Characterising Dataset Search Queries

Abstract

The amount of data generated and published on the web is increasing rapidly, but search for structured data on the web still presents challenges. In this paper we explore dataset search by analysing queries specifically generated for this work through a crowdsourcing experiment and comparing them to a search log analysis of queries on data portals. The change in search environment together with the task we gave people altered the generated queries. We found that queries issued in our experiment were much longer than search queries for datasets on data portals. They further contained seven times more mentions of geospatial and of temporal information and are more likely to be structured as questions. These insights can be used to tailor search functionalities to the particular information needs and characteristics of dataset search.