1 of 38

Linked Open Citation Database�Enabling Libraries to Contribute to an Open and Interconnected Citation Graph

Anne Lauscher, Kai Eckert, Lukas Galke, Ansgar Scherp, Szed Tahseen Raya Riyva, Sheraz Ahmed, Andreas Dengel, Philipp Zumstein, Annette Klein

JCDL, 2018�Fort Worth, Texas

2 of 38

The origins: What were libraries cataloging?

  • Everything you can put in a shelve, i.e. cataloging on the level of physical objects
  • Dependent resources, e.g. journal articles, were not cataloged

2

3 of 38

Later: Resource Discovery Systems

  • Combination of classical library content with large indexes of article data
  • Indexes are dominantly run by commercial providers (Breeding, 2015)
  • Some indexes are provided by library service centers (e.g., K10plus-Zentral)

3

4 of 38

Now: Citations?

  • Commercial providers often include citation data in their catalogs (I4OC)
  • Fully automated approaches are still erroneous
  • .. libraries?

4

5 of 38

Linked Open Citation Database

Research Question

How much would it cost, with respect to resources, if libraries catalogued everything and curated the citation graph?

​

Method

  • Development of processes and tools based on linked data technologies to enable libraries to contribute to an open and interconnected citation graph
  • Quantitative and qualitative evaluation, e.g., cost benefit analysis

5

6 of 38

Agenda

  1. Motivation
  2. Data
  3. Reference Linking Workflow
  4. LOC-DB Infrastructure
  5. Evaluation
  6. Conclusion

6

7 of 38

Data

  • Social sciences collection of Mannheim University Library
  • 522 print books and collections acquired by in 2011: ~271,000 references
  • Articles published in 2011 in 101 (mostly electronic) journals: ~298, 000 references
  • New print acquisitions of the social sciences branch library from July 2017 on

7

8 of 38

Agenda

  • Motivation
  • Data
  • Reference Linking Workflow
  • LOC-DB Infrastructure
  • Evaluation
  • Conclusion

8

9 of 38

Library Workflow

  • Integrated into the standard library workflow
  • Reuse of data from existing resources, e.g. from publishers, other projects, and standard library catalogs (high quality metadata)
  • Automated as far as possible
    • Automatic reference extraction
    • Easy-to-use editorial system
  • Distributed database and collaborative cataloging processes

9

10 of 38

Standard Library Workflow (UB Mannheim)

For Print Resources

10

Scanning of the TOC*

Arrival of a new bibliographic resource

Cataloging

*Table of Contents

  • Metadata
  • Identifier

11 of 38

Reference Linking

Extending the Standard Library Workflow

11

Reference Linking

Scanning of the TOC* + List of references

Upload to LOC-DB

*Table of Contents

  • Metadata
  • Identifier

Scan of the list of references

12 of 38

Reference Linking

Extending the Standard Library Workflow

12

Reference Linking

Scanning of the TOC* + List of references

Upload to LOC-DB

*Table of Contents

  • Metadata
  • Identifier

Scan of the list of references

Reference String

?

13 of 38

Reference Linking

Extending the Standard Library Workflow

13

Reference Linking

Scanning of the TOC* + List of references

Upload to LOC-DB

*Table of Contents

  • Metadata
  • Identifier

Scan of the list of references

Supported by our infrastructure

14 of 38

Agenda

  • Motivation
  • Data
  • Reference Linking Workflow
  • LOC-DB Infrastructure
  • Evaluation
  • Conclusion

14

15 of 38

LOC-DB Infrastructure

15

Editorial System

Automatic Reference Extraction

LOC-DB

Instance 1

Linked Open Data

LOC-DB

Instance 2

LOC-DB

Instance N

PDF

Web

Print

16 of 38

LOC-DB Infrastructure

16

Editorial System

Automatic Reference Extraction

LOC-DB

Instance 1

Linked Open Data

LOC-DB

Instance 2

LOC-DB

Instance N

PDF

Web

Print

17 of 38

Automatic Reference Extraction

Combination of text-driven and layout-driven extraction using deep learning techniques (“DeepBIBX”; Bhardwaj et al., 2017)

17

18 of 38

LOC-DB Infrastructure

18

Editorial System

Automatic Reference Extraction

LOC-DB

Instance 1

Linked Open Data

LOC-DB

Instance 2

LOC-DB

Instance N

PDF

Web

Print

19 of 38

19

20 of 38

20

21 of 38

21

22 of 38

22

23 of 38

LOC-DB Infrastructure

23

Editorial System

Automatic Reference Extraction

LOC-DB

Instance 1

Linked Open Data

LOC-DB

Instance 2

LOC-DB

Instance N

PDF

Web

Print

24 of 38

Reference Target Suggestions

Speeding up the linking process

24

Reference Query

Internal Search Index

External Suggestion Engine

Similarity Computation

Similarity Threshold Filter

Ranking

Similarity Threshold Filter

...

25 of 38

Data Model and Publishing

Ensuring optimal reusability and interoperability of the produced data

​

  • Adaption of the OpenCitations metadata model (Peroni and Shotton, 2016)
  • Publishing of the Data in RDF format by using the Semantic Publishing and Referencing (SPAR) Ontologies (Peroni, 2014)

25

26 of 38

Agenda

  • Motivation
  • Data
  • Reference Linking Workflow
  • LOC-DB Infrastructure
  • Evaluation
  • Conclusion

26

27 of 38

Results of our Public Workshop (Wilke and Retter, 2017)

  • Cataloging citation data is an important task for libraries
  • Data from union catalogs should be used
  • Suggestions help to minimize the time required to resolve a link
  • Acquired citation data should be integrated in existing systems

27

28 of 38

  • Cataloging citation data is an important task for libraries
  • Data from union catalogs should be used
  • Suggestions help to minimize the time required to resolve a link
  • Acquired citation data should be integrated in existing systems

Save the date: The LOC-DB Workshop 2018 will take place on the 6th November, 2018 in Mannheim!

​

28

29 of 38

How much time does the whole process take?

  • > 100 pages per person per hour
  • Upper bound ~ 15 minutes for scanning for an average book (26 pages of references)
  • Additional scanning time does not significantly affect other processes in the library
  • Prolongs the processing of a book on average by only 3 minutes

29

Reference Linking

Scanning of the list of references (only print)

Upload to LOC-DB

30 of 38

How much time does the whole process take?

  • Batch upload
  • Background processing for meta data retrieval and reference extraction

→ Does not affect the process in the library

30

Reference Linking

Scanning of the list of references (only print)

Upload to LOC-DB

31 of 38

How much time does the whole process take?

31

Reference Linking

Scanning of the list of references (only print)

Upload to LOC-DB

Scanning of the list of references (only print)

Upload to LOC-DB

Criterion

Minimum

Maximum

Median

Citation Linking (s)

9.93

557.20

89.45

Internal Suggestion Retrieval (s)

0.02

0.5

0.06

External Suggestion Retrieval (s)

0.50

95.65

0.89

# Searches per Reference

1

36

2

Minimum, maximum, and median time in seconds

for the reference linking step

​

​

32 of 38

How much time does the whole process take?

32

Reference Linking

Scanning of the list of references (only print)

Upload to LOC-DB

Scanning of the list of references (only print)

Upload to LOC-DB

Histogram of reference linking times

​

​

33 of 38

Agenda

  • Motivation
  • Data
  • Reference Linking Workflow
  • LOC-DB Infrastructure
  • Evaluation
  • Conclusion

33

34 of 38

How much would it cost if libraries catalogued everything and curated the citation graph?

​

​

Estimation about the number of full-time employees needed to process all literature of social sciences bought in 2011 by Mannheim University Library, depending on the time t in seconds to resolve a reference.

​

​

​

​

​

34

t

1

5

10

20

30

60

120

# employees

0.1

0.5

1

2

3

5.9

11.9

35 of 38

How much would it cost if libraries catalogued everything and curated the citation graph?

​

​

Preliminary results suggest general feasibility of the approach�Continuous improvement on the infrastructure and processes ongoing��Semi-automated approach ensures human-level quality of the generated data�

​

​

​

​

​

35

36 of 38

Thank you for your interest

36

For more information please visit

https://locdb.bib.uni-mannheim.de

https://github.com/locdb/

http://wisslab.org

​

Anne Lauscher & Kai Eckert

37 of 38

Bibliography

  • Marshall Breeding. 2015. Future of Library Discovery Systems. Information Standards Quarterly 27, 1 (2015), 24. https://doi.org/10.3789/isqv27no1.2015.04
  • Christian Wilke and Regina Retter. 2017. Zitationsdaten extrahieren: halbautomatisch, offen, vernetzt. Ein Workshopbericht. Informationspraxis 3, 2 (Dec. 2017). https://doi.org/10.11588/ip.2017.2.43235
  • Silvio Peroni and David Shotton. 2016. Metadata for the OpenCitations Corpus. Technical Report. https://dx.doi.org/10.6084/m9.figshare.3443876
  • Akansha Bhardwaj, Dominik Mercier, Andreas Dengel, and Sheraz Ahmed. 2017. DeepBIBX: Deep Learning for Image Based Bibliographic Data Extraction. Springer International Publishing, Cham, 286–293. https://doi.org/10.1007/978-3-319-70096-0_30

37

38 of 38

Bibliography

  • Silvio Peroni. 2014. The Semantic Publishing and Referencing Ontologies. In Semantic Web Technologies and Legal Scholarly Publishing. Springer, Cham, 121–193. https://doi.org/10.1007/978-3-319-04777-5_5

​

​

38