ADJ-CABOSFV for High Dimensional Sparse Data Clustering

Abstract

The classic algorithm for high dimensional sparse data clustering, CABOSFV, cannot adjust the sets once generated, which leads to the final clustering result impacted by the preceding clustering result. This paper proposes ADJ-CABOSFV that can adjust the sets clustered by CABOSFV and the objects in the same set clustered by ADJ-CABOSFV are more similar without increasing the number of parameters. The experiments on UCI data sets show that ADJ-CABOSFV maintains superiority on high-dimensional sparse data of binary variables, and the clustering quality is better than the classic CABOSFV.

Incorporating domain ontology information into clustering in heterogeneous networks

“…With the similarities between central objects, the main principle of clustering central objects is to sequentially scan the central objects to form clusters, which modifies the procedure of Squeezer (He, Xu, & Deng, 2002) and CABOSFV (Wu & Gao, 2004) to effectively conduct high‐dimensional clustering. Specifically, the first central object is assigned to the first cluster, and for the second and later central object, the process must determine whether it must be merged into the existing clusters in terms of the similarity between the central object and an existing central object cluster.…”