Document clustering based on constructing density tree?