Mapping Associations - Mapping Scientific Frontiers: The Quest for Knowledge Visualization - page 91

3.1.3 Information Visualization

Information visualization emerged as a field of study since the 1990s. There has
been a widely spread interest across research institutions and the commercial
market. Applications of information visualization range from dynamic maps of the
stock market to the latest visualization-empowered patent analysis laboratories. It is
one of the most active research areas that can bring technical advances into a new
generation of science mapping.
The goal of information visualization is to reveal invisible patterns from abstract
data. Information visualization is to bring new insights to people, not merely pretty
pictures. The greatest challenge is to capture something abstract and invisible with
something concrete, tangible, and visually meaningful. The design of an effective
information visualization system is more of an art than science. Two fundamental
components of information visualization are structuring and displaying.

3.2 Identifying Structures

The purpose of structural modeling is to characterize underlying relationships
and structures. Commonly used structural models are lists, trees, and networks.
These structures are often used to describe complex phenomena. Ben Shneider-
man proposed a task-by-data-type taxonomy to divide information visualization
(Shneiderman 1996 ).
Networks represent a wide spectrum of phenomenon in the conceptual world as
well as a real world. For example, the Web is a network of web pages connected
by hypertext reference links. Scientific literature forms another network of articles
published in journals and conference proceedings. Articles are connected through
bibliographic citations. A set of images can be regarded as a network based on
visual attributes such as color, texture, layout, and shape. In content-based image
retrieval (CBIR), the emphasis is on the ability of feature extraction algorithms to
measure the similarity between two images based on a given type of feature.

3.2.1 Topic Models

In information retrieval, it is common to deal with a set of documents, or a
collection, and to study how the collection responds to specific queries. The
similarity between a query and a document, indeed, a document and another
document, can be determined by an information retrieval model, for example, the
vector space model, the latent semantic indexing model, or the probabilistic model.
These models typically derive term-document and document-document matrices,
which are in turn equivalent to network representations. The vector space model