Martina TROGNITZ
(Universität Heidelberg, Heidelberg, Germany)
Keywords: unsupervised machine learning, clustering, Aegean seals
Abstract:
I would like to present first results of my ongoing doctoral thesis about Minoan and Mycenaean seals with more than one seal face published in the “Corpus der minoischen und mykenischen Siegel”.
As for today just over a thousand multi-sided seals are known and recorded in the Arachne-database of the DAI and the University of Cologne. Due to the large number of seals, applying unsupervised machine learning techniques may be a viable way to detect whether there exist specific patterns for combinations of different motifs on the seal faces.
Considering material, provenance, number of seal faces and represented single or combined motifs, it may be possible, by means of clustering algorithms, to discover new seal groups, which then can be analysed from an archaeological point of view. This allows to answer questions like the following: Are there only patterns for some seal groups or none at all? If patterns exist, do they allow to draw conclusions about the function of the seals? Do they bear a meaning not yet known to us? Does the place of origin play a role in the design of a seal?
The poster will give an overview of the data, point out issues while preparing the data, discuss advantages and disadvantages of the used clustering methods for this task, and display first results from the clustering process.
