Improving the performance of algorithms to find communities in networks

Darst, Richard K.; Nussinov, Zohar; Fortunato, Santo

doi:10.1103/physreve.89.032809

Cited by 20 publications

(12 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…An interesting future problem is to extend the reluctant backtracking approach to reliably detect more than two communities. Determining the number of communities in a network is a problem by itself and knowing the number of communities in a network can improve the performance of community detection methods 26 . Krzakala et al 20 suggested a heuristic to determine the number of communities in a given network when using the non-backtracking matrix B .…”

Section: Discussionmentioning

confidence: 99%

“…However, because of the approximations involved, the heuristic can fail for real 20 and simulated networks 26 , by predicting too many real-valued eigenvalues outside the bulk and thus predicting too many modules. The optimisation of modularity Q by the second eigenvector of both the flow F and normalised reluctant-backtracker P matrices suggests two further solutions for finding more than two communities.…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

Finding communities in sparse networks

Singh

Humphries

2015

Sci Rep

View full text Add to dashboard Cite

Spectral algorithms based on matrix representations of networks are often used to detect communities, but classic spectral methods based on the adjacency matrix and its variants fail in sparse networks. New spectral methods based on non-backtracking random walks have recently been introduced that successfully detect communities in many sparse networks. However, the spectrum of non-backtracking random walks ignores hanging trees in networks that can contain information about their community structure. We introduce the reluctant backtracking operators that explicitly account for hanging trees as they admit a small probability of returning to the immediately previous node, unlike the non-backtracking operators that forbid an immediate return. We show that the reluctant backtracking operators can detect communities in certain sparse networks where the non-backtracking operators cannot, while performing comparably on benchmark stochastic block model networks and real world networks. We also show that the spectrum of the reluctant backtracking operator approximately optimises the standard modularity function. Interestingly, for this family of non- and reluctant-backtracking operators the main determinant of performance on real-world networks is whether or not they are normalised to conserve probability at each node.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Discussionmentioning

confidence: 99%

Finding communities in sparse networks

Singh

Humphries

2015

Sci Rep

View full text Add to dashboard Cite

show abstract

“…Here we offer an alternative to this problem by using an effective heuristic. We first use an assistant community detection method, such as the spectral method 28 suggested by Darst et al 42 , or the widely used though often criticised modularity optimization method 29 to determine an approximate number of communities c s . Thereafter, we decrease k starting from c s until ρ k < ρ k + 1 and set c d = k + 1, and then increase k starting from c s until ρ k < ρ k − 1 and set c u = k − 1.…”

Section: Methodsmentioning

confidence: 99%

Combined node and link partitions method for finding overlapping communities in complex networks

Jin

Gabrys

Dang

2015

Sci Rep

View full text Add to dashboard Cite

Community detection in complex networks is a fundamental data analysis task in various domains, and how to effectively find overlapping communities in real applications is still a challenge. In this work, we propose a new unified model and method for finding the best overlapping communities on the basis of the associated node and link partitions derived from the same framework. Specifically, we first describe a unified model that accommodates node and link communities (partitions) together, and then present a nonnegative matrix factorization method to learn the parameters of the model. Thereafter, we infer the overlapping communities based on the derived node and link communities, i.e., determine each overlapped community between the corresponding node and link community with a greedy optimization of a local community function conductance. Finally, we introduce a model selection method based on consensus clustering to determine the number of communities. We have evaluated our method on both synthetic and real-world networks with ground-truths, and compared it with seven state-of-the-art methods. The experimental results demonstrate the superior performance of our method over the competing ones in detecting overlapping communities for all analysed data sets. Improved performance is particularly pronounced in cases of more complicated networked community structures.

show abstract

“…Fitness function f s can measure the internal and external tightness for a local community. In topology detecting community methods, it has shown a good performance, and was expanded or applied by more researchers [23,24]. Inspired by this thought, we propose the fitness function for a hyperedge e to be given as…”

Section: Similarity For Hyperedge Pairsmentioning

confidence: 99%

Detecting Topic-oriented Overlapping Community Using Hybrid Hypergraph Model

Shen¹,

Yang

Sun

2016

INT J COMPUT COMMUN

View full text Add to dashboard Cite

A large number of emerging information networks brings new challenges to the overlapping community detection. The meaningful community should be topicoriented. However, the topology-based methods only reflect the strength of connection, but ignore the consistency of the topics. This paper explores a topic-oriented overlapping community detection method for information work. The method utilizes a hybrid hypergraph model to combine the node content and structure information naturally. Two connections for hyperedge pair, including real connection and virtual connection are defined. A novel hyperedge pair similarity measure is proposed by combining linearly extended common neighbors metric for real connection and incremental fitness for virtual connection. Extensive experiments on two real-world datasets validate our proposed method outperforms other baseline algorithms.

show abstract

Improving the performance of algorithms to find communities in networks

Cited by 20 publications

References 28 publications

Finding communities in sparse networks

Finding communities in sparse networks

Combined node and link partitions method for finding overlapping communities in complex networks

Detecting Topic-oriented Overlapping Community Using Hybrid Hypergraph Model

Contact Info

Product

Resources

About