From is more advantage to randomized algorithms than the

From Figs. (4) and (5),
we can see that overlapping community structure with extra small communities
may degrade the community quality. There are two questions about the merging
strategy. First, the post-process HABM strategy an universal and merging
technique have effectively used for the post-processing procedure? Second, can
HABM boost the overlapping community structure generate by any other approach?
To answer these questions, we use post-process HABM strategy for overlapping
communities generated six competitors CPM, ABL, EPM, MCMOEA, BMLPA, and BNMTF.
The results are summarized in Table III.

TABLE III

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!


order now

IMPACT OF
MERGING STRATEGY ON COMMUNITY QUALITY IN TERMS OF NMI

 

Datasets

Algorithms

Before Merging

After Merging

Delta

Zachary

 
CPM
ABL
BMLPA
BNMTF
EPM
MCMOEA
LEPSO

 
0.335
0.409
0.347
0.346
0.857
0.885
0.825

 
0.335
0.894
0.393
0.38
0.865
0.901
0.906

 
0
0.485
0.046
0.034
0.008
0.016
0.081

Dolphins

CPM
ABL
BMLPA
BNMTF
EPM
MCMOEA
LEPSO

0.461
0.209
0.649
0.602
0.782
0.816
0.712

0.461
0.269
0.73
0.655
0.814
0.821
0.823

0
0.06
0.081
0.053
0.032
0.005
0.112
 

From Table III we can
see that all the entries in the Delta column are non-zero and the maximum
values in Delta are achieved by randomized algorithms. This means that our
proposed overlapping community enhances the quality of the merging strategy.
And HABM is more advantage to randomized algorithms than the non-randomized
algorithms. It also verifies the advantageous to use the line-graph and
ensemble clustering technique GbestGenerator. We conclude that HABM is useful
to a wide range of the approaches, particularly to randomized ones, for
overlapping community detection.

Fig 6
Performance Analysis

 

VI.CONCLUSION
AND FUTURE WORK

A meta-heuristic
algorithm LEPSO has been proposed for overlapping community discovery from
social networks. Speci?cally, a particle representation scheme based on the
ordered neighbor list and a particle update strategy proposed. Also, a
hierarchical agglomerative and bottom-up merging strategy is designed to
post-process the generated ?ne-grained overlapping communities. The experiments
and the results show that 1) compared with the randomized and non-randomized
algorithms, our LEPSO is superior in terms of strength and robustness, and 2)
the proposed hierarchical agglomerative and bottom-up merging strategy (HABM)
can improve quality of the generated overlapping communities.

As future work, K-means
clustering and the spectral clustering can be explored to achieve better
performance.