Lecture Notes in Social Networks

Lecture Notes in Social Networks Series editors Reda Alhajj, University of Calgary, Calgary, AB, Canada Uwe Glässer, Simon Fraser University, Burnaby, BC, Canada Advisory Board Charu Aggarwal, IBM T.J. Watson Research Center, Hawthorne, NY, USA Patricia L. Brantingham, Simon Fraser University, Burnaby, BC, Canada Thilo Gross, University of Bristol, UK Jiawei Han, University of Illinois at Urbana-Champaign, IL, USA Huan Liu, Arizona State University, Tempe, AZ, USA Raúl Manásevich, University of Chile, Santiago, Chile Anthony J. Masys, Centre for Security Science, Ottawa, ON, Canada Carlo Morselli, University of Montreal, QC, Canada Rafael Wittek, University of Groningen, The Netherlands Daniel Zeng, The University of Arizona, Tucson, AZ, USA

More information about this series at http://www.springer.com/series/8768

Rokia Missaoui Idrissa Sarr •

Editors

Social Network Analysis – Community Detection and Evolution

123

Editors Rokia Missaoui Département d’Informatique et Ingéniérie Université du Québec en Outaouais Gatineau, QC Canada

Idrissa Sarr Département de Mathématiques et Informatique Université Cheikh Anta Diop Dakar Senegal

ISSN 2190-5428 ISSN 2190-5436 (electronic) Lecture Notes in Social Networks ISBN 978-3-319-12187-1 ISBN 978-3-319-12188-8 (eBook) DOI 10.1007/978-3-319-12188-8 Library of Congress Control Number: 2014956200 Springer Cham Heidelberg New York Dordrecht London © Springer International Publishing Switzerland 2014 Chapter 2 was created within the capacity of an US governmental employment. US copyright protection does not apply. This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, express or implied, with respect to the material contained herein or for any errors or omissions that may have been made. Printed on acid-free paper Springer International Publishing AG Switzerland is part of Springer Science+Business Media (www.springer.com)

This book on social network analysis is dedicated to our respective families who have been our constant source of inspiration. They instill in us the drive and the power to face any challenge with enthusiasm and good spirit. Without their countless love and support, this project would not have been made possible. Rokia and Idrissa

Foreword

Creatures including humans, animals, insects, etc. avoid living in isolation and tend to form communities or societies. Though Ferdinand Tönnies distinguished between a community and a society in 1887, we may roughly say a community is a group of individuals who agreed or asked to be together in order to achieve a certain task, socialize, etc. Communities range from static and closed to dynamic and open. Some communities are persistent, while others are volatile or ad hoc. Examples of communities include families, friends, neighbors, schoolmates, employees working on a project, etc. Even birds immigrate in communities with specific leadership. Traditionally, the establishment of communities was location indexed, i.e., required the existence of individuals in the same location. However, the development in the communication technology triggered a revolution in the way human communities are established and dissolved. There is a visible rapid shift from physical to virtual communities, i.e., from expecting individuals within a community to know and see each other to accepting the ability of individuals to communicate as sufficient to form a community. The latter trend allows communities to grow and shrink without a real control. However, not all individuals within a community are likely equal when it comes to skills and influence. Thus, analyzing communities to identify and study key individuals, information propagation, evolution, behavior, structure, etc. is essential for knowledge discovery leading to informative decision-making. Thanks to the rapid development in the information technology and computing, which allows researchers to build scalable solutions capable of handling big data. Such an analysis could have been otherwise impossible. In fact, when the study of social communities started as a branch of sociology and anthropology, applications and discoveries remained limited, mainly because researchers concentrated on the study of small communities, which remained small due to the restrictions, which have been released when the ability to communicate became the only requirement and raised the need for the study and analysis of large communities. In other words, earlier studies concentrated on physical communities, while recently virtual communities do exist and are evolving and dominating. Realizing the need to handle evolving communities, researchers from various fields, including computer science, mathematics, statistics, physics, and

vii

viii

Foreword

many other domains joined the efforts to develop new and more powerful techniques capable of accomplishing various types of studies related to communities. A number of new contributions and discoveries are described well in this volume titled “Social Network Analysis – Community Detection and Evolution”, edited by two leading researchers Prof. Rokia Missaoui and Dr. Idrissa Sarr. This volume is indeed unique in its coverage and the background of the elite community of authors who have written in various chapters. Some of the important topics covered include the study of complex networks from understanding group cohesion to group detection, to internetwork community evolution, as well as dealing with Information propagation without relying entirely on the link structure of social networks. The key novelty of the approach relies on the ability to mine the published messages within a microblog platform and to extract the hidden topics to identify the seed users. The volume also discusses the notion of consensual communities and to show that they do not exist within a random graph, yet another evidence in support of the targeted formation of communities. Online communities and behavior are also discussed with emphasis on dating sites to understand how user attributes can help predict who will date whom, and hence provide a recommendation system for online dating website. Further, a group of authors discuss the modeling and visualization of hierarchical structures in large organizational email networks. The evolution of groups and communities on Twitter is also tackled by employing a technique that mixes natural language processing and social network analysis. Another interesting study covers the influence of social media in the election process with a case study on the analysis of tweets related to Iranian presidential elections. Finally, by combining all these topics related to communities and evolution this volume is an attractive source and reference for researchers, practitioners, and students who want to learn some interesting latest developments in the field. Calgary, August 2014

Reda Alhajj

Preface

Introduction Most of the contributions in the present book contain recent studies on community detection and/or evolution and represent extended versions of a selected collection of articles presented at the 2013 IEEE/ACM international Conference on Advances in Social Network Analysis and Mining (ASONAM), which took place in Niagara Falls in Canada between August 25 and 28, 2013. The topics covered by this book can be categorized into two groups: community detection and evolution in the first seven chapters, and two other related topics, namely link prediction and influence/ information propagation or maximization, in the last four chapters.

Community Detection and Evolution The discovery of cohesive groups, cliques, and communities inside a network is one of the most studied topics in social network analysis. It has attracted many researchers in sociology, biology, computer science, physics, criminology, and so on. Community detection aims at finding clusters as subgraphs within a given network. A community is then a cluster where many edges link nodes of the same group and few edges link nodes of different clusters. A general approach to community detection consists in considering the network as a static view in which all the nodes and links in the network are kept unchanged throughout the study. Recent studies focus also on community evolution since most social networks tend to evolve over time through the addition and deletion of nodes and links. As a consequence, groups inside a network may expand or shrink and their members can move from one group to another one over time. Most of the studies on community evolution use topological properties to identify the updated parts of the network and characterize the type of changes such as network shrinking, growing, splitting, and merging. However, recent work has

ix

x

Preface

focused on community evolution/detection by relying entirely on the behavior of group members in terms of the activities that occur in the network rather than exclusively considering links and network density. Another interesting feature of social networks is the cohesiveness of a group and how it varies over time. In fact, the cohesiveness of a group is a social factor that assesses how members of a group are close to each other, and may help predict a possible community splitting or disaggregation. Chapters “Entanglement in Multiplex Networks: Understanding Group Cohesion in Homophily Networks”–“The Power of Consensus: Random Graphs Have No Communities” are proposed to portray trends towards cohesiveness evaluation. Chapter about “The Emergence of Communities and Their Leaders on Twitter Following an Extreme Event” by Yulia Tyshchuk, Hao Li, Heng Ji, and William A. Wallace, combines natural language processing together with social network analysis to explore Twitter messages in order to identify actionable ones, construct an actionable network, identify communities with their central actors, and show the behavior of the community members. The approach has been evaluated on two important real-life events, namely the 2011 Japan Tsunami and the 2012 Hurricane Sandy. The results help understand the behavior of communities as a whole or as individual members of such cohesive groups. Since the two events have different characteristics, the behavior of involved people is dissimilar from one event to the other one. In particular, it was observed that there was a limited participation of Government on Twitter during the 2011 Japan Tsunami compared to an active involvement during the 2012 Hurricane Sandy. Moreover, the leadership roles were stronger in the second than in the first event, while the cohesion in virtual communities on Twitter seems weaker for the Hurricane Sandy. Chapter titled “Hierarchical and Matrix Structures in a Large Organizational Email Network: Visualization and Modeling Approaches” by Benjamin H. Sims, Nikolai Sinitsyn, and Stephan Eidenbenzof studies the visualization and modeling aspects of community detection. Indeed, the email network of a large scientific research organization is analyzed in order to visualize and model organizational hierarchies in complex network structures. To that end, formal organizational divisions and levels are integrated with network data to get an insight into the interactions between subdivisions of the organization and other external organizations. In order to manage the complexity of the large email network, the Girvan-Newman algorithm for community detection is applied. Then, a power law model to forecast degree distribution of organizational email traffic is defined based on the hierarchies that hold between managers and employees. Chapter labeled “Overlaying Social Networks of Different Perspectives for Internetwork Community Evolution” by Idrissa Sarr, Joseph Ndong, and Rokia Missaoui uses probability and possibility theories as two alternate solutions to discover perspective (temporary) communities and highlight community evolution. Starting from snapshots of the network at different time periods, the underlying social network is analyzed in order to first identify active actors (i.e., actors that participate in at least a predefined number of activities) during a set of time slots, and then delimit the perspective communities they form over time. Beside the fact

Preface

xi

that the approach tracks the evolution of the network and identifies the perspective communities, it gives a basic way to identify both active and passive users. The latter group of users can be seen as churners in customer relationship management (CRM) applications. Furthermore, mapping perspective communities to an initial (or important) network adds new links that improve the network accessibility, and hence the information flow circulation. Chapter titled “Study of Influential Trends, Communities, and Websites on the Post-election Events of Iranian Presidential Election in Twitter” by Seyed Amin Tabatabaei and Masoud Asadpour analyzes 1,375,510 tweets of Twitter users who were interested in Iranian Presidential election and its post-events. The top URLs that appeared on the tweets indicate that the most influential websites are those related to social networking and social media websites. Important keywords used in the tweets during nine days are extracted and the most popular websites among two distinct groups of users (Persian and English speaking users) are found. These groups represent the core part of the network and help in interacting with abroad to communicate the news, events, and messages. Peripheral users are identified as well as a few subcommunities within the groups. The specification of subcommunities (i.e., the supporters of political groups) is done based on the keywords extracted from the tweets using a customized version of TF-IDF. Another result shows a strong link between the posted tweets and the political events that occurred the same day. Chapter titled “Entanglement in Multiplex Networks: Understanding Group Cohesion in Homophily Networks” by Benjamin Renoust, Guy Melançon, and Marie-Luce Viaud deals with group cohesiveness in complex networks, mainly, in bipartite graphs. The authors use the homophily concept to assess similarity between actors and the group homogeneity they have. The key idea is that attributes are exploited while investigating how they interact. In other words, authors focus on measuring the cohesion of a group through the interactions that take place between attributes of actors. Hence, actor behavior is used to measure the intensity of interactions and group cohesiveness. Therefore, it can be stated that interactions between actors are a key element to identify group structure and cohesiveness. Instead of projecting a bipartite network onto a single-type network with entities of a same type, which can lead to a loss of information or hide subtle characteristics of the original data, the authors propose to directly study the multiplex networks. By doing so, they demonstrate the feasibility of detecting community structure within complex networks without the need to compute one-mode projections. Chapter titled “An Elite Grouping of Individuals for Expressing a Core Identity Based on the Temporal Dynamicity or the Semantic Richness” by Billel Hamadache, Hassina Seridi-Bouchelaghem, and Nadir Farah is related to group detection and especially to core identification in social networks. The core of a network can be seen as a central part having a high influence on the communication flows that involve the other nodes. Basically, the work can be seen as another contribution to existing studies in group detection by adding the semantic and temporal dimensions. In fact, temporal dynamic behavior or semantic concepts of social entities are an additional input to exploit in order to characterize and strengthen significantly a group structure and highlight its cohesiveness. The key idea of this work is that

xii

Preface

actors of a social network are likely to change their interactions over time by adding or removing relations with others. This has an impact on their social position in the network and/or their possible affiliation to one or more social groups. The temporal change is in fact induced by many factors influencing actor behavior. Therefore, using a semantic dimension such as the connection causality, the positive opinion of socializing, and relationship kinds may help gauge the shape of groups and their cohesiveness. Chapter by Romain Campigotto and Jean-Loup Guillaume on “The Power of Consensus: Random Graphs Have No Communities” defines the notion of consensual communities and shows that they do not exist within a random graph. The principle exploited by the authors is that the outcome of multiple runs of a nondeterministic community detection algorithm is certainly more significant than the outcome of a single run. Authors define a consensual community as a set of nodes, which are frequently classified in the same community through multiple computations. In other words, a consensual community is a repeatable outcome (set of communities) obtained from a set of community detection algorithm computations. The main reason for using consensual communities rather than classical communities comes from the fact that most techniques used to compute communities can usually provide more than one solution. This may depend on the initial configurations or the order in which nodes are considered. Moreover, consensual communities can provide a deeper insight into the structure of the network since they summarize many partitions and encode more information on the structure such as figuring out the overlapping communities. However, when considering random graphs, authors show that it is quite impossible to find consensual communities. The reason is that all pairs of nodes have the same probability to be connected in random graphs. Furthermore, authors demonstrate through various community detection algorithms the existence of a threshold beyond which a trivial consensual community containing all the nodes is found and below which each node forms a consensual community. The remainder of the book covers a few use cases of community structures that address other issues in social network analysis, namely link prediction and influence/information propagation and maximization.

Link Prediction This important topic in social network analysis aims at predicting if two given nodes have a relationship or will form one in the near future. It is exploited in many social media applications such as the ones that need an embedded recommender system to suggest new and relevant ties to the users. Like in community detection, similarity and proximity principles are widely used for link prediction. Moreover, information about network communities can improve the accuracy of similaritybased link prediction methods.

Preface

xiii

Chapter “Link Prediction in Heterogeneous Collaboration Networks” written by Xi Wang and Gita Sukthankar concerns link prediction in heterogeneous collaboration networks. It studies both supervised and unsupervised link prediction in networks where nodes may belong to more than one community, procreating different types of collaborations. Links in heterogeneous networks happen for different reasons, and hence cannot be considered in a homogeneous manner. To take into account such a fact, a new supervised link prediction framework, called Link Prediction using Social Features (LPSF), is proposed and integrates a re-weighting scheme of the network by exploiting features of nodes extracted from patterns of salient interactions in the network. It is shown that the proposed re-weighting method in LPSF better reflects the intrinsic ties between nodes and provides a better prediction accuracy for supervised link prediction methods. Chapter titled “Characterization of User Online Dating Behavior and Preference on a Large Online Dating Site” by Peng Xia, Kun Tu, Bruno Ribeiro, Hua Jiang, Xiaodong Wang, Cindy Chen, Benyuan Liu, and Don Towsley studies user behavior of an online dating website in order to understand how user attributes can help predict who will date whom. By doing so, the authors try to provide outstanding guidelines to design a recommendation system for online dating website. This means that the present work can be seen as a link prediction issue since the recommendation is done once two users are likely to date based on their profiles. An interesting aspect that this paper points out is that the connections between individuals in the underlying network are not deeply related to simple and traditional mechanisms such as preferential attachment or homophily. Actually, user attributes based on preferential attachment cannot be simply used because user behavior in choosing attributes at a given date may largely be done randomly. Moreover, authors observe that the geographic distance between two users and the photo count of users play an important role in their dating behavior, and therefore it is important to differentiate between the effective preferences of users and the random selection of attributes. The main concerns during the approach validation are: (1) How often does a user send and receive messages and how does these operation change over time? and (2) What is the correlation or link between the sender and receiver behavior based on their own profiles?

Influence/Information Propagation and Maximization Influence propagation is usually modeled using propagation models such as Linear Threshold Model and Independent Cascade Model. These models assume that a node is influenced based on the opinions of the local network neighborhood. It has been recently shown that it is more simple and realistic to model the propagation of negative influence, which is more contagious, than modeling the positive influence. Moreover, relying on community membership to study influence maximization is a viable alternate solution that researchers have considered recently as described in the last two chapters of this volume.

xiv

Preface

Chapter titled “Latent Tunnel Based Information Propagation in Microblog Networks” by Chenyi Zhang, Jianling Sun, and Ke Wang deals with Information propagation without relying entirely on the link structure of social networks. The key novelty of the approach is to mine the published messages within a microblog platform and extract the hidden topics to identify the seed users. The basic assumption is that a target message is more likely to be forwarded or re-tweeted if it is interesting to both the sender and the recipient, and an interested user is more likely to react to a message. Hence, when a topic catches the attention of two actors through previous messages, the authors conclude that both actors will probably react to the messages related to that topic and share a hidden link. They afterward identify the seeds of users that will maximize the propagation by identifying those actors, which, when they publish a message, their recipients are likely to forward it, and so on. To reach their goal, the authors unveil the latent topics associated with social links by relying on a standard topic modeling technique based on Latent Dirichlet Allocation. The modeling approach highlights the topic distribution for each link that explains its nature in information flow. These obtained distributions are used to estimate the propagation probability of a link for the target message. Chapter by Mahsa Maghami and Gita Sukthankar about “Scaling Influence Maximization with Network Abstractions” tackles the problem of influence maximization in social networks with an application in the advertising domain. A solution is developed to find the influential nodes in a social network as targets of advertisement based on the network structure, the links among the actors in the network, and the limited advertising budget. The solution is a hierarchical influence maximization approach for product marketing that constructs an abstraction hierarchy to scale and adapt optimization techniques to larger networks. An exact solution is provided on smaller partitions of the network, and a candidate set of influential nodes is selected to be propagated upward to an abstract representation of the original network. The process of abstraction, solution, and propagation is iteratively executed until the resulting abstract network becomes small enough to use an exact optimization solution. To conclude this preface, we would like to thank all authors for their significant contributions that give a broad spectrum of research work on social network analysis, mainly in community detection and evolution, link prediction, and influence propagation. Our warm thanks go also to the reviewers for their careful evaluation of the submissions and their useful comments and suggestions. August 2014

Rokia Missaoui Idrissa Sarr

Contents

The Emergence of Communities and Their Leaders on Twitter Following an Extreme Event . . . . . . . . . . . . . . . . . . . . . . . Yulia Tyshchuk, Hao Li, Heng Ji and William A. Wallace

1

Hierarchical and Matrix Structures in a Large Organizational Email Network: Visualization and Modeling Approaches. . . . . . . . . . . Benjamin H. Sims, Nikolai Sinitsyn and Stephan J. Eidenbenz

27

Overlaying Social Networks of Different Perspectives for Inter-network Community Evolution. . . . . . . . . . . . . . . . . . . . . . . Idrissa Sarr, Joseph Ndong and Rokia Missaoui

45

Study of Influential Trends, Communities, and Websites on the Post-election Events of Iranian Presidential Election in Twitter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Seyed Amin Tabatabaei and Masoud Asadpour

71

Entanglement in Multiplex Networks: Understanding Group Cohesion in Homophily Networks. . . . . . . . . . . . . . . . . . . . . . . . . . . . Benjamin Renoust, Guy Melançon and Marie-Luce Viaud

89

An Elite Grouping of Individuals for Expressing a Core Identity Based on the Temporal Dynamicity or the Semantic Richness . . . . . . . Billel Hamadache, Hassina Seridi-Bouchelaghem and Nadir Farah

119

The Power of Consensus: Random Graphs Still Have No Communities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Romain Campigotto and Jean-Loup Guillaume

145

Link Prediction in Heterogeneous Collaboration Networks . . . . . . . . . Xi Wang and Gita Sukthankar

165

xv

xvi

Contents

Characterization of User Online Dating Behavior and Preference on a Large Online Dating Site . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Peng Xia, Kun Tu, Bruno Ribeiro, Hua Jiang, Xiaodong Wang, Cindy Chen, Benyuan Liu and Don Towsley Latent Tunnel Based Information Propagation in Microblog Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chenyi Zhang, Jianling Sun and Ke Wang

193

219

Scaling Influence Maximization with Network Abstractions . . . . . . . . . Mahsa Maghami and Gita Sukthankar

243

Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

269

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

271

Contributors

Masoud Asadpour Social Networks Lab, School of Electrical and Computer Engineering, University of Tehran, Tehran, Iran Romain Campigotto Sorbonne Universités, Paris, France; CNRS, Paris, France Cindy Chen Department of Computer Science, University of Massachusetts Lowell, Lowell, MA, USA Stephan J. Eidenbenz Los Alamos National Laboratory, Los Alamos, NM, USA Nadir Farah Laboratory of Electronic Document Management LabGED, Badji Mokhtar Annaba University, Annaba, Algeria Jean-Loup Guillaume Sorbonne Universités, Paris, France; CNRS, Paris, France Billel Hamadache Laboratory of Electronic Document Management LabGED, Badji Mokhtar Annaba University, Annaba, Algeria Heng Ji Computer Science Department, Rensselaer Polytechnic Institute, Troy, NY, USA Hua Jiang Product Division, Baihe.com, Beijing, China Hao Li Computer Science Department, Rensselaer Polytechnic Institute, Troy, NY, USA Benyuan Liu Department of Computer Science, University of Massachusetts Lowell, Lowell, MA, USA Mahsa Maghami Department of EECS, University of Central Florida, Orlando, FL, USA Guy Melançon CNRS UMR 5800 LaBRI, INRIA Bordeaux Sud-Ouest, Campus Université Bordeaux I, Talence, France Rokia Missaoui Université du Québec en Outaouais, Québec, Canada

xvii

xviii

Contributors

Joseph Ndong Université Cheikh Anta Diop, Fann Dakar, Senegal Benjamin Renoust CNRS UMR 5800 LaBRI, INRIA Bordeaux Sud-Ouest, Campus Université Bordeaux I, Talence, France; Institut National de L’Audiovisuel (INA), Paris, France Bruno Ribeiro Department of Computer Science, University of Massachusetts Amherst, Amherst, MA, USA Idrissa Sarr Université Cheikh Anta Diop, Fann Dakar, Senegal Hassina Seridi-Bouchelaghem Laboratory of Electronic Document Management LabGED, Badji Mokhtar Annaba University, Annaba, Algeria Benjamin H. Sims Los Alamos National Laboratory, Los Alamos, NM, USA Nikolai Sinitsyn Los Alamos National Laboratory, Los Alamos, NM, USA Gita Sukthankar Department of EECS, University of Central Florida, Orlando, FL, USA Jianling Sun College of Computer Science, Zhejiang University, Hangzhou, China Seyed Amin Tabatabaei Social Networks Lab, School of Electrical and Computer Engineering, University of Tehran, Tehran, Iran Yulia Tyshchuk Department of Industrial and Systems Engineering, Rensselaer Polytechnic Institute, Troy, NY, USA Don Towsley Department of Computer Science, University of Massachusetts Amherst, Amherst, MA, USA Kun Tu Department of Computer Science, University of Massachusetts Amherst, Amherst, MA, USA Marie-Luce Viaud Institut National de L’Audiovisuel (INA), Paris, France William A. Wallace Department of Industrial and Systems Engineering, Rensselaer Polytechnic Institute, Troy, NY, USA Ke Wang School of Computing Science, Simon Fraser University, Burnaby, Canada Xiaodong Wang Product Division, Baihe.com, Beijing, China Xi Wang Department of EECS, University of Central Florida, Orlando, FL, USA Peng Xia Department of Computer Science, University of Massachusetts Lowell, Lowell, MA, USA Chenyi Zhang College of Computer Science, Zhejiang University, Hangzhou, China; School of Computing Science, Simon Fraser University, Burnaby, Canada