Maintaining High Utility Pattern Trees in Dynamic Databases

Abstract

We have previously proposed the high utility pattern (HUP) tree for utility mining. In this paper, we further handle the problem of maintaining the HUP tree in dynamic databases. A HUP maintenance algorithm has thus been proposed for efficiently handling new transactions. The proposed algorithm can reduce the cost of re-constructing the HUP tree when new transactions are inserted. Experimental results also show that it indeed executes faster than the batch maintenance algorithm and generates nearly the same tree structure as the batch one. The proposed maintenance algorithm can thus achieve a good trade-off between execution time and tree complexity.