Improved counter based algorithms for frequent pairs mining in transactional data streams
ECML PKDD'12 Proceedings of the 2012 European conference on Machine Learning and Knowledge Discovery in Databases - Volume Part I
| Hi-index | 0.00 |
We introduce the Pair Streaming Engine (PairSE) that detects frequent pairs in a data stream of transactions. Our algorithm finds the most frequent pairs with high probability, and gives tight bounds on their frequency. It is particularly space efficient for skewed distribution of pair supports, confirmed for several real-world datasets. Additionally, the algorithm parallelizes easily, which opens up for real-time processing of large transactions. Unlike previous algorithms we make no assumptions on the order of arrival of transactions and pairs. Our algorithm builds upon approaches for frequent items mining in data streams. We show how to efficiently scale these approaches to handle large transactions. We report experimental results showcasing precision and recall of our method. In particular, we find that often our method achieves excellent precision, returning identical upper and lower bounds on the supports of the most frequent pairs.