Transcript slides
Learning Where to Look: An
ACT-R/PM Model
Brian D. Ehret
ARCH Laboratory
Human Factors and Applied Cognition
George Mason University
ACT-R Workshop - August 7, 1999
Overview
HCI research has demonstrated that users learn
the locations of interface objects
However, not much known about mechanisms
underlying location learning
Systematically vary conditions under which
location learning may occur in order to
Infer characteristics of location learning mechanisms
Embed these characteristics into a computational
cognitive model using ACT-R/PM (Byrne & Anderson,
1998)
Color-Match Condition
Meaningful Text Condition
Arbitrary Icon Condition
Performance Time Results
16
14
16
Arbitrary
Color-Match
Meaningful
14
Trial Time (sec)
No-Label
12
12
10
10
8
8
6
6
4
4
2
2
0
0
1
2
3
4
5
6
7
8
9
Blocks
10
11
12
13
14
15
16
Overview of Model
35 rules for all four conditions
20% of rules common to all four conditions
Overlap between conditions ranges from
23% between Color-Match and Arbitrary to
86% between Arbitrary and No-Label
Interacts directly with software via ACT-R/PM
Relies on search vs. retrieve paradigm
Akin to compute vs. retrieve (arithmetic facts)
Preference for less costly strategy
Overview of Model (cont’d)
Declarative structures
Buttons
Locations
Labels
Colors
Learning Parameters
Base level learning (d=0.3)
Merging and retrieval
Associative strength learning (:al=1)
Source spread
Search Cost and Strategies
Search Cost - number of buttons evaluated per
trial
Search Cost - Buttons Evaluated
7 7
Arbitrary
Color-Match
6 6
Number of Buttons Evaluated
Meaningful
No-Label
5 5
4 4
3 3
2 2
1 1
0 0
1
2
3
4
5
6
7
8
9
Blocks
10
11
12
13
14
15
16
Search Phase Summary
Pre-attentive Search
Search for location of button with correct color
ACT-R/PM find-location command
Controlled Search
Attempt to retrieve a past use of the correct button and
its associated location chunk
If retrievals fail, attend to random button
Retrieval attempted first (search vs. retrieve)
Evaluation Cost and Strategies
Evaluation Cost - time required to determine if
currently attended button is currently needed
Evaluation Cost - Time Per Button
2.4 2.4
Arbitrary
2.2 2.2
Color-Match
Meaningful
Average Time Per Button (sec)
2 2
No-Label
1.8 1.8
1.6 1.6
1.4 1.4
1.2 1.2
1 1
.8 .8
.6 .6
.4 .4
1
2
3
4
5
6
7
8
Blocks
9
10
11
12
13
14
15
16
Evaluation Phase Summary
Color-Match and Meaningful always label
match
Arbitrary label matches only if label retrieved in
search phase
No-Label and Arbitrary (if label not retrieved)
try location assessment
Attempt to retrieve a past use of the currently
attended button and its associated location chunk
If retrievals fail, then wait for ToolTip
Retrieval attempted first (wait vs. retrieve)
Implications
Locations encoded as a by-product of attention
Default ACT-R/PM behavior
Location knowledge, once encoded into memory,
is like any other knowledge
Subject to the same learning mechanisms
Task performance characterized as rational
Attempt less costly strategies first
Pre-attentive < retrieve location < search
[search phase]
Label-match < location assessment < tip [evaluation phase]
Results in differential location learning