Describing collections with Datasheets for Datasets workshop materials

This workshop explored how web archives collections can be described using the framework provided by Datasheets for Datasets. Specifically, this work builds on the template for datasheets developed by Gebru et al. that is arranged into seven sections: Motivation; Composition; Collection Process; Preprocessing/Cleaning/Labeling; Use; Distribution; and, Maintenance. The workflow they present includes a total of 57 questions to answer about a dataset, focusing on the specific needs of machine learning researchers. For these workshops we used a later version of the framework developed by Microsoft. The Aether Data Documentation Template has nine sections and thirty one questions. This was an easier number to review in a short workshop. We considered how these questions can be adopted for the purposes of describing web archives datasets. Participants considered and assessed how each question might be adapted and applied to describe datasets from the UK Web Archive curated collections. After a brief description of the Datasheets for Datasets framework, we broke into small groups to perform a card-sorting exercise. Each pair evaluated a set of questions from the Datasheets framework and assessed them using the MoSCoW technique, sorting questions into categories of Must, Should, Can’t, and Won’t have. Groups then described their findings from the card-sorting exercise in order to generate a broader discussion of priorities and resources available for generating descriptive metadata and documentation for public web archives datasets.