Improving distributional thesauri by exploring the graph of neighbors