Knowledge Management in Association Rule Mining

Most current work on discovery of association rules assumes that the database from which the rules are determined is static. The mining operation is performed just once and therefore there is no need of knowledge management integration techniques. However, there are several domains where the database is updated on a regular basis. In these dynamic databases, it is hard to maintain the discovered rules since the updates may not only invalidate some existing rules but also make other rules relevant. We present an efficient algorithm, PELICAN, for incrementally updating the association rules when new transactions are added to or old transactions are removed from a dynamic database. Its efficiency comes from addressing several knowledge management issues in its implementation. The incremental mining operation is enabled by the maintenance and utilization of the knowledge discovered in the previous mining operations, which provides more interactivity, performance and scalability. Furthermore, PELICAN also allows obsolete transactions to be discarded, so that it can be used for short-term mining, and support counting to be relaxed, either for handling noises in the transaction stream or keeping the computational costs low. Extensive experimental data is presented that demonstrates the advantages of PELICAN.