Next: The CARMA Data Viewer
Up: Enabling Technologies for Astronomy
Previous: The ALADIN Survey Integrator
Table of Contents -
Subject Index -
Author Index -
Search -
PS reprint -
PDF reprint
Lesteven, S., Poinçot, P., & Murtagh, F. 2001, in ASP Conf. Ser., Vol. 238, Astronomical Data Analysis Software and Systems X, eds. F. R. Harnden, Jr., F. A. Primini, & H. E. Payne (San Francisco: ASP), 78
Visual Exploration of Astronomical Documents
Soizick Lesteven, Philippe Poinçot, Fionn Murtagh1
CDS, Observatoire astronomique de Strasbourg, 11, rue de l'Université, 67000 Strasbourg, France
Abstract:
The CDS bibliographical map is a tool for organizing astronomical text
documents into a meaningful map for exploration and search. The system is
based on the Self Organizing Map (SOM) algorithm that automatically organizes
documents into a two-dimensional grid so that related documents appear close
to each other and general topics appear in well defined area.
After the determination of optimal parameters for the SOM's learning process,
we have developed a graphical WWW interface which allows the visualization of
the document distribution. It shows the localization of documents related to
given topics (keyword queries). The map is clickable and provide links to the
documents. Recent developments include detailed map of small areas, full text
indexing, automatic labeling, ...
Some applications will be presented. One map is available for interactive
use on the Web (http://simbad.u-strasbg.fr/A+A/map.pl).
The continually increasing quantity of textual data requires constant
effort in order to update storage and access methods so that
the totality of information is easily accessible. Scientific publications
are no exception. Astronomy is a good example in view of the enormous mass
of data collected by modern satellites and large ground-based facilities,
and the numerous scientific articles which result from such data.
The Centre de Données astronomique de Strasbourg (CDS)
collects and organises different types of astronomical information
(Genova et al. 2000). In particular, the CDS
offers on-line access to some bibliographic data.
This article presents ongoing work on the CDS bibliographical
maps. These maps are a tool for automatically organizing collections of
astronomical text documents and for displaying them in order to
facilitate the mining and retrieval of information (Poinçot 1998, 1999).
The map is clickable and provides links to the on-line documents. The system is
based on the Self-Organizing Map, or Kohonen Map (Kohonen 1995). In the
article we present new developments, including detailed map of small
areas, full text indexing, and automatic labeling.
The SOM is an algorithm used to visualize and interpret large high-dimensional
data sets. These maps are one of the artificial intelligence techniques, and more
precisely an unsupervised neural network. The topology of the Kohonen SOM
network is shown in Figure 1. This network contains two layers of
nodes: an
input layer, and a mapping (output) layer in the shape of a two-dimensional grid.
The input layer acts as a distribution layer. The number of nodes is equal to
the number of attributes associated with the input. Each node of the mapping
layer also has the same number of attributes. Thus,
the input layer and each node of the mapping layer can be represented as a
vector which contains the number of features of the input. The
mapping nodes are initialized with random numbers. Each actual input is
compared with each node on the mapping grid. The ``winning'' mapping node is
defined as that with the smallest Euclidean distance between the mapping node
vector and the input vector. The
value of the mapping node vector is then adjusted to reduce the Euclidean
distance. In addition, all of the neighboring nodes of the winning one are
adjusted proportionally. After all of the input is processed (usually after
tens of repeated presentations), the result should be a spatial organization
of the input data organized into clusters of similar regions.
Figure:
The topology of the Kohonen self-organizing map network.
 |
We applied the method to the classification of articles
published in Astronomy and Astrophysics in the period 1994 up to now
(9450 articles). The attributes were based on the bibliographic keywords.
We kept only the 269 keywords that appear in at least
five different articles.
The documents characterized in this way constitute a set of 9450
stimuli to be applied to the network. Learning through 50 iterations
(heuristically determined) gives good results for the principal map
(
units).
We have adapted the use of the SOM to our own needs.
Documents located at a map edge have neighbors at the other side of the map.
It is then possible to reconfigure the map without losing the similarity of
closely clustered documents.
At the end of the training of a map, the number of documents assigned
to each node is known. Because
it is much easier to visualize the colours of an image than a matrix of
numbers we transformed the table of numbers into an image. For this image the colour scale
indicates qualitatively the number of documents per node.
The map is manually labeled to locate on it the different themes
associated with the nodes. The user can select one node
of the map, by clicking on the picture, to obtain some information about the
articles located in it: the number of documents and the keywords describing
them appear on the right side of the interface. The user can also access the
article content (title, authors, abstract) and all the
facilities provided by the CDS bibliographical service (including a link to
ADS and to the on-line full paper when available). The map can also be
reached by keyword queries and bibcode queries (standard definition of a
document).
One map is available for interactive use on the World-Wide Web
(http://simbad.u-strasbg.fr/A+A/map.pl).
The size of the SOM map has a strong influence on the quality of the
classification. The smaller the number of nodes in the network, the lower
the resolution for the classification. The fraction of documents assigned
to each node correspondingly increases. It can then become difficult for the
user to examine the contents of each node when the
node is linked to an overly long list of documents. However, there is a
practical limit to the number of nodes: a large number means long training.
A compromise has to be found. That is why, for each ``over-populated''
node of this map, termed the principal map, another network,
termed secondary network or map, is created and linked to the
principal map. Each secondary network is trained using the documents
associated with the corresponding node and its surrounding nodes of the
principal map (Figure 2).
In this way, a map is created with as many nodes as necessary, while
keeping the computational requirement under control.
Figure 2:
The secondary maps.
 |
To overcome the limitations due to the representation of the documents by
keywords, we use all textual information present in the documents
(title, abstracts, keywords, ...).
We built our own full text indexing tool. All the words are kept and counted,
empty words (without meaning) and the less frequent ones are
eliminated. A truncature (Porter 1980) is applied to
reduce morphologic variants of a word to a single index term.
The program allows to keep word associations (2 or 3 words).
The first applications show that the two-word associations are meaningful.
It is now possible to create bibliographical maps with a full text indexing
and to take into account documents for which keywords are not available.
It seems that the full text indexing is on one hand noisier due to the
automatic indexing
(the context is not taken into consideration) but on the other hand
more precise because it does not rely on a limited list of keywords.
For map interpretability, the different themes associated with
document/node assignments have to be indicated. Although our maps
have a relatively limited number of units, it is important to automate
the annotations.
Because it is impossible to characterize all nodes without overlapping
annotations it is preferable to select a limited number of nodes for
characterization. The selection is made according the density peaks and
the cross-positions of two peaks.
When the position is defined, the process examines the words associated with
the documents assigned to the peak and writes the most frequent one beside
the peak. When different words have the same occurrence, the automatic choice
(the first one) is not automatically the best one.
Furthermore, our system makes it possible to classify the words in various
categories. It is then possible to annotate the map according to the different
classes, corresponding to a precise application.
We developed a visual and efficient system to explore astronomical documents
based on the SOM.
Other data collections, notably catalogue information, have already been
processed in the same way, and a cartographic user interface tool has been
set up to allow catalog selection (http://vizier.u-strasbg.fr/).
Now that new tools have been developed, a number of new applications are
possible. Creation of a bibliographical maps for the SIMBAD objects,
for an ADS query, for a colloquium organisation, ...
References
Genova, F., et al. 2000, A&AS, 143, 1
Kohonen, T. 1995, Self Organizing Maps (Berlin: Springer-Verlag)
Poinçot, P., et al. 1998, A&AS, 130, 183
Poinçot, P., 1999, PhD Thesis ``Classification et recherche d'information bibliographique par l'utilisation des cartes auto-organisatrices, applications en astronomie''
Porter M. F. 1980, Program, 14(3), 130
Footnotes
- ... Murtagh1
- School of Computer Science, Queen's University Belfast, Belfast BT7 1NN, Northern Ireland
© Copyright 2001 Astronomical Society of the Pacific, 390 Ashton Avenue, San Francisco, California 94112, USA
Next: The CARMA Data Viewer
Up: Enabling Technologies for Astronomy
Previous: The ALADIN Survey Integrator
Table of Contents -
Subject Index -
Author Index -
Search -
PS reprint -
PDF reprint
adass-editors@head-cfa.harvard.edu