Interactive Clustering for Transaction Data

We propose a clustering algorithm, OAK, targeted to transaction data as typified by market basket data, web documents, and categorical data. OAK is interactive, incremental, and scalable. Use of a dendrogram facilitates the dynamic modification of the number of clusters. In addition, a condensation technique ensures that the dendrogram (regardless of database size) can be memory resident. A performance study shows that the quality of clusters is comparable to ROCK [7] with reduced complexity.