Fakultäten » Wirtschaftswissenschaftliche Fakultät » Informatik, Institut für » Prof. Dr. Abraham Bernstein » Bernstein

Completed research project

Title / Titel Parallelized probabilistic detection of complex events on triple data streams
PDF Abstract (PDF, 14 KB)
Summary / Zusammenfassung The amount of data available (e.g., on on the Semantic Web and Linked
Open Data Cloud (LOD)) is growing at an astounding speed. An increasing
number of these data-sources are dynamic (i.e. their content changes
over time) or even represent continually updating phenomena (such as the
stock exchange, sensor networks, social networks, or the continuous
arrival of intelligence data). In many cases they also contain some
temporal information either explicitly given through temporal
constraints (e.g. limiting a triple’s validity by giving a start date
and some live span) or implicitly (i.e. defining its validity from the
moment it is made available until it is superseded by an update). One
possibility to process these kinds of data are stream-processing
systems that guarantee to correctly answer queries within acceptable
time-frames whilst (i) continuously consuming triples and (ii) utilizing
a limited amount of computational resources (i.e. space and time).

usual approach to address these requirements is the definition of a time
window over which the queries are evaluated: newly arriving triples
evict the oldest ones from the window of consideration. This approach
allows for continuous updates since a triple consumption process can
continuously update the window (often defined as a “logically circular”
region of memory) and a query process can evaluate queries over the
window. It also clearly limits memory consumption via the size of the
window. The time-window approach is, however, severely limited as it (1)
cannot cope with queries that will only match in time-spans longer than
the chosen window, (2) cannot evaluate queries with temporal constraints
as all triples within the window are usually regarded as being true
without any temporal constraint, and (3) does not account for
possibly-varying validity-spans of different parts of a query (just
consider a query that involves both TBox -- and ABox -- related triple-
patterns, where the former can be assumed to be long-living and the
latter can change continuously).

The goal of this project to find well-founded approaches the the scalable
detection of complex events in scalable triple-streams.
Weitere Informationen
Publications / Publikationen Weitere Informationen
Keywords / Suchbegriffe triples, semantic web, linked data, stream processing
Project leadership and contacts /
Projektleitung und Kontakte
Prof. Abraham Bernstein, Ph.D. (Project Leader)
Thomas Scharrenbach
Funding source(s) /
Unterstützt durch
Other Public Sources (e.g. Federal or Cantonal Agencies)
Duration of Project / Projektdauer Jan 2012 to Dec 2013