Joseph T. Tennis (University of Washington, Seattle) “Casting Our Eyes Over the Threads of the Cataloguer’s Work: Population Perspective in Metadata Research”
Keynote at the KnoweScape workshop Evolution and variation of classification systems, March 4-5, 2015 Amsterdam
Albert Merono-Penuela: Understanding Change in Versioned Web-Knowledge Organi...
Joseph T. Tennis: Casting Our Eyes Over the Threads of the Cataloguer’s Work: Population Perspective in Metadata Research
1. Casting Our Eyes
Over the Threads of the
Cataloguer’s Work: Population
Perspective in Metadata Research
Joseph T. Tennis
University of Washington
Evolution and Variation of Classification Systems
KnoweScape Workshop March 4-5, 2015 Amsterdam
2. The question before us
What is the nature of the evolution* and
variation among knowledge organization
systems (KOS)?
Corollary questions
Is this a simple space or a complex space?
How often does it change?
Can we engender a common vocabulary to
describe this space?
*NB: evolution can be considered a loaded term by some – that is it
could be interpreted as fit for survival, and that is not what is intended
here. I often use change in lieu of evolution to clarify this.
3. The question before us
There are very practical reasons why we
want to ask this question.
Interoperability (sometimes called alignment*)
With widespread, yet still hopeful, collaboration
across cultural heritage sectors – those with rich
KOS, and with further development across a range
of sectors we must understand this problem of how
KOS interoperate, clarify its pressing issues, and
perhaps even incorporate this into formal
education.
*Alignment in my mind suggests more
similarities than differences, and this seems
presumptuous
4. The question before us
There are very practical reasons why we
want to ask this question.
Digital Preservation
Digital preservation is not simply the storage of
material on hard disk it is also the system of
policies and practices that guarantee digital
material a usable future. In service of that goal we
need to understand changes in our KOS.
5. The question before us
There are very practical reasons why we
want to ask this question.
Application Variations (repurposing)
By examining evolution and variety we can also
better evaluate particular applications of KOS. It
is one thing to study the standard, the ideal type, of
the KOS, but it is another see how different
institutions, sectors, and projects install and
perhaps alter that ideal type.
6. The question before us
I have, elsewhere, called the examination of
these phenomenon, how we change KOS
change over time and repurpose them, as
second-order problems [0]. The same is true for
designing for KOS interoperability.
This is because, in my mind, the first order is how to
design the KOS ex nihilo. And in many ways we
understand this problem of KOS design.
7. The question before us
So we are left to examine this universe of
KOS, how it changes, and the aspects of its
variety.
Now we can frame the question, and
establish what, from my perspective, we
know at this point.
We can then outline ways forward both in
research and development.
8. Outline
KOS as the product of problem-solving
Design of Metadata and Indexing
Languages
Metadata in the Wild
Time and Variety
Population Perspective and a Metadata
Observatory
10. KOS as the Product of
Problem-Solving
Ben Good has claimed that we are no in a
Cambrian Age of KOS [1].
In this context many different folks are trying to
solve the information organization problem.
Each of them has approached it from their
perspective, disciplinary biases, and using tools they
are familiar with (e.g., library classification, Protégé,
web browser bookmarks).
11. This is the first reason we might take a
population perspective in the study of
metadata. Namely, we expect variety.
For example, there was some debate in the
late 90s on whether or not ontologies were
the reinvention of classification. Vickery
took this up in 1997 and Soergel in 1999,
with Gilchrist taking a bird’s eye view in
2003 [2, 3, 4].
KOS as the Product of
Problem-Solving
12. As we read these accounts it becomes clear
that there are differences that make a
difference. And we are still discussing these
concepts, from various perspectives, in the
literature (cf., Barcellos Almeida, 2013 [5]).
And it is true that many think that such
variety is nothing by reinvention – recasting
old concepts and practices in new language.
Michael Gorman is one of these folks [6, 7].
KOS as the Product of
Problem-Solving
13. And yet, there are contexts where there is no
difference made.
KOS as the Product of
Problem-Solving
Are there differences
made in this LOV
service?
14. I have done some work on this problem. I
will introduce it a bit later. I have called it
framework analysis [8, 9, 10].
Suffice it to say here, that I believe it is useful
and generous to consider the program of
creating KOS as problem solving done by
many folks in many contexts.
KOS as the Product of
Problem-Solving
16. Design of Metadata and
Indexing Languages
Metadata
Machine and human readable assertions
about resources.
Indexing Languages
A set of representations, that is systematically
ordered, that provides access to the
conceptual content, and indicates or
establishes relationships, between terms to
denote concepts and between natural
language and terms used to denote concepts
17. Design of Metadata and
Indexing Languages
Indexing languages are, in my mind, the superclass
under which thesauri, classification schemes,
ontologies, taxonomies of various sorts hang.
Having said that, indexing languages can and are
used for other things than indexing. But we’ll not
take that up in this talk.* Soergel [3] offers a good
starting list of functions.
*But these may be of interest to
studying a wide variety of metadata
– articulating fully their purposes
18. Design of Metadata and
Indexing Languages
Metadata, confusingly, sometimes simply refers to
one subset of KOS or sometimes to the whole
universe of KOS.
This requires that we further clarify the form and
function that we assume we find in the universe of
KOS.
NB: KOS in my mind is both metadata AND
indexing languages.
19. Design of Metadata and
Indexing Languages
For me, metadata is human and machine readable
assertions about resources, where resources are the
W3C definition of anything with an identity.
Your definition may differ, and that is perhaps part
of our building a common vocabulary. So let’s
discuss.
However, I do not find it important to retrofit non-
machine readable description into the definition of
metadata. It has its own names (e.g., cataloguing).
20. Design of Metadata and
Indexing Languages
It has been helpful in the context of Dublin Core
Metadata work to clarify between schemes and
schemas. These are naïve distinctions, if you will,
made of convenience, and so through more
thorough research may be revised; but in this
context it is helpful, I think to distinguish between
the attributes of a resource and values you might
use to describe that resource.
21. Design of Metadata and
Indexing Languages
Attribute: Value
Author: Joseph T. Tennis
Subject: Evolution of KOS
Drawn from a schema: Drawn from a scheme (or
not)
We may find these don’t work well in some
contexts, but let’s try it out for now.
22. Design of Metadata and
Indexing Languages
Review
Metadata
Indexing Languages
KOS
Schemas
Schemes
23. Design of Metadata and
Indexing Languages
There is a large literature on the right way to
design metadata and indexing languages.
There is good reason for this, and it is a
useful body of literature. For one thing it is
not as straightforward as one might assume
to construct an indexing language.
24. Design of Metadata and
Indexing Languages
Whether one consults the literature or not,
the result of trying to solve problems in
information organization results in some
form of KOS.
And they are out there. Multiplying and
evolving.
26. Metadata in the Wild
If we take away the research on the design of
KOS, we are left with the literature that
describes how it is implemented, maintained,
and evaluated.
We are also left with literature that reads KOS
in particular ways.
27. Metadata in the Wild
In both of these cases we are talking about
metadata in the wild.
In 2005 we saw a declaration in the form of
a call for papers by Jack Andersen of the
then Royal School calling for, what I now
term, a descriptive turn in knowledge
organization research.
28. Metadata in the Wild
He said,
“Much classification research, and
knowledge organization research in general,
has tended to be concerned with rules,
principles, standards or techniques; that is,
with prescriptive issues. This workshop will
focus on descriptive issues,” [11].
29. Metadata in the Wild
Of course we had seen work well before this
time that could be described as descriptive
rather than prescriptive as well. We could
cite Richardson’s bibliography from 1901 or
earlier works that inventoried extant schemes
[12, 13].
30. Metadata in the Wild
And Bowker and Star have been famously
critical of decisions of classification as
infrastructure – where professional work
around changing what was there or in
faithfully representing controversial topics is
seen as compromise and therefore fruitful for
investigation. For example, representing the
full range of nurses work from medical
procedures to counseling is not
straightforward [14].
31. Metadata in the Wild
And finally, both Melanie Feinberg’s work
and Melissa Adler’s work, while quite
different, provide us ways in which we can
read KOS as authored rhetorical arguments
or institutions of dominance, power, and
instruments that promulgate particular
worldview if not prejudice, respectively [15,
16]
NB: Both at Local/Global Knowledge
Organization Workshop in Copenhagen
in August
32. Metadata in the Wild
And it is in this context, that we again ask the
question and its corollaries.
What is the nature of the evolution and variation
among knowledge organization systems (KOS)?
Corollary questions
Is this a simple space or a complex space?
How often does it change?
Can we engender a common vocabulary to
describe this space?
33. Metadata in the Wild
And it is here that we can begin to discuss what has
been done and how we might go forward.
35. Time and Variety
Time
I think it is safe to assume that we all know that
KOS change over time. We revise, edit, sunset,
phoenix, and otherwise rework our schemas and
schemes.
I have been curious about this since 2002.
36. Time and Variety
In an ISKO paper I looked at the entry from
EUGENICS relative index of the DDC at two points
in time, at edition 16 and edition 20. This simple
case study was enough to demonstrate there is
sometimes dramatic change in long-lived large
indexing languages.
I wanted to learn more.
37. Time and Variety
For those that do not know, EUGENICS is the body
of knowledge and the practice of creating better
human beings through selective breeding and
sterilization measures. It was once considered, by
the DDC to be a biological science. It is now a
widely debunked science, but the term persists in
many different contexts (even legitimate scientific
ones).
38. Time and Variety
It makes sense that if I was curious to see how
indexing languages (schemes) change over time I
could use this example and a couple of other
subjects to see how things change.
To that end I began data collection. This took a
village, but it was fun and worth the effort.
39. Time and Variety
We reviewed all editions of DDC for Eugenics and
Anatomy*
We identified where in the classification we could
find these subjects from 1876-2010. These were
often in different places (because of the nature of
DDC – variety cue!), but it showed us where
cataloguers might put books on these subjects.
*Among others, like Gypsies, Algebra,
Woman, Civil Disobedience, etc.
41. Time and Variety
The second set of data were gathered using Z39.50
protocol, harvesting MARC records from 572
catalogues that both (1) used EUGENICS or
ANATOMY as a first subject heading (in the 650 field
of the MARC record, the subject added entry for
topics) [17], and (2) used the DDC in the 082 field
of the MARC record. After automatically
removing duplicate records we were left with c. 927
records for EUGENICS and c. 1965 for ANATOMY.
43. Time and Variety
Combining this data would give us insight into
where some cataloguers were putting books on
EUGENICS and ANATOMY.
44. Time and Variety
A note about data, and this data specifically is that
it is MESSY and we do not necessary trust our
sources. So at best this is an exploratory look at
this phenomenon and we should improve on
methods of data collection and analysis.
45. Time and Variety
In this dataset we have
Date derived from LCCN
DDC class number
Date of publication
Date of publication cleaned (removing c. etc.)
Year differing between LCCN date and pub. date
Title
Server
Abridged notation present or not
Classification edition number if present
Record from Library of Congress?
Total count of identical records
46. Time and Variety
In this dataset we have
Date derived from LCCN
DDC class number
Date of publication
Date of publication cleaned (removing c. etc.)
Year differing between LCCN date and pub. date
Title
Server
Abridged notation present or not
Classification edition number if present
Record from Library of Congress?
Total count of identical records
DDC edition date
DDC classes possible
Discontinued classes
See alsos
Edition number
Notes
47. Time and Variety
We can now line these two datasets up and
explore our question about subject change
over time.
That is, we can see its ontogeny.
Ontogeny is the totality of changes of an
individual of a species from conception to
full maturation.
61. Time and Variety
There are many questions that can be asked
of this data and I will be talking more about
this tomorrow. I have some things here in
appendixes if we have time. I can also
provide citations.
62. Time and Variety
Variety
Now we can talk about variety in this
context. This is a harder problem for me,
because there may be infinite ways we
describe variety in KOS.
63. Time and Variety
Let’s take a (potentially) simple example.
What is the difference and similarity between
Descriptor Set (Mooers)
Thesauri
Classification Schemes
Schemes for Classification (Ranganathan)
Ontologies
Folksonomies?
64. Time and Variety
In the past I have looked at this in two ways.
By establishing a hierarchical or nested
method of comparative analysis
Through exploratory naïve linguistic
expression.
65. Time and Variety
I have tried to establish rubrics or
frameworks whereby we could lay various
standards of KOS against. These
frameworks include [18]:
Structure
Work Practices
Discourse
66. Time and Variety
Elsewhere, Elin Jacob and I say,
“The structure of a social tagging system, a metadata
scheme, or an indexing language must be understood within
the framework in which it occurs. The information
organization framework itself is comprised of three distinct
but interrelated components: the discourse that establishes
the goals, priorities and values of the system; the work
practices involved in the application and maintenance of
the system; and the structure that instantiates both the
discourses underlying the framework and the work practices
that make it visible,” [10].
67. Time and Variety
Elsewhere, Elin Jacob and I say,
“For example, ontology curation (or engineering) is an
information organization framework, and the Gene
Ontology (GO) is a specific instance of ontology curation.
The discourses revolving around GO reflect the fact that its
work practices are focused on representation of the natural
(or biological) world; and the structure of GO is therefore
informed by this scientific and representationalist focus and
the work practices and discourses that follow from that
focus.” [10].
68. Time and Variety
In an earlier project trying to make sense of the then
popular social tagging work (folksonomies), I tried to
compare that work to cataloguing in a similar way.
70. Time and Variety
The second way I have tried to characterize similarities and
differences has been with naïve linguistic expression.
In this exercise, Ben Good and I were trying to see if there
was a way to quantify a gold standard of indexing languages,
such that through automatic inspection we could assess and
modify those that were not satisfactory.
I must say that I was not convinced this was the right way to
go, but I was curious about what clusters would form and
why when we reduced all indexing languages to a bag of
terms and ran analysis over them.
72. Time and Variety
Excerpt
from [1]
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
%
OLP
uniterms:
%
OLP
duplets:
%
OLP
triplets:
%
OLP
quadplus:
OLP
flexibility:
%
containsAnother:
%
containedByAnother:
Number
disInct
terms:
Mean
Term
Length
Max
Term
Length
Min
Term
Length
Median
Term
Length
Standard
DeviaIon
-‐
Term
Length
Skewness
-‐
Term
Length
Coefficient
of
variaIon
-‐
Term
Length
OLP
max
number
sub
terms
per
term
OLP
mean
number
sub
terms
per
term
OLP
median
number
sub
terms
per
term
21
Connotea
73. Time and Variety
Excerpt
from [1]
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
%
OLP
uniterms:
%
OLP
duplets:
%
OLP
triplets:
%
OLP
quadplus:
OLP
flexibility:
%
containsAnother:
%
containedByAnother:
Number
disInct
terms:
Mean
Term
Length
Max
Term
Length
Min
Term
Length
Median
Term
Length
Standard
DeviaIon
-‐
Term
Length
Skewness
-‐
Term
Length
Coefficient
of
variaIon
-‐
Term
Length
OLP
max
number
sub
terms
per
term
OLP
mean
number
sub
terms
per
term
OLP
median
number
sub
terms
per
term
16
CHEBI
74. Time and Variety
Excerpt
from [1]
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
%
OLP
uniterms:
%
OLP
duplets:
%
OLP
triplets:
%
OLP
quadplus:
OLP
flexibility:
%
containsAnother:
%
containedByAnother:
Number
disInct
terms:
Mean
Term
Length
Max
Term
Length
Min
Term
Length
Median
Term
Length
Standard
DeviaIon
-‐
Term
Length
Skewness
-‐
Term
Length
Coefficient
of
variaIon
-‐
Term
Length
OLP
max
number
sub
terms
per
term
OLP
mean
number
sub
terms
per
term
OLP
median
number
sub
terms
per
term
1
MeSH
PrefLabels
75. Time and Variety
Excerpt
from [1]
0
0.2
0.4
0.6
0.8
1
%
OLP
uniterms:
%
OLP
duplets:
%
OLP
triplets:
%
OLP
quadplus:
OLP
flexibility:
%
%
Number
disInct
Mean
Term
Length
Max
Term
Length
Min
Term
Length
Median
Term
Standard
DeviaIon
Skewness
-‐
Term
Coefficient
of
OLP
max
number
OLP
mean
number
OLP
median
20
Bibsonomy
0
0.2
0.4
0.6
0.8
1
%
OLP
%
OLP
duplets:
%
OLP
triplets:
%
OLP
OLP
flexibility:
%
%
Number
disInct
Mean
Term
Max
Term
Min
Term
Median
Term
Standard
Skewness
-‐
Coefficient
of
OLP
max
OLP
mean
OLP
median
21
Connotea
0
0.2
0.4
0.6
0.8
1
%
OLP
uniterms:
%
OLP
duplets:
%
OLP
triplets:
%
OLP
quadplus:
OLP
flexibility:
%
containsAnother:
%
Number
disInct
Mean
Term
Length
Max
Term
Length
Min
Term
Length
Median
Term
Length
Standard
DeviaIon
-‐
Skewness
-‐
Term
Coefficient
of
OLP
max
number
sub
OLP
mean
number
OLP
median
number
22
CiteUlike
76. Time and Variety
I do not know if these means anything, but I have kept
collecting similar data.
I have about 36 single versions
of this data. Including English
dictionaries.
And here is where the two come
together. I need multiple versions
to make sense of this over time.
78. Population Perspective and a
Metadata Observatory
I have tried to demonstrate through my past research
that there is sufficient reason to investigate KOS from
a population perspective.
We have a wide range of standards, types, and a
potentially even wider range of implementations that
change over time.
In order for us to better understand this universe I
believe we need to work toward a metadata
observatory.
79. Population Perspective and a
Metadata Observatory
Like scanning the night sky for different instances of
blue dwarf stars or gassy giant planets, we can look
for various instances of schemes and schemas.
We can then see how they change over time. How
they are similar to or different from others.
Currently I’m interested in wikipedia’s category
system and its nature and changes. I’m also
interested in building a view of all the DDC numbers
in use. There would be a lot we could see from a
metadata observatory.
80. The question before us
What is the nature of the evolution* and
variation among knowledge organization
systems (KOS)?
Corollary questions
Is this a simple space or a complex space?
How often does it change?
Can we engender a common vocabulary to
describe this space?
*NB: evolution can be considered a loaded term by some – that is it
could be interpreted as fit for survival, and that is not what is intended
here. I often use change in lieu of evolution to clarify this.
81. Population Perspective and a
Metadata Observatory
Possible features of this observatory might be:
Real Time Metadata Feeds
Metadata Viz
Run Analysis on Metadata
Metadata Maps (geographic and conceptual)
Upload Your Metadata
Version Comparisons
82. Thank you
jtennis@uw.edu
Joseph T. Tennis
University of Washington
Evolution and Variation of Classification Systems
KnoweScape Workshop March 4-5, 2015 Amsterdam
83. Appendix A.
Time and Variety
Now that we have these visualizations in our
minds (perhaps), we can talk about
Semantic Gravity
Collocative Integrity
84. Appendix A.
Time and Variety
Semantic Gravity
Cataloguer privileges collection over updated
scheme (theory)
Collocative Integrity
Degree to which scheme comports with
cataloguing practice
86. Appendix A.
Time and Variety
0%
20%
40%
60%
80%
100%
1899
1911
1913
1919
1922
1927
1932
1942
1951
1958
1965
1971
1979
1989
1991
2003
Anatomy
Old
Out
In
0%
20%
40%
60%
80%
100%
1899
1911
1913
1915
1919
1922
1927
1932
1942
1951
1958
1965
1971
1979
1989
1996
2003
Eugenics
Old
Out
In
[19]
87. Appendix A.
Time and Variety
0%
20%
40%
60%
80%
100%
1899-‐2003
Eugenics
Old
Out
In
0%
20%
40%
60%
80%
100%
1899-‐2003
Anatomy
Old
Out
In
[19]
88. References
0 Tennis, J. T. (2010). Form, Intention, and Indexing: The Liminal and Integrated
Conceptions of Work in Knowledge Organization. In Advances in Classification
Research. Vol. 21. Available:
http://journals.lib.washington.edu/index.php/acro/issue/archive
1 Good, B. M. & Tennis, J. T. (2009). Term based comparison metrics for controlled and
uncontrolled indexing languages. In Information Research 14(1). Available:
http://www.informationr.net/ir/14-1/paper395.html
2 Vickery, B. V. (1997). Ontologies. In Journal of Information Science 23(4): 277-286.
3 Soergel, D. (1999). The rise of ontologies or the reinvention of classification. In JASIST
50(12): 1119-1120.
4 Gilchrist, A. (2003). Thesauri, taxonomies and ontologies – an etymological note. In
Journal of Documentation 59(1): 7-18.
5 Barcellos Almeida, M. (2013). Revisiting Ontologies: A Necessary Clarification. In
JASIST 64(8): 1682-1693.
6 Gorman, M. (1990). A Bogus and Dismal Science; or, the Eggplant That Ate Library
Schools. In American Libraries 21(5): 463-465.
7 Gorman, M. (1999). Metadata or cataloguing? In Journal of Internet Cataloging 2: 5-22.
89. References
8 Tennis, J. T. (2006). Comparative Functional Analysis of Boundary Infrastructures,
Library Classification, and Social Tagging. In Information Science Revisited: Approaches to
Innovation. Proceedings of the Annual Meeting of the Canadian Association for
Information Science/L'Association canadienne des sciences de l'information. York
University, Toronto.
9 Tennis, J. T. (2006). Function, Purpose, Predication, and Context of Information
Organization Frameworks. In Knowledge Organization for a Global Learning Society: Proceedings of
the 9th International Conference for Knowledge Organization. International Society for Knowledge
Organization 9th International Conference. (Vienna, Austria. Jul, 2006). Advances in Knowledge
Organization vol 10. Ergon. Würzburg: 303-310.
10 Tennis, J. T. and Jacob, E. K. (2008). "Toward a Theory of Structure in Information
Organization Frameworks." (2008). In Culture and Identity in Knowledge Organization:
Proceedings of the 10th International Conference for Knowledge Organization. (Montreal, Quebec
August 5-8, 2008). Advances in Knowledge Organization vol. 11. Ergon: Würzburg:
262-268.
11 Andersen, J. (2005). Call for papers. 16th ASIS&T SIG-CR Classification Research
Workshop, 2005, “What knowledge organization does and how it does it: Critical Studies
in and of Classification and Indexing.” Available: http://dhhumanist.org/Archives/
Virginia/v18/0597.html
90. References
12 Ricahrdson, E. C. (1901). Classification: Theoretical and Practical. Scribner’s Sons.
13 Horne, T. H. (1825). Outlines for the classification of a library; respectfully submitted to the
consideration of the trustees of the British Museum. G. Woodfall.
14 Bowker, G. and Star, S. L. (2000). Sorting Things Out: Classification and Its Consequences.
MIT Press.,
15 Feinberg, M. (2011). How information systems communicate as documents: the
concept of authorial voice. Journal of Documentation 67(6), 1015-1037.
16 Adler, M. (2015). Broker of Information, the “Nation’s Most Important Commodity”:
The Library of Congress in the Neoliberal Era. In Information and Culture 50(1): 24-50.
17 Library of Congress. (2007). 650-Subject Added Entry –Topical Term. http://
www.loc.gov/marc/bibliographic/bd650.html
[18] Tennis, J. T. (2006). Social tagging and the next steps for indexing. In Advances in
classification research, Vol. 17: Proceedings of the 17th ASIS&T SIG/CR Classification Research
Workshop (Austin, TX, November 4, 2006), ed. Jonathan Furner and Joseph T. Tennis.
[19] Tennis, J. (2013). Collocative Integrity and Our Many Varied Subjects: What the
Metric of Alignment between Classification Scheme and Indexer Tells Us About
Langridge’s Theory of Indexing. NASKO, 4(1). Retrieved from http://
journals.lib.washington.edu/index.php/nasko/article/view/14660