Clustering is sometimes applied to multiple sets of
items, with each set being clustered separately. For
example, in the noun-phrase coreference task, a single
document’s noun-phrases are clustered by which nounphrases refer to the same entity (MUC-6, 1995), and
in news article clustering, a single day’s worth of news
articles are clustered by topic. In our method, users
provide complete clusterings of a few of these sets to
express their preferences, e.g., provide a few complete
clusterings of several documents’ noun-phrases, or several days’ news articles. From these training examples,
we learn to cluster future sets of items.