Convert-a-Card: Extracting Entities from Catalogue Cards to Create E-Records
Name
AKS_TUC24_Poster.pdf
Description
visibility:open
Size
651.43 KB
Format
Adobe PDF
Checksum (CRC64NVME)
HzlreCDgt6c=
Resource type
ConferenceItem Conference poster (unpublished)
Creator (person)
Keinan-Schoonbaert, Adi
Date published
2024
Abstract
This poster aims at showcasing the implementation of Transkribus within a workflow to convert physical catalogue cards into e-records, in order to make them discoverable via a library's online catalogue. With many workflow elements previously crowdsourced, we have been exploring new ways of extracting text, identifying relevant entities, and using those to search the OCLC WorldCat database for e-record retrieval. In this workflow, we use Transkribus to extract typewritten text, with the most important fields being shelfmark, author and title. Training and using P2PaLA layout models worked pretty well in one case, but less well in another. An alternative entity extraction method is the use of python code that identifies key elements on the cards based on their location and format. This work considers pros and cons for these information extraction methods, and provides a case study for efficient and accurate entity extraction as a cataloguing aid in heritage organisations.
Event title
Transkribus User Conference 2024
Official URL