Cornell University Library Metadata Working Group Forum May 20, 2016

1. Assessing Subject
Access for Images
HANNAH MARSHALL
METADATA LIBRARIAN FOR IMAGE COLLECTIONS
MWG MAY 20, 2016

2. Outline
Background
Research questions
Research design
Findings
Applications
Conclusions

3. The Arts & Sciences Images for
Teaching Collection

4. Challenges of Subject Analysis for
Images
"Image indexing is a complex socio-cognitive process that involves
processing sensory input through classifying, abstracting, and mapping
sensory data into concepts and entities often expressed through
socially-defined and culturally-justified linguistic labels and identifiers"
(Heidorn, 1999)
"Concept-based indexing has the advantage of providing higher-level
analysis of the image content but is expensive to implement and suffers
from a lack of inter-indexer consistency due to the subjective nature of
image interpretation" (Chen, Rasmussen, 1999)

5. Assessment Goals
• Determine retrieval rates
• Approximately how successful are subject searches for images in our teaching collection?
• Analyze search utility
• Is subject metadata a good access point for images?
• Test a framework for training image catalogers and structuring visual
literacy outreach
• Is there an easy way to improve the search utility and retrieval rates for these images?

6. Study Design
Data Sets
Existing
Existing images and metadata from the
Teaching Collection
Metadata created by many different image
catalogers from varying backgrounds over
a long period of time
Gathered
Survey to gather subject terms from
participants for each image (for
comparison against subject terms in
existing metadata set)
Participants = undergraduate A&S
students currently enrolled in an Art
History or Classics course
VS.

7. Study Design
Variable Group
Control version Variable version

8. Research Questions
1. What is the level of correspondence between the existing subject terms
for these images and the participant-assigned terms?
2. What is the level of correspondence in the types of subject terms
assigned by participants and those in the existing metadata?
3. Does providing users with a framework for analyzing the subject of an
image change the nature and content of the subject terms they choose to
assign to the image?

9. Research Question #1
What is the level of
correspondence between the
existing subject terms for these
images and the participant-
assigned terms?
TRANSLATION: Do users search
for images using the same
terms we use to describe them?
Existing Metadata:
Still lifes ; Fruit ; Vessels ; Clocks ;
Roemers ; Lemons ; Oysters ;
Goblets
Participant responses:

10. Research Question # 1
Example
Ethiopian Alphabet and Numerals
Wosene Worke Kosrof
mid-20th century
Existing Metadata:
Participant responses:

11. Research Question # 2
What is the level of correspondence in the types of subject terms assigned by
users and those in the existing metadata?
TRANSLATION: Do users search for images using the same TYPES of
terms that we use to describe them?
Primary Terms Secondary Terms Tertiary Terms
Objects and elements are
identified and named
“What is the image of?”
Objects and elements are
interpreted: characters are
identified, facial expressions are
interpreted, gestures are ascribed
meaning
“What is the image about?”
An awareness of the work/image
as an expressive cultural output
that is the product of a time,
place, and culture
“What is the image a good
example of?”

12. Research Question # 2
Example
Primary Terms Secondary Terms Tertiary Terms
“What is the image
of?”
“What is the image
about?”
“What is the image
a good example
of?”
Women; candles;
severed head;
maidservant;
interior; drapes;
swords; murder
Judith; Holofernes Tenebrism;
chiaroscuro
Judith and Maidservant with the Head of Holofernes
Artemesia Gentileschi
Ca. 1625
The Detroit Institute of Arts

13. Research Question #3
Does providing users with a framework for
analyzing the subject of an image change
the nature and content of the subject terms
they choose to assign to an image?
TRANSLATION: Does it make a
difference if we ask them these three
questions about each image?
1. “What is the image of?”
2. “What is the image about?”
3. “What is the image a good example
of?”

14. Findings!

15. Findings - types of terms
Search Utility
64%
34%
12%
13%
19%
16%
5%
37%
EXISTING DATA USERS
TYPES OF TERMS
Primary Terms Secondary Terms
Tertiary Terms Non-Subject Terms
Primary Terms
Objects and elements are identified and
named
“What is the image of?”
Secondary Terms
Objects and elements are interpreted:
characters are identified, facial expressions
are interpreted, gestures are ascribed
meaning
“What is the image about?”
Tertiary Terms
An awareness of the work/image as an
expressive cultural output that is the product
of a time, place, and culture
“What is the image a good example of?”

16. Findings - types of terms
Search Utility
71.70%
45.30%
0
47.20%
26.40%
15.30%
16%
0
5%
8.20%
13%
19%
0
32%
16.80%
0%
19.70%
0
15.80%
48.60%
EXISTING
DATA
USERS EXISTING
DATA
USERS
2D WORKS VS. 3D WORKS
Primary Terms Secondary Terms
Tertiary Terms Non-Subject Terms
2D works 3D works
Higher levels of
correspondence for
images of 2D works
Users were 2.5 times
more likely to use non-
subject terms when
describing images of 3D
works

17. Findings- types of terms
Search Utility
Non-subject terms
tended to capture other
key descriptive access
points: Culture,
Materials/Techniques,
Style/Period, Worktype
0% 10% 20% 30% 40% 50% 60%
Worktype
Style/Period
Materials/Techniques
Culture
Most common types of non-
subject access points

18. Findings - literal terms
Retrieval Rates
Literal matches = successful image
retrieval
Non-matches = unsuccessful image
retrieval
Successful retrieval = 8.5%
Unsuccessful retrieval = 91.5%
91%
9%
Correspondence between
existing metadata and users’
search terms
Non-matches Literal Matches

19. Findings
Retrieval Rates
Of that 8.5%...
• Primary Terms (75%)
• Secondary Terms (3%)
• Tertiary Terms (16%)
• Non-subject Terms (6%) - Other
descriptive metadata that does not
address subject meaning (i.e. materials
and techniques)
Corresponding literal terms
broken down by type
Primary Terms Secondary Terms
Tertiary Terms Non-Subject Terms

20. Conclusions
Primary terms yield the greatest search utility and higher levels of
successful image retrieval.
◦ Application: Focus image cataloging on assigning primary terms to images
High numbers of non-subject terms applied to images of 3D and non-
representational works suggest that subject metadata is a weak access
point for them
◦ Application: Forego full subject cataloging for these works and focus on non-
subject descriptive access points