Agglomerative clustering using cosine and Jaccard distances: a computational approach to Roman vessel taxonomy

This paper addresses the issue of standardization in the cross-comparability of different vessel assemblages. It presents a computational method for building vessel categories from the bottom up, by comparing the specified attributes of a collection of vessel-types, and grouping like with like. Thus, it provides a platform for translating vessel data which may have been classified or divided by type using one taxonomy, bringing them into communication with those categorized by another. Two different methods of measuring the similarity among vessel-types (cosine similarity and the Jaccard index) are explored, toward providing a control on the resulting ‘synthetic’ categories. An exploratory dataset, collected from published data of archaeological projects in Italy focusing on ceramic vessels of the last two centuries BCE, was used to test the performance of this approach. Project data and results are open source and are available online at https://github.com/scollinselliott/synthkat/.