Architecture of Dynamic Query User Interface for Networked Information Systems
Khoa Doan
Catherine Plaisant
Ben Shneiderman
Human Computer Interaction Laboratory
A.V Williams Buildings
University of Maryland
College Park, MD 20742
Email: {doan,plaisant,ben}@cs.umd.edu
phone: (301) 405 2725
fax: (301) 405 6707
For the past several years, research at the Human-Computer Interaction
Laboratory (HCIL) has focused on creating dynamic query interfaces that
apply the principles of direct manipulation to the database environment:
o visual representation of the query's components;
o visual representation of results;
o rapid, incremental, and reversible control of the query;
o selection by pointing, not typing; and
o immediate and continuous feedback.
Dynamic queries involve the interactive control by a user of visual query
parameters that generate a rapid (100 ms update), animated, visual display
of database search results.
In a networked information system such as the EOSDIS (Earth Observing
System -- Data and Information System), there are three major obstacles in
supporting dynamic queries for searching datasets in data directories:
o Network Performance: Dynamic queries requires rapid performance (100
queries or more returned per second). But querying different Data
Archive Centers via the network is too slow to support dynamic
queries.
o Data Volume: Dynamic query interfaces support loading data points
into memory, and the queries are performed locally. Our existing
dynamic query interfaces only handle up to a few thousands data
points while in the context of the EOSDIS, the data volume can reach
hundred thousands data points.
o Data Complexity: In a dynamic query interface, query parameters are
visually represented in a single screen. The large number of
attributes of the EOS dataset poses an interface challenge in
displaying it in a single screen.
In order to resolve the above problems in developing a dynamic query user
interface for the EOSDIS, a two-step approach using volume predictors is
proposed:
o Initial Search: An initial search for data sets is performed in the
Data Volume Predictor Dynamic Query Interface . The user starts by
selecting a few attributes of the data sets (eg. parameter, sensor,
time and location ). The volume predictor displays graphically the
estimated total number of the data sets for all values of the
selected attributes in the form of the charts or the shaded maps (eg.
the time chart would show that there are much more data sets in the
recent years than ten years ago). All charts are visually updated
when users indicate a rough range of values of the attributes (eg.
when selecting humidity for the parameter, hygrometer for the sensor,
the updated location chart would show that there are more data sets
in the South East Asia region than elsewhere in the world). These
graphical results give users a rough indication of the results of the
search before an actual search is formulated and submitted to the
Data Archives. The initial search therefore aids users to rapidly
eliminate the bulk of undesired data sets, and focus on a manageable
number of data sets in a dynamic manner. The initial search relies on
the regular publishing by the Data Archives of estimates which are
merged in a volume predictor matrix. The volume predictor is thus
updated outside the interaction session. In addition, the volume
predictor is local, therefore rapid dynamic queries can be achieved
in the Data Volume Predictor Dynamic Query Interface . When the
estimated number of the data sets is low enough, the initial search
is then submitted to the Data Archives, which returns the metadata of
the data sets for browsing and further refinement in the Refined
Search step.
o Refined Search: Data sets metadata which are returned from the
Initial Search, are stored in a local database. The search is then
further refined by adjusting the sliders, pressing buttons or using
the pointing devices (eg. a mouse or a trackball) to select more
precise attribute values of the data sets in the Data Browser Dynamic
Query Interface for viewing and ordering. With this dynamic query
approach, users are not only able to refine the search dynamically,
but can possibly discover data set patterns and exceptions.
The above two-step approach aims to overcome the network performance, and
breaks down the data volume and data complexity in the EOSDIS in order to
facilitate dynamic queries. This method is named the dynamic query
extraction method via the use of volume predictor , and its architecture
is illustrated in the figure below:
[Image: http://www.cs.umd.edu/~doan/abstract_arch.gif]
The dynamic query extraction method via the use of volume predictor
supports both novice and expert users with diverse technical backgrounds.
For novices, dynamic query approach can help them formulate queries, and
present results graphically. Expert users are able to formulate more
complex queries and interpret intricate results in a dynamic manner. In
conclusion, the dynamic query extraction method via the use of volume
predictor in the EOSDIS demonstrates how dynamic queries can be used in a
networked environment.