PNNL - Streaming Data Characterization

Stream reasoning systems are typically built using a fixed-size data windows, into which new elements from the stream are continuously added, and from which other elements are dropped. Our project develops a new class of window management algorithms based on semantic importance, using RDF data semantics to rank the elements of the window for their importance. Using semantics achieves higher reasoning performance, particularly for smaller window sizes, and takes account of background knowledge in the domain. We evaluate our approach using streaming soccer game data as well as AIM's insider threat data.

View publications

Funding Agency/Sponsor