Collection-Level User Searches in Federated Digital Resource Environment

1
Collection-Level User Searches in Federated
Digital Resource Environment

• Oksana Zavalina
• IMLS Digital Collections and Content project
• Graduate School of Library and Information
  Science
• University of Illinois at Urbana-Champaign
• 2007 ASIST Annual meeting

2

• IMLS Digital Collections and Content project at
  UIUC
• Started in 2002 with National Leadership Grant
• Aggregation of over 200 cultural heritage
  collections
• Collection Registry
• Provides access, services, and additional
  functionality to a database of collection
  descriptions
• Collection-level schema developed based on DC and
  RSLP (Research Support Libraries Programme, UK)
• Metadata repository
• Harvested metadata aggregated in one location
• Acts as a portal to the item-level records for
  digital content in NLG collections

3

• Participating institutions, 2006

4

• Collection Registry access through search and
  browse

5

• Subject Representation in the Registry
• Gateway to Educational Materials (GEM) subject
  headings
• Alternative subject headings (e.g., LCSH, AAT,
  locally-developed)
• Geographic coverage headings (Getty Thesaurus of
  Geographic Terms).

6

• Top GEM (Gateway to Educational Materials)
  subjects in Collection Registry
• Social Studies 80
• United States history
• State history
• Arts 46
• Visual arts
• Photography
• Science 17

7

• Collection records

8

• Collection records

9

• Research questions
• What is the distribution of the two major search
  types (subject and known-item) in the Registry?
• What are the typical user search categories in
  the Registry? Can FRBR set of 10 entities be used
  for user search categorization?
• What are the quantitative characteristics of a
  typical user search query in the Registry?
• How suitable is GEM subject scheme for describing
  diverse collections in the Registry compared to
  alternative controlled vocabularies?
• semantic similarity measures
• user keywords extracted from transaction logs
• subject terms in 3 different controlled
  vocabularies GEM, Library of Congress Subject
  Headings (LCSH), and Art and Architecture
  Thesaurus (AAT).

10

• Dataset
• MS Access file (7 months 19,000 records)
• 936 user keyword search query strings
• Minimal data processing
• manual extraction of keyword query strings
• aggregation of repetitive identical queries and
  morphological variants
• no query parsing
• stop-word list prepositions, conjunctions and
  articles
• Methods
• Transaction log analysis
• qualitative (subject analysis similarity
  measures)
• quantitative (basic descriptive statistics).

11

• Methods search categories

• 7 FRBR entities
• work collection as a work any intellectual or
  artistic creation that has a title attribute
• (individual) person
• corporate body
• concept
• object
• event
• place

• FRANAR/FRAD entity
• family
• Additional categories
• classes of persons (e.g., abused children,
  prisoners) 2 of queries
• ethnic/national group (Irish Americans, Sioux
  Indian) 5 of queries
• unknown search category (e.g., beyond, LU65)

12

• Operational definition of search types
• searches where the user queries either the title
  or the author individual or corporate of the
  digital collection belong to collection-level
  known-item search type
• all the other searches in the Registry belong to
  a collection-level subject search type.

13

• Methods similarity measures
• exact matches
• synonymous matches (semantic variants)
• near-exact matches
• syntactic variants (e.g., French art and Art,
  French)
• morphological variants (e.g., automated speech
  recognition and automatic speech recognition)
• acronyms (e.g., WW1 and World War, 1914-1918)
  
• NO broader and narrower terms matches

14

• Findings search categories
• Object, concept, place, individual person are
  heavily used
• surprisingly low level of event searching (4)

15

• Polysemy and search intent ambiguity problem
• Concept or Object?
• books, tools
• Amusement park, Ballrooms, Highways,
  interstates, detroithistoricalmuseums
• Industrial models, Lessonplans,
  dissertations
• Landscape
• Work or Person or Object?
• donquijote
• TomSawyer
• Event or Concept?
• Civil rights movement
• Census
• Single or multiple categories/entities?
• Lettersfrom19thcentury (object? object AND
  event?)
• childrenthatareabused (class of persons?
  class of persons AND event?)
• henryfordmuseumandgreenfieldvillage
  (corporate body? corporate body AND person AND
  place?)

16

• Findings search types
• Prevalence of subject search
• Higher than usual level of subject searching
• general shift towards subject searching in Web
  2.0?
• conceptual difference between collection-level
  and item-level search?

17

• Findings search query length

18

• Findings frequency of unique search query use
  (query popularity)

19

• Findings user queries by search category

20

• Findings semantic similarity

21

• Semantic similarity findings at a glance
• Weak semantic match between searches and GEM/AAT
  terms strong match for LCSH
• GEM represents only concepts, AAT only concepts
  and objects

22

• Findings semantic similarity overlap

23

• Semantic similarity overlap findings
• at a glance
• LCSH on its own (without any overlap with AAT or
  GEM) covers 48 of the user search terms.
• Only 12 terms (7) matched in AAT were not also
  matched in LCSH.
• All the terms matched in GEM were also matched in
  LCSH.
• 27 of user search terms were not matched in any
  of the three controlled vocabularies.

24

• Semantic similarity map

25

• Conclusions
• Unusually high for catalog use / transaction log
  analysis studies level of subject searching
• Strong semantic match to user queries offered by
  a traditional library subject scheme Library of
  Congress Subject Headings
• Combination of two or more standardized
  controlled vocabularies may be beneficial for
  collection-level subject description in IMLS DCC
  Registry
• Based on user searches, we recommend to update
  FRBR model to cover class of persons and
  ethnic/national group.

26

• Further research
• reasons for subject search prominence (interviews
  and observations of the Registry users)
• user conceptualization of the collection-level
  search and its possible difference from the
  concept of the item-level search
• investigate more flexible than LCSH controlled
  vocabularies, which, unlike GEM or AAT, represent
  a wide variety of search categories.

27

• Acknowledgements
• This research has been funded by IMLS NLG
  Research and Demonstration grant LG-02-02-0281
  http//imlsdcc.grainger.uiuc.edu/
• Special thanks to
• Timothy W. Cole Principal Investigator
• Carole L. Palmer Co-Principal Investigator
• Michael Twidale Co-Principal Investigator
• Amy Jackson, Sarah Shreeves, and Jenny Benevento
  current and former Project Coordinators

28

• Questions and comments always welcome
• Oksana Zavalina zavalina_at_uiuc.edu