Local Community Detection Based on Small Cliques

Abstract

Community detection aims to find dense subgraphs in a network. We consider the problem of finding a community locally around a seed node both in unweighted and weighted networks. This is a faster alternative to algorithms that detect communities that cover the whole network when actually only a single community is required. Further, many overlapping community detection algorithms use local community detection algorithms as basic building block. We provide a broad comparison of different existing strategies of expanding a seed node greedily into a community. For this, we conduct an extensive experimental evaluation both on synthetic benchmark graphs as well as real world networks. We show that results both on synthetic as well as real-world networks can be significantly improved by starting from the largest clique in the neighborhood of the seed node. Further, our experiments indicate that algorithms using scores based on triangles outperform other algorithms in most cases. We provide theoretical descriptions as well as open source implementations of all algorithms used.

Local Community Detection: A Survey

“…The proposed method is applied on both unweighted and weighted networks. Extensive experiments with synthetic and real networks (social, collaboration, biological) prove the accuracy of the framework to detect local communities compared with local methods like [46], [127], [36] and a global one [108], considering precision, recall and F1 score. The largest dataset that they experiment with is Friendster [113] with 65608366 nodes.…”

“…Node(s) Bagrow J. et al [30],Xu B. et al [12],Zhang T. et al [31],Chen Q. Et al [32],Moradi F. et al [33],Xia S. et al [10],Fanrong M. et al [34],Wang P. et al [35],Hamann M. et al [36],Ding X. et al [37],Tasgin M. et al [38],Xu X. Et al [39],Guo K. et al [8],Luo W. et al [40],Liu J. et al [16],Aghaalizadeh et al [41],Hu Y. et al [42],Ji P. et al [43] Greedy Focus on Quality metric Clauset A.…”

“…[17],Luo F. et al [44],Chen J. et al [45],Huang J. et al [46],Branting L. et al [47],Wu Y. et al [48],Ngonmang B. et al [49],Kudelka M. et al [50],Ma L. et al [51],Cui W. et al [52] are greedy and non-greedy. Regarding greedy algorithms, these methods focus on the technique used to add nodes: A node is selected for inclusion in a community if and only if that node has the maximum value of a quality function that assigns a value to each node [36]. The majority of existing algorithms belong to this category.…”

“…Hamann M. et al [36] also suggest to start the expansion of the community from the largest clique in the neighbourhood of the seed node. Besides, they introduce Triangle-based Community Expansion (TCE) as an alternative strategy for greedy community expansion.…”

“…Many methods, useful for community detection, are implemented and executed in optimal time, such as Louvain, Static networks/ Greedy/ Focus on starting Node(s) Reference Algorithm Key features Source Code Bagrow J. et al [30] use of l − shell to find the starting nodes, parameter dependent -Xu B. et al [12] use of k top leader nodes as starting nodes, community size is parameter dependent Zhang T. et al [31] use nodes with maximum degree as starting ones, low accuracy -Chen Q. Et al [32] LMD local degree central node as seed -Moradi F. et al [33] similarity score to define starting nodes, parameter-free -Xia S. et al [10] ILCDSP uses selection probability to find starting nodes, not very stable -Fanrong M. et al [34] LCD-MC start LCD from maximal clique containing the seed node -Wang P. et al [35] MAGA-LC tightest nodes as seeds -Hamann M. et al [36] start LCD from maximal clique containing the seed node, TCE method, unweighted and weighted networks, overlapping communities Provided Ding X. et al [37] RTLCD use core node as starting one -Tasgin M. et al [38] use of highest score node to start the label propagation method, instability in determining the final communities, weak performance in identifying a reasonable number of communities Provided Xu X. Et al [39] DLCD-CCE expands community centers, can run in a distributed environment, weighted networks -Guo K. et al [8] InfoNode local degree central nodes and Jaccard coefficient to detect starting nodes, overlapping communities, parameter dependent, more than one seeds -Luo W. et al [40] LCDNN nearest nodes with greater centrality as seeds, multiscale local communities -Liu J. et al [16] HqsMLCD multiple communities, higher quality starting nodes -Aghaalizadeh et al [41] use core nodes as starting, weighted networks -Hu Y. et al [42] S-LM local centrality measurements used for selecting starting nodes -Ji P. et al [43] CAELCD finds core nodes as seeds and expand them -PageRank Nibble and others.…”