Understanding the Importance of Community Detection Method

Business // Understanding the Importance of Community Detection Method

Understanding the Importance of Community Detection Method

Author’s Name

Institutional Affiliation

Table of Contents

Title Page ……………………………………………………………………………………………………            1

Table of Contents ………………………………………………………………………………………..            2

Initial Thoughts …………………………………………………………………………………………..            3

Community Structures …………………………………………………………………………………            4

Composite Community Structures …………………………………………………………………    5

Use of Divisive Procedures ……………………………………………………………………………   7

Broad Perspectives for Community Detection …………………………………………………           7

The Cut-Based Perspective ………………………………………………………………….   9

The Clustering Perspective ………………………………………………………………….   10

The Stochastic Block Model Perspective ………………………………………………   10

The Dynamical Perspective …………………………………………………………………    11

Community Detection as a Nonspecific Tool …………………………………………………..         11

Performance of these Community Detection Algorithms ………………………………….      14

Statistical Inference Model ……………………………………………………………………………   16

Final Thoughts …………………………………………………………………………………………….            17

References …………………………………………………………………………………………………..            18

Understanding the Importance of Community Detection Method

Initial Thoughts

The underlying concept of communities allows researchers and experts to grasp the technical meaning of the term “community.” To define, a community as a technical term represents a group of individuals that interact with each other in a network. In a vast scale map of networks, communities operate within their scales or even outside the groups. Groups of professionals, such as information technology experts and computer scientists, recognize the complex structures of communities. To generate a giant scale map of networks, information technology experts and computer science professionals distinguish the importance of the community structures including their individual communities. These individual communities make the study of certain meta-nodes easier and simpler to understand (Shuo & Chai, 2016, p. 122). Additionally, the community structures serve as the most widespread possessions in multifarious systems. Information technology experts and computer scientists from various scientific divisions and subdivisions must realize the wide-ranging tools to use to develop the community detection procedures (Shuo & Chai, 2016, p. 124). The question of how to detect a certain group of communities is something that experts need to do. Using a community detection method, information technology connoisseurs and computer technologists can easily find groups of communities within the complex nets. Since the detection of the communities within the large maps of networks takes time and effort, a novel community detection method is necessary to exceed the traditional means approaches in terms of accuracy and firmness while endorsing additional computational costs. This method identifies communities that interact with each other. Despite the challenge to identify communities within the sets of nodes of certain networks, extensive experiments and other studies conducted on the computer-generated networks and real-world datasets will only exemplify agreeable presentations and results of the presented algorithm in comparison with other customary community detection processes.

Community Structures

The contemporary science of systems makes sizeable and substantial progress to the understanding of composite structures. The germane structures of graphs, which signifies real systems, is “community structure” (Fortunato, 2010, p. 78). In a number of sources, this community structure is grouping, in which it represents an organization of vertices in groups, with numerous boundaries linking vertices of the similar group and reasonably few boundaries linking vertices of diverse groups. Such communities are objectively unrestricted sections of a graph with a comparable character of human organs and tissues in the human body. Of course, detecting groups remain an important task in various fields of disciplines, such as biology, computer science, and sociology, where graphs represent the systems (Fortunato, 2010, p. 79). It even becomes a big problem to solve. In spite of the enormous effort of the interdisciplinary group of scientists who work this problem for decades, this paper tries to explore the topic to bring the light of the central rudiments of the problem.

Figure 1. Communities in Complex Real-World Networks

The study of complex networks fascinates biology, social networks, epidemiology, computer science, and transportations experts. These multifaceted systems kindle more logical and systematic advances in numerous grounds. Several articles from different authors and experts explore dissimilar properties in complex networks and identify the prime problems in the varying concepts (Rosvall, Delvenne, Schaub, & Lambiotte, 2017, p. 1). Information technology experts and computer scientists believe that learning on community structure is extensive to determine the groups in the given networks. They need to discover the comparatively thick connections of the groups by distinguishing the divisions of nodes in networks. It means that web grouping exposes the fundamental dynamics, mechanisms, and semantic structures of certain systems.

Composite Community Structures

A straightforward demarcation of communities in intricate networks proves to be composite. In this case, the information technology experts, computer scientists, and network technologists have community detection algorithms on certain networks with a specific procedure of community structure based on the measured and scientific techniques to employ. Nonetheless, this assessment seems deceptive since apparent parallels in their mathematical mechanism tend to camouflage unrelated whys and wherefores. Information technology connoisseurs, computer scientists, and network technologists work with community detection. Different urges must reinforce community detection since this problem-driven plan is appropriate in truthful network knowledge. It is important to select an appropriate procedure for the assumed drive. Moreover, underlining the diverse approaches to community detection allocates the many appearances of research and uncluttered guidelines for forthcoming research. As a result, an open separation of communities in difficult networks proves to be complex.

Information technology mavens, computer scholars, and network technologists, as well as the Internet Service Providers (ISPs), can set up a mirror server for focused web visits from the similar topographical areas to advance clients’ Internet surfing involvements. This mirror server helps these groups of professionals comprehend efficient recommendations to clienteles on the side of crafting a friendlier environment. On top, information technology experts and computer scientists wish to introduce a first-hand community detection method emphasizing the differential evolution process. Such a community detection method focuses on the modularity density of the communities to advance its function and parameter to analyze the diverse communities in the network. The self-possessed concerns on the computer-generated and real-life networks reveal that the anticipated algorithms offer an active method in determining community structures. To address the community detection problem, information technology experts, computer scientists, and researchers must find ways to detect these groups within the networks using the numerous advanced measures. Social network academics and engineers, for instance, will find certain groups of users and web visitors to access the different social networks such as Facebook, MySpace, LinkedIn, and the like and can solve this community detection problem by ordinary approaches such as “graph partitioning, hierarchical clustering, partitional clustering, and spectral clustering” (Shuo & Chai, 2016, p. 124). They can find those groups of friends who access their social media platforms. Moreover, they can determine the groups of friends who access within days, weeks, and months and those who groups who post pictures, who surf the web pages, and who tag friends with comments and messages. Using the community detection method, groups of professionals, as well as the Internet Service Providers (ISPs), can concentrate on the users’ use of the social network sites from the parallel geographical portions to develop clients’ Internet surfing systems and to determine the communities in the group.

Use of Divisive Procedures

One way to determine the community structures is the utilization of the divisive measures. This process breaks connections of betweenness as philosophies to appraise the edge links to normalize the maximum notable community structure (Bickel & Chen, 2009, p. 21068). Girvan’s and Newman’s algorithm, labeled modularity, points out the quality function as a benchmark to choose when to discontinue the calculation of finding communities. This algorithm facilitates a landmark that facilitates the grouping ways to think of the problem of modularity optimization (Bickel & Chen, 2009, p. 21069). Even if these groups of professionals know the practicable partitions to know the paramount approach, information technology mavens, computer scholars, and network engineers use some “fairly accurate optimization techniques, comprising greedy algorithms, random walks, fast unfolding algorithm, information-theoretic framework, belief propagation, simulated strengthening, and genomic procedures to resolve the challenge” (Wang, Zhang, Chai, & Xia, 2018, p. 1). For instance, the spectral algorithms use traditional clustering techniques. Information technology mavens, computer scholars, and network scientists will use “the correspondence matrix factorization” and “the blocking system” to define the community configurations (Shuo & Chai, 2016, p. 124). Perhaps, these groups of professionals can use the label propagation technique in which they attach labels or names to the nodes based on the operative clustering method.

Broad Perspectives for Community Detection

Community detection represents an umbrella term with numerous aspects and features. Information technology mavens, computer scholars, and network scientists understand these surfaces, which materialize from the different objectives. These groups of professionals want to appreciate the diverse perspectives on how to verbalize the problem of community detection. Experts must be aware of the vital motivations to select and equate community detection methods (Bickel & Chen, 2009, p. 21070). Instead of focusing on the technical details of diverse algorithmic implementation, groups of professionals can stress the theoretical alterations between the dissimilar views on community detection and emphasize the standpoints to determine these groups within the networks. This paper does not discuss the idea that the different viewpoints are unrelated but that the mathematical problem constructions will lead to the creation of comparable procedures. Perhaps, they can determine the undirected networks and the function modularity from a grouping point of view. In some circumstances, the famous associations will dissolve. Even if some experts argue that their methods work better than them, this paper does not assert that this particular standpoint works best at the given network. In terms of data clustering, experts think that no method can consistently achieve all categories of systems. Unquestionably, the idea of community detection remains an unsubstantiated learning task. None can detect that certain procedures to determine groups and communities operate. All processes remain unidentified by the experts and researchers upon doing the analysis. For that reason, information technology mavens, computer scholars, and network scientists can recognize the usefulness and relevance of the particular methods. As experts disclose the dissimilar purposes behind community detection, information technology mavens, computer scholars, and network scientists work best within the compactness and scarce internal connections. Furthermore, four broad perspectives exist to serve as motivation for community detection a) the cut-based perspective, b) the clustering perspective, c) the stochastic block model perspective, and d) the dynamical perspective. These viewpoints can operate within the contexts of the community detection processes. For the cut-based perspective, this method reduces the restriction of links between clusters of nodes. For the clustering perspective, certain information technology pundits, computer scholars, and network scientists can take full advantage of the internal density in collections of nodes. When it comes to the stochastic block model perspective, these professionals can identify stochastically corresponding groups of nodes. Finally, for the dynamical perspective, these professionals can identify groups of nodes to form extended building blocks of dynamics on networks. Of course, this classification is not exceptional. It can help illuminate models and theories about community detection to determine the suitable technique to uncover a specific purpose.

Figure 2. Four Broad Perspectives for Community Detection

The Cut-Based Perspective

The use of the cut-based perspective suggests a network partitioning application. In fact, information technology mavens, computer scholars, and network scientists initially use this network partitioning application to form a circuit layout and design. This application encourages modification to propose the practice of eigenvectors for network partitioning. The prime goal of these experts to describe the signal flows and to design a circuit in a well-organized way is to divide the network into a permanent figure of roughly correspondingly sized groups. Information technology mavens, computer scholars, and network scientists consider these communities for the well-adjusted load with a small number of edges. Experts can illustrate the edges, which run between clusters, like a cut. Moreover, these professionals can design the most efficient circuit within the networks if they learn to minimalize the cut with comparatively balanced clusters. In the end, the practice of the cut-based perspective recommends a system partitioning presentation to detect communities in the complex networks.

The Clustering Perspective

In the framework of data clustering, certain motivations for community detection ascend. For instance, these information technology mavens, computer scholars, and network scientists use the clustering viewpoint to advocate the high-dimensional space. The objective is to partition the points into a number of clusters so that communities come closer to one another. It will mean that the points in different groups will be aloof from one another. To achieve the goal of this clustering perspective, these information technology mavens, computer scholars, and network scientists can construct proximity between and among these groups in the networks. Of course, this tactic will lead to a community detection problem. The nearness between these nodes typifies the display and weightiness of the limits of the communities between them.

The Stochastic Block Model Perspective

In Stochastic Block Model Perspective, these information technology mavens, computer scholars, and network scientists will group similar nodes, which also link to the similar nodes within the public. They will find the assortative group structure in which the tiered clusters will group with clusters of clusters. However, the manifestation of the assortative structural groups remains too precautionary to choose groups based on the common connectivity patterns. The disassortative groups have insubstantial relations within the clusters rather than between the groups. These information technology pundits, computer intellectuals, and network geniuses who perform the social network analysis can find a collective objective to recognize those nodes within a network system. These nodes with comparable organizational features in terms of connectivity profile share parallel connection patterns to other nodes. It means that nodes are recurrently comparable if they are correspondingly connected to equivalent others. However, experts must realize that the algorithms used in classifying clusters of “roughly equal” nodes divulge block structures in the block models.

The Dynamical Perspective

The dynamic perspective as a substitute drive for community detection problem concentrates on the courses to take place on the system. Experts know that these ideas of communities are successfully mechanical and that these concepts work with the opus of the system. Conversely, one core reason to apply tools from network science is to realize the behavior of a system. However, the network topology cannot clarify the system behavior. Rather than finding a coarse-grained report of the adjacency matrix, these information technology pundits, computer intellectuals, and network geniuses may find a coarse-grained account of the dynamics and undercurrents performing on top of the system with diverse tracks beyond the contiguous groups.

Community Detection as a Nonspecific Tool

The discussion of community detection requires further analysis and review. Through a range of dissimilar lenses, community detection as a nonspecific tool that works in a generic framework considers associating dissimilar approaches. The diverse viewpoints of a real-world network spawned by a probably multifaceted arbitrary task of edges underline the specific undercurrents on this network called epidemic spreading (Stegehuis, van der Hofstad, & van Leeuwaarden, 2016, p. 29748). Of course, with enough freedom of the Internet, interrelating and partaking awareness empower a newfangled technique of overwhelming, learning, cooperating, and making associations, unchecking a huge quantity of data to scrutinize to acquire a restored understanding of human conduct. In terms of social network sites, people use data to construct systems of associations between communities (Stegehuis, van der Hofstad, & van Leeuwaarden, 2016, p. 29748). For instance, business leaders hire IT who can determine the networks between and among friends on Facebook, to buy products on Amazon, and the like. The appointed IT experts who find communities in the system to categorize enormously related groups of individuals in these networks (Kernix, 2016). These groups called communities, which have similar motives and interests, form groups in a network. The use of the community detection method helps business directors learn a brand that grasps the dissimilar clusters of outlook toward the products and the target groups and influencers. Business leaders can use this method to help an e-commerce website build an approval system. Finally, business leaders like experts can learn about community detection method and problem in details.

Some of the information technology specialists, computer intellects, and network masterminds work with the principal algorithms for community detection. They also assess their performance to construct a yardstick contingent on the field of disciplines. In fact, diverse types of networks can cause variations in the algorithms’ performance and attempt to comprehend the elements of the structures of communities within the networks (Stegehuis, van der Hofstad, & van Leeuwaarden, 2016, p. 29748). However, these structural elements touch the performance of the algorithms. Over the years, the dissimilar approaches appear to disclose communities professionally in composite networks. Information technology specialists, computer intellects, and network masterminds make use of certain measures or modularity in the network. These measures nearly are correspondent to take full advantage of the number of edges or relationships inside the groups and minimalize the number of edges between the groups. For example, “the greedy algorithm on modularity comes from Newman in 2004 that eliminates edges or relationships to be between groups from the system” (Peel, Larremore, & Clauset, 2017, p. e1602548). To distinguish groups, information technology specialists, computer intellects, and network masterminds can make use of the simulating random walks inside the system. They think that random walkers have a tendency to stay inside heavily and thickly associated zones of the chart. This concept comes from the impression of Pascal Pons and Matthieu Latapy who introduce the inkling behind the walktrap algorithm. For years, Martin Rosvall develops another algorithm called infomap after the walktrap algorithm.

On the other hand, these algorithms can only envisage one community per node. This notion means that it is a wrong hypothesis to think that those algorithms can only anticipate one group per node. In some cases, experts know that those algorithms base their representations on the superimposing groups where nodes can belong to some groups. They even use the group percolation technique “as the most acknowledged way to distinguish intersecting communities” (Fortunato, 2010, p. 78). Based on the concept of the network, experts know that numerous groups and subgroups of nodes overlay. For example, one clique belongs to more than a few groups. Because of this discovery, Jaewon Yang and Jure Leskovec publicize their overlying community detection algorithm named BigClam, “a very scalable algorithm based on the non-negative matrix factorization” (Yang & Leskovec, 2013, p. 589). Yang and Leskovec assert that groups overlap due to dense communities (Yang & Leskovec, 2013, p. 589). It means that more groups that share nodes can connect themselves.

Figure 3. Infomap and Fastgreedy’s Results on the Twitter Graph

Figure 3 focuses on the visualizations of Infomap and Fastgreedy since this paper does not have “the ground-truth groups” (Peel, Larremore, & Clauset, 2017, p. e1602548). However, Figure 3 works well on the graph structure. The fundamental dissimilarity of this paper is that infomap crafts one big community in red but that fastgreedy breaks “this community into small groups” (Newman, 2004, p. 066133). As the graph grows in size, the behavior of the algorithms gets reversed. In this case, the fastgreedy progressively skirmishes to differentiate smaller groups whereas infomap enormously turns out to be more clear-cut.

Performance of these Community Detection Algorithms

Additionally, community detection in multifarious systems remains a challenging arena. Numerous procedures emerge to competently classify groups inside large nets. The recognized and reliable method to be used is infomap. Conversely, appropriately appraising the performance of these algorithms remains an issue, particularly in the case of intersecting groups within the network. Experts can only appreciate the feature of algorithms if they can test such a method on a graph with an incorporated group structure (Bedi & Sharma, 2016, p. 116). Possibly, they can complete a community detection on large graphs with semantic analysis to have a positive mechanism on the consequences. Of course, experts know that there can be no broad-spectrum purpose algorithm to serve all data types or applications. They accentuate that each perspective highlights a specific fundamental aspect. To recall, a cut-based method offers good partition of well-adjusted clusters. A clustering method affords forceful interconnection of the groups with great core density. The stochastic block model, for instance, delivers sturdy sameness of nodes inside the clusters based on the connectivity profiles. This method, which views communities as dynamical building blocks, is responsible “for the node groups to influence the dynamics of the group or groups in the same way” (Peel, Larremore, & Clauset, 2017, p. e1602548). Information technology authorities, computer experts, and network architects can accumulate wide-ranging types of data, which lead to a supplementary complex network structure. On the other hand, they want to reveal that the differences between the standpoints will turn out to be even more arresting since such similar network holds manifold and valid partitions. In addition, experts may lose their interests in dividing the nodes and in separating edges due to the complexity of the task to determine the groups within the network. People must have a thoughtful assessment of the network aspects when they implement a community detection method to integrate “the best technique and to understand complex networks” (Zachary, 1977, p. 463). Readers must know that the prime goal of the information technology specialists, computer intellects, and network masterminds is to decide the segmental community structure in multipart network with the data and statistics controlled in the chart topology configuration. This key network analysis remains positive and useful in the fields of biology, computer science, physics, and sociology. Of course, it seems very essential for people to realize “the features of a complex system” (Sun, 2016, p. 368). A workable community detection system finds a range of network assemblage to manage all system classifications. Experts contemplate that a gargantuan system can administrate the time complexity and space complexity. In addition, a community detection technique can pigeonhole the structure of the network based on statistical interpretation. The community detection method requires a theory and interpretation in order to meet the standard of the community detection algorithm (Rosvall, Delvenne, Schaub, & Lambiotte, 2017, p. 6). As a final point, experts need to elaborate on the future development of community detection method and its predictions to solve the problem based on statistical inference.

Statistical Inference Model

The use of statistical inference model explains the process of data analysis and the distribution. Accordingly, statistical inference is the method of using data analysis to infer properties of a fundamental possibility distribution. Experts even make use of the inferential statistical analysis that conjecture properties of a population. They can even do that, for example, “by testing theories and stemming calculations of the population from a larger population” (Gibbons & Chakraborti, 2011, p. 978). This idea falls under the inferential statistics that contrast with the descriptive statistics, solely concerned with properties of the observed data. Certainly, the use of the statistical inference model takes account of the vertex community (Gibbons & Chakraborti, 2011, p. 979). In fact, the information technology specialists, computer intellects, and network masterminds, then, point out the notions of the “Planted Partition Model, Newman’s Mixture Model, Mixed Membership Model, Mixed Membership Stochastic Block Model, and Degree-Corrected Stochastic Block Model” as models to determine the connections of the groups within the network (Shuo & Chai, 2016, p. 123). In the end, the use of statistical inference model, which clarifies the procedure of data investigation and the distribution of groups, makes community detection problem solve.

Final Thoughts

Certainly, information technology experts and computer scientists grasp that the community detection method offers an unpretentious technique to appreciate intricate structures of networks. The imbalance problem in community detection may appear in contexts in which experts have to find ways to acquaint with such a disparity problem. These professionals work hard to provide certain measures to assuage the disparity problem and even study two options of the measures to examine the resolution scale. This community detection approach tends to separate densely linked subcategories differently. Through numerous scientific domains, certain components interrelate. Some experts contemplate that the all-encompassing tasks called community detection in networks can assess the performance of the detection algorithms based on the ground truth communities. It means that no algorithm can exclusively disentangle and resolve community detection. Nonetheless, certain community detection stays a controlling tool and node metadata to have a certain value. Perhaps, experts can create a cautious survey of the relationship with network structure to yield understandings of genuine value. Using the statistical techniques, one can measure the connections between metadata and group structures for the comprehensive models to establish these approaches using both artificial and practical systems.

References

Bedi, P., & Sharma, C. (2016). Community detection in social networks. Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery, 6(3), 116.

Bickel, P. J., & Chen, A. (2009). A nonparametric view of network models and Newman–Girvan and other modularities. Proceedings of the National Academy of Sciences, 106(50), 21068-21073.

Fortunato, S. (2010). Community detection in graphs. Physics Reports, 486(3-5), 78.

Gibbons, J. D., & Chakraborti, S. (2011). Nonparametric statistical inference (pp. 977-979). Berlin, Heidelberg: Springer.

Kernix. (2016). Community detection in social networks. Retrieved May 10, 2019, from <https://www.kernix.com/blog/community-detection-in-social-networks_p13>.

Newman, M. E. (2004). A fast algorithm for detecting community structure in networks. Physical Review E, 69(6), 066133.

Peel, L., Larremore, D. B., & Clauset, A. (2017). The ground truth about metadata and community detection in networks. Science Advances, 3(5), e1602548.

Rosvall, M., Delvenne, J. C., Schaub, M. T., & Lambiotte, R. (2017). Different approaches to community detection. Retrieved May 10, 2019, <from https://arxiv.org/pdf/1712.06468.pdf>.

Shuo, L., & Chai, B. (2016). Discussion of the community detection algorithm based on statistical inference. Perspectives in Science, 7, 122-125.

Stegehuis, C., van der Hofstad, R., & van Leeuwaarden, J. S. (2016). Epidemic spreading on complex networks with community structures. Scientific Reports, 6, 29748.

Sun, P. G. (2016). Imbalance problem in community detection. Physica A: Statistical Mechanics and its Applications, 457, 364-376.

Yang, J., & Leskovec, J. (2013). Overlapping community detection at scale: A non-negative matrix factorization approach. In Proceedings of the sixth ACM international conference on Web search and data mining, pp. 587-596.

Zachary, W. W. (1977). An information flow model for conflict and fission in small groups. Journal of Anthropological Research, 33(4), 452-473.

2019/5579

Author: admin