(based on Shneiderman Chapter 15)

Introduction.

Information search goes by many names:

• information retrieval
• database management
• information gathering/seeking/filtering/visualization
• data mining/warehousing

• Use of complex query language (e.g. SQL)
• Adv: yields best results
• Dis: knowledge/experience required to form queries
• Use of natural query language (which is translated internally into SQL)
• Adv: easy for novice to form query
• Dis: the query may lose something in translation, and yield poor/confusing results

In either case, user mental focus is pulled away from task and drawn to the tool!

• Productivity declines.
• User may feel loss of control.

Loss of control also a problem with WWW searches : user does not know criteria for ranking matches! (VERY proprietary -- should it be??)

Novices need DMI that combines expressiveness with ELU (ease-of-learning-and-use)

Experts need suite of query tools

OAI can help: ELU requires intuitive translation (metaphor) from task OA to interface OA

Example of the type of translation problems that have to be solved: use of boolean logic

• Boolean AND may be difficult for novice to learn.
• Natural language use of AND: expands the scope (John and Paul and George and Ringo)
• Boolean AND: restricts the scope (search for user and interface and design)
• Similar problems with boolean OR.

Good filtering metaphor is needed.

Shneiderman suggests a four phase framework for designing search tool:

1. formulation : ability of user to express search. Let user specify phrases, multiple phrases, limitation of search space to certain fields (such as title), opening search up to variants/partial matches/stems/synonyms/abbreviations/thesaurus of specified words.
2. initiation of action : usually explicit (button) but may be implicitly triggered by change to search criteria or similar manipulation of objects (dynamic query).
3. review of results : users may be given control over how many results are presented, which fields to display, the order in which results are presented.
4. refinement : support for user ability to refine an unsatisfactory search. E.g. keep a history of searches and results for later consultation; allow easy search criteria modification.

• Photos: Consider query by image content (QBIC) : Current technology limits such searches to very restricted data collections (such as fingerprints).
• Maps: Some mapping systems support searching based on features (not lattitude/longitude).
• Designs: Some CAD/CAM systems support searching based on design/diagram features. This is simplified if document being searched is formatted as codes rather than image.
• Spoken audio: Searches specified by speaking then translate into waveform. Speaker-independence difficult to achieve.
• Music audio: search based on score (rather than music waveforms) due to signal complexity.
• Video: same problems as photos. Can compare adjacent frames for frame-to-frame structural changes such as scene changes.

• Provide orientation and context
• Provide dynamic feedback
• Support selection based on location (rather than name)
• Reveal data patterns (clusters, gaps, outliers)

Quote from Shneiderman (p 522) : "Overall, the bandwidth of information presentation is potentialy higher in the visual domain than it is for media reaching any of the other senses. Humans have remarkable perceptual abilities that are greatly underutilized in current designs."

Shneiderman visual-information-seeking mantra: "Overview first, zoom and filter, then details on demand." This should be basic principle for designing info visualization systems.

Seven basic tasks:

• overview (visualize entire data set, possibly with field-of-view box),
• zoom (control zoom factor and focus),
• filter (remove uninteresting items, much as is done with search query specifications),
• details-on-demand (select item, browse its displayed attributes),
• relate (methods for finding, exploring, displaying relationships),
• history (for undo/redo/refinement),
• extract (save items or filter settings).

Data may be :

• 1-D spacial (streamed data such as unix file)
• 2-D spacial (newspaper layout)
• 3-D spacial (car; any real-world object with volume)
• n-D (arbitrary number of attributes)
• temporal (has start and finish time)
• tree relationships (visualize using nodes/edges, indented lists, etc)
• network relationships (visualize using nodes/edges, matrix)

Last reviewed: 9 December 1998