Transcript bottom time

User Characteristics
& Application design
Introduction
• The starting point of this course:
– Users want to understand information about
something to make an informed decision
• So far we learnt techniques that act as
means to support users in comprehending
information
– Data mining (DM) and
– Information visualization (InfoVis) techniques
• Real world applications require integration
of these two technologies
• This is one of the grand challenges in
informatics
– No easy and readymade solutions
2
Integrating Data Mining and InfoVis
• High level architecture
Input
Data
Data Mining
Information
Visualization
End-User
Interaction
• Data Mining (DM)
– Compute patterns or models (in general abstractions) from
raw input data
• Information Visualization (infovis)
– Present the relevant abstractions (patterns or models) in a
form suitable to the end user
– Support user interaction
• Integrating Data Mining and InfoVis is the main goal
of this course
• Two Options for integration
– Option 1 - Loose Coupling
– Option 2 – Theory Driven
3
Loose Coupling
• Separate libraries of data mining and infovis are
offered to the user
• User is given freedom in exploiting the available
methods to understand data
• Certain constraints may be defined in linking a
specific data mining method with a specific infovis
method
• Already available in many existing tools
– In Excel and Weka to lesser extent
– R, SPSS and other statistical packages
• Unlimited growth of the libraries
– Because there is no notion of what is required
• Users may use only small portions of the tools
– Using sledge hammer for domestic work
• Suitable only to expert users
– A few experts gain monopoly over critical data owned by all
4
Theory driven
• A general Human Computer Interaction (HCI)
theory is required that focuses on
– Systematic linking of application design to users, their
tasks and task contexts
– Taking into account human perception, comprehension and
reasoning of information
• Visual Analytics (VA) is a new discipline that aims
to develop such a theory
– VA currently focuses on grand applications
• intelligence analysis
• genomic data analysis
• Simpler versions of VA (?) should be developed for
simpler applications
– E.g. Scuba dive computer
• HCE, TimeSearcher and GIS studied in this course
integrate DM and InfoVis
5
HCE and TimeSearcher
• HCE incorporates sense of statistical analysis into
an InfoVis tool
– GRID (Graphics, ranking and interaction for discovery)
principle is the simple theory here
– (refer to lecture 7)
• TimeSearcher adds a InfoVis driven front end to
time series similarity matching
– Visual query tools such as Timeboxes form the simple
theory here
• Both tools are general purpose
• How to design simple applications in specific
domains with
– Real users and
– Real tasks
6
HCI Approach to user
characterization
• Users vary along several dimensions
–
–
–
–
–
Age (e.g children vs adults)
Personality (e.g. extrovert vs introvert)
Physical disabilities (e.g visual impairments)
Skills (e.g. expert vs novice)
Etc
• Identify user groups among the complete set of
users
– Subsets of users with similar characteristics
• Identify the tasks (goals) of user groups
• Use this information to drive system design
7
Configurable Systems
• Designing systems for different user
groups comes under the study of
‘Accessibility’
• This is an important topic in itself
– Unix Vs Windows for sighted Vs visually
impaired users
• The main principle behind improving
accessibility is to allow system
configuration
– A configurable system is an accessible system
• We follow this principle in application
design
– But what features should be available for
configuration?
8
Implications of user characterization
• System design in our case involves mainly
designing two components
– Data mining
– Visualization
• Designing data mining component involves
– Collecting the right data (both attributes and
instances)
– Selecting an appropriate data mining task
– Selecting an appropriate data mining method to
achieve the task
• representation language
• search method
• pruning method
9
Implications of user characterization
(2)
• Role of user tasks in designing data mining
component
– User tasks (goals) help collecting the right data
• E.g. For judging the safety of a dive, you need data about dive
depth, duration and rapid ascents
– User tasks determine the appropriate data mining task(s)
• E.g. Cluster together dives with similar characteristics
• Applications usually require multiple data mining
methods
– Because performance of data mining methods varies widely
• User group characteristics determine the level of
configurability offered to the users
– E.g. expert Vs novice
10
Implications of user characterization
(3)
• Visualization techniques present information using a
suitable visualization
• What is a suitable visualization?
– Visualization is suitable if it enables users (with their
characteristics) to understand the presented information
– E.g. a learner scuba diver with poor graph reading skills
might need visualizations that clearly mark dive depth and
bottom time
• Design of visualization involves
– Choosing a visualization technique
– Mapping data features to graphical features
• The visualization technique used can vary with user
characteristics
– E.g. a doctor inspecting scuba dive data may like to view
tissue saturation values and model predicted micro-bubble
data
11
Implications of user characterization
(4)
• The mapping scheme used in the visualization
can vary with user characteristics
– E.g. for a user with red-green colour blindness to
avoid using red for marking rapid ascent patterns
on a green dive profile line graph
• User tasks too have some influence on design
of visualization
– E.g. a researcher on diving safety requires
visualizations that are lot more technical than a
regular scuba diver
12
Practical problem with the HCI
Approach
• Acquiring knowledge of user
characteristics and user tasks is not easy
• HCI recommends two approaches
– explicit characterization – e.g. asking users
directly for user characteristics
• But users do not always know the required
information
– Implicit characterization – e.g. start with no
explicit user information (cold start) but infer
user characteristics from observable user
behaviour
• But user behaviour is not always rational
13
Our Approach
• Experts know the implications of user
groups and their tasks
• One practical solution is to allow domain
experts who regularly deal with different
user groups and tasks to configure the
system for different users
– E.g, a weather forecaster may know how to
analyse and present weather forecast
information for more technically oriented oilrig
staff
14
ScubaText
• ScubaText project analyses scuba dive computer
data and presents the results of analysis
– Graphically – annotated graph
– Textually – summary of safety related information
• It is assumed that learner divers may find the
textual descriptions and their links to graphical
displays useful for judging the safety of a dive
– User group – learner divers
– User task – judging the safety of a dive
• Based on user characterization (learner diver)
textual descriptions are included in the
presentation
• Real user evaluation showed that the simple user
model did not work!
15
Text+Annotated Graphics (D)
Depth-Time Profile
Surface
00
'2
0
01 "
'4
0
03 "
'0
0
04 "
'2
0
05 "
'4
0
07 "
'0
0
08 "
'2
0
09 "
'4
0
11 "
'0
0
12 "
'2
0
13 "
'4
0
15 "
'0
0
16 "
'2
0
17 "
'4
0
19 "
'0
0
20 "
'2
0
21 "
'4
0
23 "
'0
0
24 "
'2
0
25 "
'4
0
27 "
'0
0
28 "
'2
0
29 "
'4
0
31 "
'0
0
32 "
'2
0
33 "
'4
0
35 "
'0
0
36 "
'2
0
37 "
'4
0
39 "
'0
0
40 "
'2
0
41 "
'4
0
43 "
'0
0
44 "
'2
0
45 "
'4
0
47 "
'0
0"
0
-5
-10
-15
85% MaximumDepth
MaximumDepth
Depth
-20
-25
-30
-35
Bottom Zone
-40
-45
-50
A
A
Bottom Time
Time
Risky dive with some minor problems. Because your bottom time of 12.0min
exceeds no-stop limit by 4.0min this dive is risky. But you performed the ascent
well. Your buoyancy control in the bottom zone was poor as indicated by ‘saw
tooth’ patterns marked ‘A’ on the depth-time profile.
16
Revised Text
One of the subjects revised the output text as follows:
Potentially risky dive with some minor problems. The bottom
time of 12.0min exceeds no-stop limit by 4.0min requiring
mandatory decompression stops. The ascent was at a
constant rate within the recommended rate. The saw tooth
patterns marked ‘A’ on the depth-time profile should be
avoided if possible as this increases the chance of
developing DCI even within the recommended decompression
limits. The re-descent from 5m to 10m in the later stages
of the dive should also be avoided for the same reason as
saw-tooth profiles.
• Revisions mostly aimed at neutralising the emotional content
– But doctors who regularly treat divers with DCI prefer text
with emotional content because of the direct impact it can have
17
Discussion
• To communicate the safety message
effectively
– Good understanding of user personality
required
• In the department we work on
Affective NLG
– Generating emotionally appropriate text
• Elsewhere in NLG, emotional issues
such as politeness are explored
18