Adaptive Frequency Counts over Real-time Data Streams

Speaker: Bill LIN

Abstract

Computing frequency counts is an essential step in data mining. In several emerging applications, data takes the form of continuous data streams, which can be considered as streams of data such that each record can be processed or read only once. Traditional mining algorithms, however, are not applicable in mining data streams because they require multiple passes over the data.

In this talk, methods of frequency counting over a data stream will be presented. In addition to the one-pass property, real-time requirements are considered. We propose a flexible algorithm which computes frequency counts adaptively depending on the time constraints. A practical system based on the algorithm will be showed for mining data streams with bursty traffic.