Abstract
In this paper, we propose a clustering method for non-metric proximity
data based on the $\epsilon$-indiscernibility. First, we introduce a
hierarchical grouping method based on bi-links, which groups objects
when bi-directional links are established between objects that have
asymmetric dissimilarities. Next, we incorporate the concept of
$\epsilon$-indiscernibility into the process of establishing
bi-directional links in order to allow 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
may have a possibility of producing better clusters compared to the
straightforward use of bi-links.
