Kanji similarity hierarchical clustering