Leben gelabelt Computergestützte Inhaltverzeichnisse für Oral-History-Interviews

Abstract

Machine-based methods for content analysis are not only used to obtain statistical results. The quantitative results can also be used qualitatively to identify patterns and thematic trends in the texts under investigation. This essay documents how a corpus of life history interviews, compiled from collections on the Oral History portal, was used to create a thematic index for all interviews represented in the portal using topic modelling, a machine learning method. This thematic index serves as an index for cross-archive searches and as a basis for thematic analysis. After providing an overview of the most common methods for automatic content indexing of texts, the entire process, from the compilation of the corpus to the finished topic index and sample tables of contents, is presented in a transparent manner. The default setting of all parameters, from the size of the sections to the number of topics, has been a major challenge so far. Automated processes were unable to deliver clear results, which is why a qualitative approach is proposed, in which the approximation of the optimal number of topics is evaluated using a zoom-in and zoom-out procedure (scalable reading). A group of experienced oral historians then labelled the individual topics and grouped similar topics into clusters. Finally, the labelled topics were implemented as a register in the oh.d platform. On this basis, tables of contents can be created for all interviews.

Bibliography: Bayerschmidt, Philipp/Möbus, Dennis: Leben gelabelt. Computergestützte Inhaltverzeichnisse für Oral-History-Interviews, BIOS – Zeitschrift für Biographieforschung, Oral History und Lebensverlaufsanalysen, 1+2-2025, pp. 83-104.