Skip to main content
. Author manuscript; available in PMC: 2019 Oct 1.
Published in final edited form as: IEEE Trans Affect Comput. 2017 Mar 6;10(1):115–128. doi: 10.1109/TAFFC.2017.2678472

Fig. 7.

Fig. 7.

The agnostic Precision-Recall curve (by valence) based on manually annotated spammers. The top 20, top 40 and top 60 precision is 100, 95, 78 percent respectively (black line). It is expected that precision drops quickly with increasing recalls, because the manually annotation process can only identify a special type of spammers, while other types of spammers can be identified by the algorithm. The PR curves at γ = 0.3,0.37,0.44 are also plotted. Two baselines are compared: the Dawid and Skene (DS) approach and the time duration based approach.