Dr Anusuriya Devaraju1, Mr Dominic Hogan2
1CSIRO Mineral Resources, Perth, Australia,
2CSIRO, Brisbane, Australia
With an increase in the rate of data publication on CSIRO’s Data Access Portal (DAP) comes the challenge of helping users to discover relevant datasets. Research data repositories typically allow searching via keywords and faceted navigation, where users benefit from an existing familiarity with the content, whether browsing by subject or navigation approaches. Research was conducted into the feasibility of a hybrid recommendation approach, presenting users with recommended datasets. The approach leverages content-based similarity and usage patterns, tuned to a feature weighting model obtained through a survey involving real users. The results of the model were then evaluated in a user study, indicating which ranks of recommended results were deemed relevant. Following this research, the model was implemented in the DAP and released to the public in March 2018. To our knowledge this is the first implementation of such a recommendation system for research datasets. We present a preliminary analysis of the use of these recommendations by general users of the DAP, discussing the proportion of users following recommendations and their activity compared to the wider population of DAP users.
Dominic Hogan is a business analyst at the Commonwealth Scientific and Industrial Research Organisation. From 2012 to 2017 he worked as a data librarian supporting data management across CSIRO, and has been heavily involved in development work for CSIRO’s Data Access Portal. In his work he has supported work in various research domains, including terrestrial ecology, marine research, computer visualisation and materials science. Recently he has contributed to a project implementing a recommender system for research datasets in CSIRO’s data repository.