Towards intelligent open data platforms: Discovering relatedness in datasets

Open data platforms are central to the management and exploitation of data ecosystems. While existing platforms provide basic search capabilities and features for filtering search results, none of the existing platforms provide recommendations on related datasets. Knowledge of dataset relatedness is critical for determining datasets that can be mashed-up or integrated for the purpose of analysis and creation of data-driven services. When considering data platforms, such as data.gov with over 193,000 datasets or data.gv.uk with over 40,000 datasets, specifying dataset relatedness relationship manually is infeasible. In this paper, we approach the problem of discovering relatedness in datasets by employing the Kohonen Self Organsing Map (SOM) algorithm to analyze the metadata extracted from the Data Catalogue maintained on a platform. Our results show that this approach is very effective in discovering relatedness relationships among datasets. Findings also reveal that our approach could uncover interesting and valuable connections among domains of the datasets which could be further exploited for designing smarter data-driven services.