Convert to tfidf space and search for the documents with the highest correlations (using, i.e., a clustering algorithm).