A robust iterative refinement clustering algorithm with smoothing search space
Knowledge-Based Systems
An analysis of collaborative patterns in large-scale ontology development projects
Proceedings of the sixth international conference on Knowledge capture
| Hi-index | 0.00 |
We present a novel algorithm called PR, which initializes cluster centers based on pointer ring. PR partitions traditional hyper-rectangular units further to hyper-triangle subspaces. It takes the geometric center of a unit as its symmetrical center. With the interconnections between the center and vertices of the unit, we get 2d hyper-triangle subspaces for each ddimensional unit. Next it decides direction pointers for the hyper-triangle subspaces according to the difference of space density. After that, PR gathers the hyper-triangle subspaces which have the uniform direction to the center of pointer rings in different units which connected one by one .Thus PR finds all the pointer rings by scanning the grid units Candidate central fields appear from extending pointer rings. Finally, PR takes the geometrical centers among total data points contained in candidate central fields as initial cluster centers. Experiments have shown that PR is helpful to finding cluster centers quickly and effectively. Moreover, the running time of PR is linear with respect to number of instances, number of grid units and dimensions. This makes PR scalable to large datasets.