Elsevier

Information Sciences

Volume 369, 10 November 2016, Pages 402-418
Information Sciences

Multi-objective optimization of community detection using discrete teaching–learning-based optimization with decomposition

https://doi.org/10.1016/j.ins.2016.06.025Get rights and content

Abstract

Community detection has been an active field of study in complex network analysis in recent years. It can be modeled as a seriously nonlinear optimization problem. Many intelligent optimization techniques have shown promising results for this problem. The teaching–learning-based optimization (TLBO) algorithm is a recently proposed swarm intelligent algorithm. In this paper, a discrete variant of TLBO (DTLBO) is proposed to address discrete optimization problems. In the proposed method, the learner representation scheme is redefined, and the updating rules for learners are also redesigned. Moreover, based on the proposed discrete variant DTLBO, a multi-objective discrete method (MODTLBO/D) is proposed to solve community detection problems for complex networks. The multi-objective decomposition mechanism is adopted and neighbor-based mutation is introduced to maintain the diversity of the population and avoid being trapped in the local optima. Finally, to verify the performance of the proposed algorithm, real-world networks are examined. The experimental results indicate that MODTLBO/D is effective compared with other algorithms used for community detection in complex networks.

Introduction

Real-world networks involve many important and complex interactions among individuals [3], for example, social networks [46], biological networks [6], biochemical networks [13], and communication networks [22]. These networks can be naturally represented as complex networks. In mathematics and computer science, complex networks are often modeled as graphs. More specifically, a complex network consists of nodes (or vertices) and edges (or links), which represent the individual members and their relationships. Generally, the cluster or community is defined as a subset of vertices that are closely knit in a relatively sparse neighborhood. Hence, the cluster or community is composed of individuals with common properties inherent to complex networks [48]. Community detection or network clustering is essential in complex networks; it is one of the most relevant features of networks that represent real systems and a key aspect of network analysis, the purpose of which is to discover complex network topology structures and understand complex network functions [16]. Community detection or the network clustering problem is typically viewed as the division of networks into subsets of vertices according to their similarity, other characteristics related to the nodes and edges of the graph, or both. Naturally, it can be considered an optimization problem [2], [38].

The teaching–learning-based optimization (TLBO) algorithm [43] is a population-based algorithm. The algorithm requires common control parameters rather than algorithm-specific control parameters. Because TLBO is characterized by simple computation and rapid convergence, it is used in many optimization fields. Although the TLBO algorithm and its variants have achieved effective results in continuous optimization, little progress has been made for discrete TLBO variants. To solve job shop scheduling problems (JSSP), Keesari and Rao used a learner in the TLBO algorithm to represent an operation scheduling list in JSSP, where each dimension of a learner represents one operation of a job [27]. Xia et al presented a simplified teaching–learning-based optimization (STLBO) algorithm for solving disassembly sequence planning problems (DSP) with discrete solution spaces [50]. In this method, a Feasible Solution Generator (FSG) is used to generate a feasible disassembly sequence, and a Teaching Phase Operator (TPO) and a Learning Phase Operator (LPO) are used to learn and evolve the solutions towards better solutions by applying the method of precedence preservation crossover operation. Baykasoglu et al analyze the performance of the TLBO algorithm with respect to flow shop (FSSP) and job shop scheduling problems (JSSP) [5]. In the proposed method, a random key-based approach with the largest position value rule is used to obtain a job permutation for FSSP and JSSP, and G&T algorithm is used for the construction of active schedules within the TLBO search. Xu et al. proposed an effective teaching–learning-based optimization algorithm for the flexible job-shop scheduling problem with fuzzy processing time (FJSPF) [51]. In this method, a solution is expressed by an operations sequence vector and a machine assignment vector, and a decoding method is employed to transfer a solution to a feasible schedule in the fuzzy sense. Then, a bi-phase crossover scheme based on the teaching–learning mechanism and special local search operators are incorporated into the search framework of the TLBO to balance the exploration and exploitation capabilities. Dede utilized TLBO for the discrete optimization of truss structures [12]. Li et al. proposed a discrete teaching-learning-based optimization for solving the flow shop scheduling problem (FSSP) [29]. In the proposed DTLBO algorithm, four discretization operators are introduced into the teacher phase and learner phase and a modified iterated greedy (IG)-based local search is embedded to enhance the search ability of the proposed algorithm.

In essence, community detection in a complex network is a clustering optimization problem. For this reason, evolutionary computation and swarm intelligence can be used for the community detection or network clustering problem (Seen in Section 2.1). Compared with traditional algorithms, intelligent optimization algorithms can effectively find a proper, high-quality solution within a reasonable period of time. This paper attempts to apply the multi-objective discrete teaching–learning-based optimization algorithm based on decomposition (MODTLBO/D) to address communities in complex networks. First, unlike the above method, we proposed a novel discrete TLBO for community detection problems. In the proposed DTLBO, the problem-specific locus-based adjacency representation strategy is used to initialize a population with N learners in which each learner represents the cluster values of all nodes; the difference between the Teacher (the best learner in the current generation) and NMean (the mean solution of the local learners) of the learner's neighbors in the Teacher Phase and the difference between the learner and another learner, which is randomly selected from the class in the Learner Phase, are updated using the XOR operator; the niching mechanism is applied to update the new learner according to the learner and the difference; and the mutation operation is implemented after both the Teacher Phase and Learner Phase at each iteration instead of the duplicate elimination in the original TLBO. The features of complex networks are taken into account, and DTLBO is extended to MODTLBO/D, which is a multi-objective optimization variant with a decomposition strategy. Experimental results and relevant comparative analyses show that the proposed algorithm achieves high-quality detection results.

The remainder of this paper is organized as follows. Related works are introduced in Section 2. Section 3 describes the proposed method for community detection. Section 4 presents tests performed on real-world complex networks, and the experiments are conducted along with statistical tests. Conclusions are given in Section 5.

Section snippets

Literature review

In recent years, community detection in complex networks has attracted an increasing amount of attention, and several popular algorithms for community detection have been presented. Most community mining algorithms can be classified into several categories:

Graph partitioning-based methods: In these methods, the network is divided the vertices into a number of groups with pre-defined size so that the number of edges between groups is minimized. The Kernighan-Lin algorithm [14] and Spectral

Teaching–learning-based optimization

Rao et al. [43] first proposed a novel teaching–learning-based optimization (TLBO) inspired from the philosophy of teaching and learning. TLBO mimics the teaching and learning process of a typical class. In optimization algorithms, the population consists of different design variables. For the TLBO algorithm, a group of learners or a class of learners is considered the population. The teacher and learners are vital components of the TLBO algorithm. The output of the TLBO algorithm is considered

The DTLBO and MODTLBO/D algorithm

In this section, the proposed MODTLBO/D method for community detection problems is described. First, the motivation for proposing the discrete TLBO (DTLBO) algorithm for complex community detection problems is given. Next, the DTLBO algorithm is described in detail. Then, based on the proposed discrete DTLBO variant, a novel multi-objective discrete method (MODTLBO/D) is proposed. The following sections describe the updating strategy and implementation process of the proposed algorithm.

Application of MODTLBO/D to community detection

We used real-world datasets to test the efficiency of MODTLBO/D. To evaluate the performance of MODTLBO/D, we also simulated the representative single-objective algorithms (GA [44], FM [9], BGLL [7]) and multi-objective algorithms (MOCD [44], MOGA-net [40], MOEA/D-net [21], MODPSO [19]) and then presented a comparison of their results.

Conclusion

One of the simplest and most efficient techniques, TLBO has been empirically shown to perform well on many optimization problems. In this paper, a discrete variant of the teaching–learning-based optimization algorithm (DTLBO) is first proposed. In this discrete TLBO framework, the learner vectors and their updating operators have been redefined. Moreover, based on the proposed discrete variant DTLBO, a novel multi-objective discrete optimization algorithm is first presented to solve the

Acknowledgment

This work is partially supported by the National Natural Science Foundation of China (Grant Nos. 61572224, 61304082) and the National Science Fund for Distinguished Young Scholars (Grants No.61425009). This work is also partially supported by the Major Project of Natural Science Research in Anhui Province (Grant No.KJ2015ZD36) and the Natural Science Foundation in colleges and universities of Anhui Province (Grant No.KJ2016A639).

References (58)

  • LiJ. et al.

    A discrete teaching-learning-based optimisation algorithm for realistic flowshop rescheduling problems

    Eng. Appl. Artif. Intell.

    (2015)
  • Li.W.

    Visualizing network communities with a semi-definite programming method

    Inf. Sci.

    (2015)
  • LiZ. et al.

    A multi-agent genetic algorithm for community detection in complex networks

    Physica A

    (2016)
  • MaL. et al.

    Multi-level learning based memetic algorithm for community detection

    Appl. Soft Comput.

    (2014)
  • R.V. Rao et al.

    Teaching–learning-based optimization: an optimization method for continuous non-linear large scale problems

    Inf. Sci.

    (2012)
  • ShiC. et al.

    Multi-objective community detection in complex networks

    Appl. Soft Comput.

    (2012)
  • XiaK. et al.

    Disassembly sequence planning using a simplified teaching–learning-based optimization algorithm

    Adv. Eng. Inf.

    (2014)
  • XuY. et al.

    An effective teaching–learning-based optimization algorithm for the flexible job-shop scheduling problem with fuzzy processing time

    Neurocomputing

    (2015)
  • YuK. et al.

    Constrained optimization based on improved teaching–learning-based optimization algorithm

    Inf. Sci.

    (2016)
  • ZhouX. et al.

    Multiobjective biogeography based optimization algorithm with decomposition for community detection in dynamic networks

    Physica A

    (2015)
  • A.L. Barabási. Scale-free networks: a decade and beyond. Science, 2009,...
  • L. Bennett et al.

    Detection of composite communities in multiplex biological networks

    Sci. Rep.

    (2015)
  • V.D Blondel et al.

    Fast unfolding of communities in large networks

    J. Stat. Mech

    (2008)
  • A. Clauset et al.

    Finding community structure in very large networks

    Phys. Rev. E Stat. Nonlinear Soft Matter Phys.

    (2004)
  • J.D. Cruz et al.

    Entropy based community detection in augmented social networks

  • L. Danon et al.

    Comparing community structure identification

    J. Stat. Mech

    (2005)
  • T. Dede

    Application of Teaching-Learning-Based-Optimization algorithm for the discrete optimization of truss structures

    KSCE J. Civ. Eng.

    (2014)
  • M. Fiedler

    A property of eigenvectors of nonnegative symmetric matrices and its applications to graph theory

    Czechoslovak Math J.

    (1975)
  • M. Girvan et al.

    Community structure in social and biological networks

    Proc. Natl. Acad. Sci.

    (2002)
  • Cited by (47)

    • A multiobjective Cuckoo Search Algorithm for community detection in social networks

      2022, Multi-Objective Combinatorial Optimization Problems and Solution Methods
    View all citing articles on Scopus
    View full text