Interactive Mining of Frequent Patterns in a Data Stream of Time-Fading Models

Abstract

Mining frequent itemsets in data streams is an emergent research topic. Previous approaches generally assume a fixed minimum support threshold on mining patterns in the stream. However, allowing users to interactively specify minimum supports is more desirable in practice. In addition, the importance of stream data tends to decrease as time goes by. Thus, mining frequent patterns in streams of time-fading models is important for many applications. In this paper, we propose an algorithm that allows users to change the minimum support at any time in mining recently frequent itemsets in data streams of time fading models. A synopsis vector with a support decaying mechanism is constructed to summarizing past transactions. A batch of transactions will be incorporated into the synopsis for potential re-mining if the support changes. Extensive and comprehensive experiments were conducted over various datasets. The experimental results show that our approach has high precision and recall for mining recently frequent itemsets over the data streams with variable support thresholds.