abc

1155 Words3 Pages

Defining and evaluating network communities based on Ground truth

In this paper, the authors define ground truth communities by selecting networks where the nodes define their relationship with the groups. After determining the ground truth communities, a comparison is performed between the network communities and the ground truth communities to find out difference of result in 13 chosen structural definitions of network communities. These 13 structural definitions gets partitioned into four classes and tested on the basis of three parameters: sensitivity, reliability and performance to determine the ground truth. Besides this, author also tries to find the network communities in case of a single node. To achieve the task author applies spectral clustering along with heuristic parameter-free algorithm to detect the communities of the node. The advantage of this algorithm is that it is extremely scalable and can be applied to networks with millions of nodes. We have studied clustering in our lecture; here the clusters might overlap as a node can have relationships with many communities, it is not confined to only one community, it means that the nodes are not exclusive. Also, these nodes in the network form densely linked clusters.
As discussed above, in order to achieve the goal the process is divided into three steps: firstly, study a large set of networks to find the ground truth (here the size of the set is 230 that consists of social, collaborative and information networks); secondly, compare the 13 selected structural definition for network communities to ground truth communities and lastly, apply the spectral clustering along with heuristic parameter free algorithm.
In order to define the ground truth community, authors h...

... middle of paper ...

...o find out the communities whereas they build ground truth communities based on explicitly formed groups. They point out that their algorithm is parameter free and totally different from [2][28] as they only use sweeping under Conductance as a scoring function whereas here many scoring functions are used for sweeping.

Final remarks:
The paper by Yang et al is well represented. The idea of calculating scoring functions seems to be a bright one as it helps in them in many sort of ways. As a reader, the way of writing kept me hooked to the technical details performed as it was easy to understand and presented in a coherent manner. The authors have done a great job in presenting their research work with fine details required for novel readers to understand properly. For the future work, these ground-truth communities can help in fresh ideas for community detection.

More about abc

Open Document