Mining frequent patterns in a data stream is very
challenging for the high complexity of managing patterns
with bounded memory against the unbounded data. While
many approaches assume a fixed support threshold, a
changeable threshold is more realistic, considering the
rapid updating of the streaming transactions in practice.
Additionally, mining of itemsets over various time
granularities rather than over the entire stream may
provide more flexibility for many applications. Therefore,
we propose a interactive mechanism to perform the
mining of frequent itemsets over arbitrary time intervals
in the data stream, allowing a changeable support
threshold. A synopsis vector having tilted-time tables is
devised for maintaining statistics of past transactions for
support computation over user-specified time periods
The extensive experiments over various parameter
settings demonstrate that our approach is efficient and
capable of mining frequent itemsets in the data stream
interactively, with variable support thresholds. |
| Cite as: Lin, M.-Y., Hsueh, S.-C. and Hwang, S.-K. (2008). Interactive Mining of Frequent Itemsets over Arbitrary Time Intervals in a Data Stream. In Proc. Nineteenth Australasian Database Conference (ADC 2008), Wollongong, NSW, Australia. CRPIT, 75. Fekete, A. and Lin, X., Eds. ACS. 15-21. |
 (from crpit.com)
(local if available)
|