Top-k Pattern Matching Using an Information-Theoretic Criterion over Probabilistic Data Streams

As the development of data mining technologies for sensor data streams, more sophisticated methods for complex event processing are demanded. In the case of event recognition, since event recognition results may contain errors, we need to deal with the uncertainty of events. We therefore consider probabilistic event data streams with occurrence probabilities of events, and develop a pattern matching method based on regular expressions. In this paper, we first analyze the semantics of pattern matching over non-probabilistic data streams, and then propose the problem of top-k pattern matching over probabilistic data streams. We introduce the use of an information-theoretic criterion to select appropriate matches as the result of pattern matching. Then, we present an efficient algorithm to detect top-k matches, and evaluate the effectiveness of our approach using real and synthetic datasets.