A Novel Similarity-based Modularity Function for Graph Partitioning - Clustering Challenges in Biological Network - page 231

Fig. 11.3. Accuracy curves of GA using Q N and Q S

Table 11.3. Results of FHAC optimizing Q S

K in : K out

Best Q S #ofclusters Errors

11:5 0.614885 4 0

10:6 0.565117 4 4

9:7 0.481685 5 18

8:8 0.458222 8 44

Fig. 11.4. Accuracy curves of FHAC using Q N and Q S

when K out > 6, when the cluster structure becomes confused. However, one
may notice that the FHAC algorithm works much better when combining with the
proposed Similarity- based Modularity. It demonstrates that the proposed mea-
surement cooperates with the greedy search algorithm FHAC much better even
when the pre-defined cluster structure is not in favor of it.
While the synthetic network is a Type I network that does suffers from no hubs
or outliers, it is a significant problem to study to demonstrate the stability that
similarity adds. In that regard, hubs and outliers are elements that add confusion.

11.6.2. Real Applications

The first real application we report in our experiment is a social network - de-
tecting communities (or conferences) of American college football teams [6, 7,
12, 13]. The NCAA divides 115 college football teams into 13 conferences. The