methods, the visualization
aspect has been less taken into consideration.
Traditional browsers which allow the user to follow
links in hypertext are unable to present the whole result
in the available space. A graphical user interface (2D
or 3D) helps users to see more results as a global view
and to discern their relation. Unlike flat 2D
representations, 3D visualization helps to present more
information in the enlarged space by adding it an extra
Computer Graphics, Imaging and Visualisation (CGIV 2007)
0-7695-2928-3/07 $25.00 © 2007