A former member of a student-run university broadcasting station at Kyonggi University in South Korea has developed the VOKU (Voice of Kyonggi University) Archive, a digitization project including detailed documentation of a workflow for LLM-assisted name redaction to preserve the privacy of students and staff whose names appear on the over 66,000 cue sheets used in the project. Jeoung Geun-Yeong, the independent creator of the project and a former station member of the radio station, digitized cue sheets created for the station’s student-produced campus radio broadcasts from 1988-2019, which can be explored through the project’s main website.
The project does not make publicly accessible the documents themselves nor the OCR’d full-text of the cue sheets, but it does attempt to preserve the history of student broadcasting through the application of human-and-agent-redacted versions of many of the cue sheets, making the materials accessible to the public in a way that prioritizes privacy by replacing real names of students and staff with consistently reused/ID-linked pseudonyms, carrying the synthetic name identifiers of people mentioned in the cue sheets across multiple documents. The project workflow is documented in a GitHub repository and the project data is hosted on Hugging Face, including some selected document images and text but not the full OCR of the complete real scripts, operational databases, real-name mappings, or private review records.
The project’s README discusses how Jeoung utilized GPT-6 to map named entities to persistent identifiers and perform the redaction of the documents in the archive.
dh+lib Review
This post was produced through a cooperation between Melissa Horak-Hern, Mariam Ismail, Trip Kirkpatrick, and Mimosa Shah (Editors-at-Large), Molly McGuire and Rachel Starry (Editors for the week), Claudia Berger, Ruth Carpenter, Caitlin Christian-Lamb, Nickoal Eichmann-Kalwara, Pamella Lach, Hillary Richardson, and Christine Salek (dh+lib Review Editors), and Tom Lee (Technical Editor).