Database Reference
In-Depth Information
Many of the traditional database operators for stream processing can-
not effectively handle the detection of arbitrary event patterns. These
techniques are quite effective for regular expression matching, though
not quite as effective for pattern matching. The latter is much more
important in the stream scenario. Therefore, the work in [6] proposes
a method for matching arbitrary patterns in data streams in order to
perform event detection. This work proposes a formal query evaluation
model, NFA b that combines a finite automaton with a match buffer.
This is used to create query evaluation plans that can be executed over
event streams. One characteristic of this approach is that it uses storage
sharing of all possible pattern matches as well as in automaton execution
to produce these matches. This results in more ecient query execution.
4.1 Probabilistic Event Extraction
It has been observed in [45, 47] that there is an inherent ambiguity
in the cleaning and determination of high level events of RFID data.
Since, the collection of RFID data is prone to errors, it is natural that
such data is best represented by probabilistic databases as discussed
in [17, 31, 77]. The importance of using probabilistic representations
for event extraction in pervasive computing applications has been dis-
cussed in depth in [25], though the approach discusses the design of
an inference engine for event extraction. In the context of RFID data
management applications, it is more critical to design a query processing
engine for probabilistic event extraction. Therefore, the work in [45]
proposes a probabilistic event language PeexL for defining probabilis-
tic events. An implementation of the approach is proposed in a system
called Probabilistic Event EXtractor (PEEX) , a middleware layer on
top of a relational database management system (RDBMS). The idea is
that uncertainty propagates as events are aggregated into higher level
events. For example, a MEETING event can be inferred from a sequence
of ENTERED-ROOM events by different participants. However, if there is
limited confidence in the ENTERED-ROOM events, then the confidence in
the MEETING events will also be lower. The work in [45, 47] uses confi-
dence tables in order to track the confidence of the different events and
then aggregate these probabilities into higher level event probabilities
with the use of the PeexL language. Another interesting probabilistic
event processing system known as Lahar has been proposed in [61]. This
approach uses a framework which is similar to the Cayuga system [18]
for event processing, except that it is focussed on querying probabilistic
representations of the underlying data.
Search WWH ::




Custom Search