Figure 3.

The effects on precision and recall rates from using a cutoff score. Test sets containing chemical names and words were evaluated with an MM trained on both types of data and cutoff scores ranging from 10 to zero were used to define which entries were valid. The data points from left to right reflect the precision and recall rates obtained by using each cutoff value, shown in descending order from 10 (far left) to zero (far right). The optimal tradeoff between precision and recall appears to be somewhere between a cutoff of one and two.