Transcript Introduction to HCI - Department of Computer and Information

Information Search and
Visualization
Human Computer Interaction
CIS 6930/4930
Section 4188/4186
Intro
How can we design interfaces to search through large amounts of data?
► We’ll look at different approaches to sift through information
► Old approach: Information Retrieval
► New approaches:
►





►
►
►
Information Gathering
Seeking
Filtering
Visualization
Data mining and warehousing
Difficulty increases with data volume and diversity
Ex. Find a news story, find a picture
How can we design an interface for
 New users (how do I express what I want? boolean operators are not that easy to
use)
 Experienced users (powerful search methods)
►
►
Use research from:
Perceptual psychologists, statisticians, graphics designers
Searching Textual Documents and
Databases
► Most
widely used (and understood)
► Web
searches relevance still needs work
► To satisfy both users
 create two interfaces (advanced and basic)
 Multilayer interface
► User
satisfaction increases with more search
control (Koenemann ’96)
► Clustering into meaningful hierarchies might be
effective (Dumais ’01)
Multimedia Searches
Currently: Requires metadata (captions, keywords, properties)
► Query by Image Content (QBIC) – Find pictures of the Florida Gators Football
team (w/o using descriptors, webpage info, etc.)
►
 Approaches:
►
►
►
►
Search for distinctive features
Give example images
Image spaces
Major research area
 Best: restrict database if possible (like medical, etc.)
►
Map Search
 Easy: search by lat and long
 Harder: search by features (find all cities near a seaport and a moutain > 10000 ft)
 App: Find businesses for mobile GPS systems
►
Design or diagram search
 CAD models, engineering apps
 Ex: Find 6 cylinder engine designs with pistons > 6 cm
 Some basic structured document searching for things like newspapers and magazine
layouts
Multimedia Searchse
►
Sound Search
 Music-information retrieval (MIR)
 New approaches:
► Query
with musical content (Hu ’02)
► Query by recognized patterns like singers
► Using speech recognition and TTS as inputs to audio databases
►
Video Search
 Only preliminary research on this topic
► Infomedia
(screenshot) uses visual features + text (TTS) for esarching
 Currently: show timeline to allow quick browsing of contents
►
Animation Search
 Untapped, but growing need
 Might be easier with standard definitions like Flash
Advanced Filtering and Searching
Interfaces
►
►
Alternatives to form-fillin
Filter with complex Boolean queries



Research into how we can make them easier to specify
Difficulty is in the colloquial use of English (all classes in weil and NEB, or I’ll take ketchup or mustard)
Novel metaphor approaches (doesn’t scale well)
►
►
►
►
►
Automatic Filtering
►

Aesthetic Computing (screenshot)



Users create rules for data
Ex. Email filters, news stories filters
Similar to : Selective Dissemination of information (SDI)





Adjusting numerical sliders (www.bluenile.com)
Appealing and easy to understand
a.k.a. Direct-manipulation queries (objects, [rapid, reversable, immediate] actions, feedback)
Reduces errors and encourages exploration
Large databases can give previews given the user defined ranges (Fig. 14.8)
Dynamic Queries
►
►
Venn diagrams
Decision Tables
Water through filters
Although having such large ‘hits’ might seem poor, (Tanin ’00) showed 1.6 to 2.1 performance and satisfaction increase
Faceted metadata search


Combine category browsing with dynamic previews (Yee ’03)
Search on a topic (car price), then restrict on feature (car year), then on # of doors, then widen on all years



Groups of users to combine evaluations to find interesting results
Amazon.com’s lists or “other people who bought this item also bought…”
Good for organizational databases, news files, music, shopping

Research areas to use perhaps restricted domain-specific translation dictionaries (like medical ones)




Use visual representations like maps instead of text lists to select and refine searches
Trees to represent product catalogs
Very powerful. (Also the calendar and plane layout methods)
User error reduced
Collaborative Filtering
►
Multilingual Searches
►
Visual Searches
Information Visualization
►
►
►
►
Visualize information data in novel methods to amplify
cognition (Card ’99, others)
Different the scientific viz because of the abstract nature
Goal: Present compact graphics representations and user
interface for manipulating large # (or subset) of items
Visual data mining




Apply visual bandwidth (which is very great) and human perception
Make discoveries, decisions, hypothesis
Underutilized in most interfaces
Humans are good at:
► Detecting patterns
► Recall images
► Detect
subtle changes in size, color, shape, texture
 Research: new dynamic info viz. Go beyond icons and illustrations
Info Viz
►
►
►
►
Provide useful tools to allow users to find trends in data
Go beyond novelty and address true business concerns
Share insights easily with others
Some user resistance (esp. if text really is better!)
 Solution: Measure benefits
►
Visual information mantra:
 Overview first, zoom and filter, then details on demand
►
Figure out data type and task, then look at different
current methods to visual them. Box 14.2 (pg 583)
Data Types
►
1D Linear –





►
Ex: source code, text, audio
Info Viz approaches: zoom, color coding
Consider: Layout, color, size, overview approach
Tasks: Find # of items, changes
Apps: Document Lens, SeeSoft, Info Mural
2D Map - Planar data
 Ex: GIS, floorplans, newspaper layouts
 Approaches: Multilayer (each layer is 2D), spatial displays
 Consider: Data (name, owner, value) and interface (size, color,
opacity) attributes
 Tasks: Find adjacent items, regions, paths
 Apps: GIS, ArcInfo, ThemeView
 screen shots
Data Types
► 3D
World – more than just geometry
 Ex: 3d molecules, body, buildings
 Approaches: landmarks, overviews, multiple views,
tangible UI
 Consider: both geometry and relationships, navigation
can be difficult for many
 Tasks: focus on meta-relationship patterns
 Apps: Medical imaging, CAM, chem structure, scientific
sims, flythroughs
 Making things 3d that aren’t or don’t fit well, doesn’t
make the results better, could hamper performance
 Screen shots
Data Types
► Multidimensional
data