The semantic web provides more and more resources that are amenable to machine interpretation, for example, RDF files or RDF-annotated web contents that can be processed via the query language SPARQL. However, mobile devices such as smartphones are hard-pressed to utilize this potential due to their limitations with respect to connection bandwidth, processing power, and energy consumption, which make regular queries over large datasets infeasible.

The paper addresses this problem with a novel kind of mobile query service that incorporates three optimizations: (1) a source index model (SIM) that indexes data in a such a way that, from an analysis of a query with respect to the involved predicates and resource types, only a small subset has to be considered; (2) a cache component that locally preserves downloaded data for reuse in later queries; and (3) a cache replacement strategy, least popular sources (LPR), that removes from a full cache all data from that source that is last in a ranking that considers the popularity of both source data and metadata.

These ideas are clearly presented in the first part of the paper; its second part is dedicated to a thorough experimental evaluation of the optimizations. The authors conclude that a combination of an SIM that considers predicates and subject/object types with a cache whose organization is based on shared source metadata and LPR performs best. A problem that remains to be addressed in future work is the cost of maintaining the cache due to newly inferred resource types.