This document summarizes JSTOR's efforts to develop a sustainability collection and utilize their controlled vocabulary thesaurus (JTHES) to semantically index and organize content in the collection. Subject matter experts were enlisted to review and provide feedback on key sustainability terms. A prototype sustainability portal was created that uses the thesaurus and semantic indexing to power auto-suggest, discovery, and refinement of results. Refinements are ongoing, including automating the calculation of documents' sustainability scores and topic labeling.

1. JSTOR Sustainability
Collection
Sharon Garewal, JSTOR Senior Metadata Librarian
Ron Snyder, ITHAKA Labs Director of Research and Development

2. Overview
 Sustainability collection defined
 Utilization of the thesaurus within the sustainability collection
 Subject matter experts enlisted
 Results
 Live demo

3. JSTOR- a quick primer
 3,200+ journals & 30,000+ books
 9.3 million full length articles
 70 million pages
 2.9 million book reviews
 138 million content accesses in
2013
 100 million searches per year
http://www.jstor.org/

4. Sustainability Collection: what will it
be?
 Driver: Emerging interdisciplinary
area that JSTOR wanted to
support in both research and
teaching needs.
 Core topics of Cities and
Urbanization, Food and
Agriculture, Industrial Ecology,
Resource Economics, Forestry and
Land Use and Environmental
Policy and Law
 Composed of journals, books,
grey literature (working reports,
research reports, technical reports
etc.)
 Specialized functionality to
support research by including
semantic indexing to help
researchers locate related terms
and concepts. This is where the
JSTOR Thesaurus (JTHES) comes
into play!

6. JTHES
19 Top terms, 57,470 Terms;
103,129 rules

7. The challenge
 To assemble a list of key terms in Sustainability
 The terms will be used to organize and tag sustainability-related research articles
on JSTOR starting in 2015.
 These terms will also be used for an auto complete function in the search
component.
 Utilize the JTHES in a live prototype
 This was the first project where we looked at how to use the thesaurus as an
intelligence layer within a collection. How should it work? How do we do this?

8. How do we get this done? The
options…
 Create a new thesaurus for sustainability:
 Pros: Specific to sustainability
 Cons: Remembering to make changes in more than one place. Cost associated with
creating and maintaining a separate thesaurus
 Create a sustainability branch within JTHES:
 Pros: Could BT (Broader terms) all relevant branches and terms from elsewhere in the
JTHES into 1 branch
 Cons: Redundant; Multiple BT’s clutter up the JTHES
 Create a facet to tag terms within JTHES as “Sustainability”:
 Pros: Creates a flat list (in faceted view) of all of the terms in that facet; Easy to
maintain
 Cons: Does not show a hierarchy; Cannot have multiple facets

9. The road to sustainability…
 Research: examined existing
glossaries and thesauri created
by research libraries, discipline
associations and individual
scholars in each of the disciplines.
 Existing terms (pulling lists)
 Existing branches (clean up)
 Adding new terms
 Adding new branches: Food
studies, Urban studies, etc.
 Constructing new rules and
refining existing rules
 Testing content

10. Enlisting Subject matter experts
 Contacted faculty members in ten disciplines to go over the subset of terms
assembled in their discipline and review those terms with an eye toward:
 Is this how people in the field express this concept?
 Is it correctly included in the sustainability facet?
 Are there any important terms or concepts that we've missed? (including
acronyms, synonyms, variant spellings, inverted phrases)

11. SME spreadsheets
Each SME was slightly
different in how they
approached their
subject areas with
some SMEs being
reluctant to give much
feedback and others
giving large amounts
of feedback to sift
through.
Example of terms pulled from Law, Public administration/policy and International/global studies

12. View- Facet
provides
alphabetical
list of all
tagged terms

13. labs.jstor.org/sustainability

14. Implementation of the Sustainability
Prototype
 The thesaurus and semantic index are used for content discovery and
presentation
 The identification of a “sustainability collection” from the JSTOR corpus was
performed using topic modeling (specifically LDA – Latent Dirichlet
Allocation)
 A model of 100 topics was generated from the content
 Staff assigned sustainability scores for each of the topics based on a review of
the top words in each topic
 Each document in the JSTOR corpus was then assigned a sustainability score of
0-9 based on the sustainability scores for the topics most closely associated with
the document

15. Weighting of document-level indexed
terms
 Document-level weights were computed for each sematic term using TF-
IDF
 TF-IDF is a measure of how important a word is to a document in a collection
 The TF-IDF value increases proportionally to the number of times the word
appears in a document (the ‘TF’ or term frequency), but is offset by how
common the word is in a corpus (the ‘IDF’ or inverse document frequency)
 The TF-IDF weighted terms are used to:
 order the terms displayed for each document
 boost document relevancy when index terms are used in discovery

16. Auto-suggest and refining results
[Thesaurus slide: a new thing, metadata we create, screenshot(s) of
Sustainability Portal]

17. Refinements in our use of the thesaurus
and semantic index in sustainability
 Auto calculation of sustainability score using LDA topics and thesaurus
sustainability facet
 Calculate topics and term correlations
 Compute sustainability score for each topic based on the most relevant terms and
sustainability facet
 Compute a sustainability score for each corpus document based on topic weights
and topic sustainability score
 Automated LDA topic labeling
 Labeling topics generated by unsupervised topic modeling is an ongoing challenge
 We’re investigating the feasibility of using the same topic/term correlations used to
compute sustainability scores to assign labels
 Attempts to find the thesaurus term that best characterizes the most highly correlated
terms for each topic

18. Other JSTOR Labs projects/tools using
the thesaurus and semantic index
http://labs.jstor.org/jthes/
http://labs.jstor.org/snap/
http://labs.jstor.org/readings/
Thesaurus Visualization Tool

19. And some other JSTOR Labs projects
http://labs.jstor.org/reflowit/
http://labs.jstor.org/shakespeare/

20. Thank you!
Sharon.Garewal@ithaka.org
Ronald.Snyder@ithaka.org