|
Information management - Topic Maps visualization
|
 |
 |  | Find |
|
Abstract
|
Topic maps - the new ISO standard ISO-IEC 13250 - provide a bridge between
the domains of knowledge representation and information management. Topics
and topic associations build a structured semantic link network above information
resources. Our research aims at visualizing this semantic layer efficiently,
which is a critical issue as topic maps may contain millions of topics and
associations. This paper is divided into 3 parts. First, we depict briefly
basic topic maps concepts. Then, we review a few graph visualization techniques.
Finally, we describe the visualization tool we developed at the Laboratoire
d'Informatique de Paris 6 and study how this tool may be used - and enhanced
- for topic maps visualization.
|
Keywords
|
|
Contents
|
Introduction: basic topic maps concepts
According to [Pepper 99], topic maps - the new ISO
standard ISO-IEC 13250 [ISO 99] - will become the answer
for organizing and navigating through large and continuously growing information
pools. They provide a "bridge" between the domains of knowledge representation
and information management. This standard defines both an abstract data model
and a serialization syntax to represent knowledge structures and to link them
to information resources.
Figure 1 [Pepper 99] describes
topic maps basic concepts: topics, occurrences of topics and relationships
(associations) between topics.
For example, consider an information pool consisting of conference material.
These resources have different roles (occurrence roles) - articles, videos,
charts, call for participation, etc. Examples of topics are XML Europe, Paris,
France. Topics also have different types - conference, city, country, etc.
- and associations exist between them: XLM Europe takes places in Paris, Paris
is in France. In this example, there are two types of associations - "takes
place in" and "is in".
Figure 1
. Topic maps basic concepts
As shown in this diagram, a topic map is divided into a topic domain
(consisting of topics and associations) and a resource domain.
Topics and topic associations build a structured semantic link network
above information resources (topic occurrences). This network allows an easy
and selective navigation. However, if topic maps are very large, users may
have problems to understand them and find relevant information. Thus, it is
necessary to represent them efficiently.
In this paper we will focus on the visualization of the "portable semantic
network" and disregard the information resources layer. We thus need to represent
topics and topics associations. In the future, we will also represent occurrences
(i.e. the links between individual topics and the occurrences of information
about those topics).
The portable semantic network can be viewed as a graph in which topics
are nodes (vertices) and associations are arcs (edges), both of which are
typed. Visualizing this semantic layer may be a critical issue as real topic
maps will consist of millions of topics and associations.Therefore, it is
interesting to study how existing large graphs visualization techniques may
be applicable to topic maps.
Previous
Table of Contents
Existing graph visualization techniques
NicheWorks [Wills 99] is a 2D interactive visualization
tool for the investigation of very large graphs that cannot be represented
on one static display. NicheWorks allow users to examine a variety of nodes
and edges attributes in conjunction with their connectivity information. Parts
of a graph may be shown or hidden using interactive manipulation of views
of node and edge attributes; nodes and edges have different colors and shapes
according to their attributes.
Figure 2 and
Figure 3
are examples of NicheWorks visualizations.
Figure 2
. Example of graph visualization with NicheWorks
Figure 3
. Node appearance details
MAGE software [Freeman 98] uses color, three dimensional
representations and animation to help users see their data in different ways
and develop new insight about their data.
In the Laboratoire d'Informatique de Paris 6, we developed a 3D visualization
tool for large information hierarchies which are very difficult to use and
represent. This tool is fitted to any type of hierarchical data; as explained
in [Desclefs 99], we used XML [XML 98]
and the DOM [DOM 98] to design a "universal visualization
tool". However these structures may not be completely hierarchical: there
can also be non-hierarchical links, called "cross connections". We decided
to investigate its use for topic maps visualization.
Our tool speeds access to relevant information and helps users find
their way within the hierarchy. Fundamental factors for a good visualization
interface are ( [Kurnar]):
- An overview of the structure for a global understanding of the structure and of the relationships within the hierarchy,
- The ability to zoom and to select some nodes,
- Dynamic requests in order to filter data in real time.
Two visualizations are provided: a traditional 2D view (
Figure 4)
so as not to confuse users and a 3D view of the whole structure (
Figure 5).
3D allows a more efficient use of screen space. In particular, links between
nodes do not intersect. We used cone trees [Robertson 91] as
3D representation model. Nodes are spheres, cones or cylinders whose color
changes according to their level in the hierarchy. Consistency between 2D
and 3D views is achieved: when a node is selected in the global view, the
corresponding line in the 2D tree is highlighted and node's properties are
displayed - name, attributes and so on.
These visualizations are highly interactive; interesting nodes can be
put in the foreground with zooms, translations and rotations. Users can delete
irrelevant branches of the tree and expand interesting ones. They can also
select specific elements and display them in separate 3D windows called detailed
views.
However, these representation and navigation methods cannot solve the
whole problem: there is still a large amount of information to display. Therefore,
the initial hierarchy must be pruned with filtering and aggregating techniques.
This tool was used to supervize telecommunication network equipment. Two filters
were added to answer network managers' needs. They are particularly interested
in elements where an alarm was activated; they need to know the consequences
of these problems on connected ports. "Alarmed" nodes are stressed on the
visualization and connections starting from these elements can be displayed,
as shown on
Figure 5. Different colors are used according
to alarm types.
Figure 4
. 2D tree and XML source file
Figure 5
. Global view with cross-connections visualization
Application to topic maps
We decided to use the tool we developed at LIP6 to visualize topic maps.
In the following we will describe what modifications have to be made so as
to adapt our tool to topic maps visualization.
Filtering techniques
Topic maps may contain millions of topics and associations. Therefore
it is essential to select relevant information as it is impossible to display
the whole data efficiently. Filtering techniques are needed in order to select
and display only relevant information. Our tool enables users to filter topics
and associations according to their name and/or type, provided that name and
type are implemented as XML attributes, as explained in [Desclefs 99].
In the future, we will enhance our tool so that it can handle scope:
users will specify which themes they are interested in and the tool will filter
names, associations, etc. on that basis. This can be done with ontologies [Guarino 95]. Ontologies are widely used instruments for knowledge
sharing and reuse; they may be applied to specify domain knowledge in a generic
and consensual way. The key ingredients that make up an ontology are a vocabulary
of basic terms. Each concept is associated to a term (i.e. a symbol) used
to designate the concept, a description in natural language and a formal specification
in an appropriate language such as KIF (Knowledge Interchange Format) [Genesereth 92]. Terms are stored in dictionary-like structure. The
dictionary is open-ended to allow the addition of new concepts.
Let us compare topic maps with geographical maps. You will never find
a map of a country with the whole information about the country on it. There
will be a topographical map, a political map, an economic map, etc. In the
same way, topics and associations can be classified into different ontologies
and different topic maps will be provided to the user according to his interest.
If the user is interested in theatre, relevant topics are "play", "author",
"tragedy", "culture", etc. This is a way of filtering information according
to a specific scope.
Topics and associations representation
Once topics and associations are filtered, they need to be represented
efficiently.
Topics
Topics are nodes and their type may be symbolized by different colors,
shapes and textures. However, the number of different shapes, colors and icons
is limited. Class hierarchies can be used to reduce topic types to a small
number of "super-types", as stated in [Pepper 99]. In this
case, we only need a specific shape and/or color for each super-type. Consider
the following topic types: "artist", "painter" and "poet" ; they will look
alike because they all derive from the super-type "person". In the same way,
"was created by" and "was composed by" association types derive from "was
caused by".
Currently the topic map standard does not define a standardized way
of specifying type hierarchies. However this could be done in our tool - at
the application level - by using a kind of stylesheet mechanism. This would
allow users to specify which association types represent the supertype / subtype
relationship in a particular topic map. Nevertheless, for navigation purpose
in a topic map, a user may want to visualise hierarchies that have not been
specified by the designer of the map.
We suggest another way of reducing the number of types to represent,
by aggregating topics and associations with a classification algorithm. Galois
lattices [Godin 95] can group objects that share common
properties automatically. These groups are called classes. Therefore we only
need to distinguish classes of topics in the representation instead of differentiating
all topics. If we consider the example of topic map in
Figure 1,
"Paris", "Ile de France" and "France" will be represented the same way because
they belong to the same class "location". In the same way, associations "is
in" and "takes place in" are both "information about location". This classification
mechanism makes it possible to display topic maps with different levels of
details, as shown in
Figure 6.
Figure 6
. Level of detail in topic maps
Of course information displayed is less precise but this is acceptable
for navigation purposes. However it is possible to display more precise information
when users focus on specific parts of the topic maps. For example, if the
mouse cursor is on "Paris", textual information appears on the screen that
states Paris is a city (which is more precise than "location").
Associations
As far as associations are concerned, two solutions are possible: they
may be represented as arcs or nodes.
In the first case associations are arcs and their type may be symbolized
by the style of lines (full line, dotted line, etc.). However, this representation
is limited to binary relationships, which means that n-ary relationships have
to be decomposed into several binary relationships. This is possible but might
add too much complexity to the resulting visualization.
On the other hand, associations can be represented the same way as topic
- as nodes. More complex associations can be represented, in particular associations
that involve more than two topics. In this case, associations types are distinguished
the same way as topic types, as described in part 3.2.1.
A mouse-over event can display topics and associations names and types
when the cursor is positioned over these elements.
Figure 7 is an example of topic maps visualization
with our tool, in which associations are represented by nodes. Topics and
associations are symbolized by different shapes according to their type. We
use three different shapes in the current version of our prototype - cones,
cylinders and spheres - but more are possible.
Figure 7
. Example of topic map visualization. Associations
= nodes
In
Figure 8, associations are symbolized by arcs
between topics.
Figure 8
. Example of topic map visualization. Associations
= arcs
Conclusions and future work
In this paper, we investigated how the visualization tool we developed
at LIP6 may be used for topic maps visualization. This prototype provides
3D interactive visualizations. Topic types can be distinguished with different
colors and shapes, but we will add class hierarchies functionalities to reduce
the number of types to represent. In the future, we will also handle scope,
so as to allow users to specify their need for information more precisely.
Acknowledgements
We would like to thank Steve Pepper for his great help. He originated
our interest in topic maps and his ideas and comments allowed us to complete
this research.
Bibliography