Figure 3

Accuracy change with different parameter settings for minCov and minCons on refining cluster overlaps from DIP data. These figures show the effect of different parameter values for minCov and minCons on the average f-score of the clusters refined. We used as input the preliminary clusters produced by (a) CFinder, (b) DPClus and (c) the modified entropy-based method. For minCss, we used the optimal values selected from the experimental results in Figure 2, 0.7 in (a) and (b) and 0.8 in (c). In (a) and (c), the best average f-scores of the refined clusters are achieved with the lowest values of minCov and minCons. In (b), the best accuracy is captured in the ranges between 0.2 and 0.5 for minCov and between 0 and 0.2 for minCons.