Research And Application Of Consensus Clustering-based Community Detection Method For Multilayer Networks

In the real world,there are multiple connections or interactions between things at different levels and dimensions,and the traditional single layer network models are often unable to describe these multi-level and multi-dimensional connections.In contrast,multilayer networks have higher flexibility and expressiveness,and can reflect multiple different levels and dimensions of information at the same time,which can better depict the complex connections and change patterns among things in the real world.Therefore,multilayer networks have received more and more attention from scholars and have been widely applied in various disciplines.Community detection is an important task in multilayer network analysis.Its goal is to group highly relevant nodes in the network into a community,so that the nodes in the community have strong correlation and the nodes between different communities have weak correlation.The traditional single layer network community detection algorithm only needs to consider the node association on a single layer,but the nodes in a multilayer network have different associations and interactions at different layers.When conducting community detection,we need to consider the node association of each layer comprehensively.In addition,the inherent complexity of a multilayer network makes community detection on a multilayer network need to consume more computing resources,Therefore,there are many challenges in multilayer network community detection.In order to solve the community detection problem in multilayer networks,many new algorithms have emerged in recent years.Among them,some algorithms aggregate multilayer networks into a single layer network and then apply the single layer network community detection algorithm on the aggregated network.However,directly aggregating multilayer networks into a single layer network destroys the structure of multilayer networks and may lead to severe information loss,and this approach is only applicable to multilayer networks with small inter-layer differences.In addition,there is also a method of community partition by first using single layer network community detection for each layer of a multilayer network,and then aggregating the community set of different network layers to obtain the final community segmentation results,however,this method requires community detection for each layer,which has high computational complexity,and the community attributes of nodes are diverse in the case of many network layers,which easily affects the accuracy of the final results.In this regard,this paper proposes a consensus clustering-based community detection method for multilayer networks.The method first applies a single-layer network community detection algorithm to perform community detection for each layer of the multilayer network separately,and further explores the community structure characteristics of each layer of the network.Then consensus clustering is used to construct a consensus matrix based on the community division structure of each layer of the network,and then community detection is performed again on the consensus matrix to obtain the final community division results.Since the constructed consensus matrix usually becomes very dense and poses a challenge to the accuracy of the algorithm,this paper proposes a consensus matrix filtering threshold calculation method based on the network importance index to improve the accuracy of consensus clustering.In addition,for the low accuracy of the multilayer network community detection algorithm based on consensus clustering when there are more network layers,this paper proposes an improvement of the algorithm using a network reduction strategy,which improves the accuracy of the algorithm when there are more network layers.Since multilayer network community detection has a wide range of applications in various fields,for example,in the field of social network analysis for inferring users’ interpersonal relationships and interests,in the field of biomedicine for analyzing the complex regulatory relationships between proteins and genes,and in the field of ecology,multilayer network community detection is also used to identify the interactions between different populations on habitat utilization,etc.Therefore,this paper details how the multilayer network community detection method based on consensus clustering proposed in this paper can be used in concrete practice through a practical application case in the ecological field,with the aim of promoting the use of this method in practical applications.Specifically,the main work of this paper is as follows.(1)A consensus clustering-based multilayer network community detection algorithm(CCMCD algorithm)that uses a consensus matrix of network importance indicators to calculate filtering thresholds is proposed,and its community detection accuracy is verified by designing experiments and an artificial dataset based on the LFR benchmark.(2)The improvement schemes of CCMCD algorithm based on network von Neumann entropy-based multilayer network reduction strategy and network dissimilarity-based multilayer network reduction strategy are proposed respectively,and the community detection accuracy of CCMCD algorithm based on these two multilayer network reduction schemes with different number of layers of multilayer network reduction methods is verified by design experiments and artificial data sets based on LFR benchmark..(3)Through a practical application case in ecological domain,we elaborate how to use the consensus clustering-based multilayer network community detection algorithm proposed in this paper in concrete practice,and discuss the feasibility and effectiveness of the method in practical application.