Usage-Driven Dublin Core Descriptor Selection

Dublin Core schemas are the core metadata models of most repositories, and this includes recent repositories dedicated to datasets. DC descriptors are generic and are being adapted to the needs of different communities with the so-called Dublin Core Application Profiles. DCAPs rely on the agreement within user communities, in a process mainly driven by their evolving needs. In this paper, we propose a complementary automated process, designed to help curators and users discover the descriptors that better suit the needs of a specific research group. We target the description of datasets, and test our approach using Dendro, a prototype research data management platform, where an experimental method is used to rank and present DC Terms descriptors to the users based on their usage patterns. In a controlled experiment, we gathered the interactions of two groups as they used Dendro to describe datasets from selected sources. One of the groups had descriptor ranking on, while the other had the same list of descriptors throughout the whole experiment. Preliminary results show that 1. some DC Terms are filled in more often than others, with different distribution in the two groups, 2. selected descriptors were increasingly accepted by users in detriment of manual selection and 3. users were satisfied with the performance of the platform, as demonstrated by a post-study survey.