Identifying User Interests within the Data Space - a Case Study with SkyServer

Many scientific databases nowadays are publicly available for querying and advanced data analytics. One prominent example is the Sloan Digital Sky Survey (SDSS)—SkyServer, which offers data to astronomers, scientists, and the general public. For such data it is important to understand the public focus, and trending research directions on the subject described by the database, i.e., astronomy in the case of SkyServer. With a large user base, it is worthwhile to identify the areas of the data space that are of interest to users. In this paper, we study the problem of extracting and analyzing access areas of user queries, by analyzing the query logs of the database. To our knowledge, both the concept of access areas and how to extract them have not been studied before. We address this by first proposing a novel notion of access area, which is independent of any specific database state. It allows the detection of interesting areas within the data space, regardless if they already exist in the database content. Second, we present a detailed mapping of our notion to different query types. Using our mapping on the SkyServer query log, we obtain a transformed data set. Third, we aggregate similar overlapping queries by DBSCAN and gain an abstraction from the raw query log. Finally, we arrive at clusters of access areas that are interesting from the perspective of an astronomer. These clusters occupy only a small fraction (in some cases less than 1%) of the data space and contain queries issued by many users. Some frequently accessed areas even do not exist in the space spanned by available objects.