Fig. 7.

Classifier performance comparison: Snorkel-based classifier (red curve$\hspace{0.17em}$, $N=415$) vs. Botometer (green curve$\hspace{0.17em}$, $N=289$) vs. Bot Sentinel (blue curve$\hspace{0.17em}$, $N=288$), given proxy, community-based truth. The dashed gray horizontal line at 63% is the fraction of presumptively true examples in the community-based proxy for known IO accounts and therefore represents random chance precision performance.