Clustering of non-metric proximity data based on bi-links with ϵ-indiscernibility

摘要

In this paper, we propose a hierarchical grouping method for non-metric proximity data based on bi-links and ϵ-indiscernibility. It hierarchically forms directional links among objects according their directional proximities. A new cluster can be formed when objects in two clusters are connected with bi-directional links (bi-links). The concept of ϵ-indiscernibility is incorporated into the process of establishing bi-links. This scheme enables users to control the level of asymmetry that can be ignored in merging a pair of objects. Experimental results on the soft drink brand switching data showed that this approach is capable of producing better clusters compared to the straightforward use of bi-links.