Figure 9.

Performance comparison between Snowball 2.0 and Snowball 1.0 in terms of precision (left), recall (middle), and F-score (right) as a function of τ_{sim_l}. All evaluations are cut off at τ_{sim_l} = τ_{sim_h}. Results of the “single term” formulation, a modified eq 5, are also shown for reference.