Abstract - In this turbulent business environment of global recession, traditional organizational structure is reaching its limits. In order to accommodate itself to these changes, managing informal communication beyond old framework is indispensable. It is critical for innovation management to recognize communities of practice and informal leaders. In previous studies we have demonstrated our method was effective to indentify informal communities and potential leaders from one month email log data collected in September 2008 within an organization through a case study of a global manufacturing company. In this paper we collect the second set of one-month email log in June 2009 so as to chronologically compare with the first set of data collected in September 2008 and to analyze changes before and after major organizational changes triggered by the bankruptcy of Lehman Brothers. Email network analysis helps management systematically view its organization as a whole. Keywords - email, network analysis, organizational management, leadership, innovation I. INTRODUCTION In this turbulent business environment, traditional organizational structure is reaching its limits. By accommodating itself to these changes for its survival and prosperity, business organizations need to manage communication networks beyond old framework. As for innovation management, it is indispensable to identify communities of practice and deploy informal leaders. Through a case study of a global manufacturing company, in our previous studies we have demonstrated our method was effective to indentify informal communities and potential leaders with the network analysis of the first set of one-month email log data collected in September 2008 within the organization [1]. As the results of the previous case study with interviews, we identified communities and hierarchical structures reflect actual status of organization structures of the organization. Most of people who have high network centralities are recognized as key persons in the firm. We found that both betweenness and pagerank is a good indicator to detect hidden leadership in their communities. In this paper, we collect the second set of one-month email log data in June 2009 and chronologically compare and analyze any changes. We use the same methodology of the previous studies for the email network analysis in which we construct an email network from a set of log data, and then identify communities in the email network by performing a topological clustering of the networks. We calculate degree centrality, betweenness centrality, closeness centrality, and pagerank centrality. Clustering process is visualized by a dendrogram which is a hierarchical tree diagram. Then, we interview the managers of the company. Our data are unique in three ways. (1) The email log of a fairly large size organization is collected. (2) Two sets of data are collected for chronological analysis. (3) The collection of data sets coincides with the drastic organizational change owing to the unprecedented business impact triggered by the bankruptcy of Lehman Brothers in September 2008. Consequently, we have the data sets for organizational analysis before and after the impact of global recession from a perspective of informal community by an email network analysis. According to the interview with the managers of the company, the top management team resolutely carried out organizational changes for its survival through the global depression, aiming for (1) restructuring of highly paid managers, (2) rejuvenations for organizational vitality, and (3) reintegration of divisions for innovation. We challenge to evaluate the organizational changes for verification with the email network analysis. As well as informal community analysis, we compare before and after leader characteristics with network centralities and communication patterns. The informal networks coexist with the formal structure of the organization and serve many purposes, such as resolving the conflicting goals of the institution to which they belong, solving problems in more efficient ways [2], and furthering the interests of their members. Despite their lack of official recognition, informal networks can provide effective ways of learning and with the proper incentives actually enhance the productivity of the formal organization [3, 4]. Along with the growth of the informal communities, leadership roles in the communities have been distributed [5]. Given the dynamics of forming communities and distributed leadership, it is important to extract such hidden patterns of collaboration and leadership for organizational management that could lead to innovation. The previous approach to identify informal community was to gather data from interviews, surveys, or other fieldwork and to construct links and communities by manual inspection [6] or an internet-centric approach [7]. These methods are accurate but time-consuming and Email Network Analysis for Organizational Management H. Tashiro 1 , J. Mori 1 , N. Fujii 2 , and K. Matsushima 1 1 Graduate School of Engineering, the University of Tokyo, Tokyo, Japan 2 Faculty of Science and Engineering, Waseda University, Tokyo, Japan {jmori, tashiro}@ipr-ctr.t.u-tokyo.ac.jp nf_tomo_home@ybb.ne.jp matsushima@biz-model.t.u-tokyo.ac.jp 958 978-1-4244-6567-5/10/$26.00 ©2010 IEEE labor-intensive, prohibitively so in the context of a very large organization. Given the recent development of online communications in an organization, several studies have been working on identification of communities using online information resources [8]. Adamic showed that the communities, identified from online mailing lists and Web, resemble the actual social communities of the represented individuals [9]. Among several communication means, email has widely become the means of communication in an organization. Therefore email has been established as an indicator of collaboration and knowledge exchange [8, 10, 11]. Since email provides plentiful data on personal communication in an electronic form which enables automatic processing of data, several studies have addressed using email to discover shared interests, relationships, and social networks [12, 13]. Providing the structure and communication patterns within an organization [14, 15], email networks are useful information resources to find informal communities. Several studies have proposed automated methods for using email data to construct a network, and then identify informal communities within an organization [16, 17]. However, there is not yet enough understanding and evaluation regarding how identified communities from email data can be exploited for management of organization and leadership which is important to enhance organizational innovation. In this paper, we collect and analyze the second set of the one-month email log data with the method for indentifying informal communities and potential leaders. We use the clustering method that can rapidly detect dense communities within an email network. The result of the clustering process reveals informal communities and hierarchical structures with an organization. To characterize people in the informal communities, we calculate several network centralities of a person using the structure of an email network. Through the interviews with the managers, these measures enable us to identify leadership roles with the informal communities. Then, we compare two sets of email communication networks to see if we can conclude any managerial implications with significance to the top management. II. METHODOLOGY A. Email network We construct an email network from email log data. We extract the information about sender and receiver from each email. The sender or receiver corresponds to a node in the network. If there is at least one email communication between persons, an edge is then drawn between these persons. As a sum of the nodes and edges, we finally obtain the email network. Since we distinguish a sender from a receiver, an email network is expressed as a directed graph. Given the network, we find a maximal complete sub-graph as a clique which becomes a target of following network analysis. B. Email network analysis We first identify communities in the email network. To this aim, we perform a topological clustering of networks. Although such a methodology had been difficult to achieve due to the difficulty in performing cluster analysis of non-weighted graphs consisting of the large number of nodes, recently proposed algorithms [18, 19] facilitate fast clustering with calculation time in the order of O((l+n)n), or O(n 2 ) on a sparse network with l links; hence this could be applied to large-scale networks. The algorithm proposed was based on the idea of modularity. Modularity Q was defined as follows [18, 19, 20]: ¦ » » ¼ º « « ¬ ª ¸ ¹ · ¨ © §  m N s ss l d l l Q 1 2 2 where N m is the number of modules, l s is the number of links between nodes in module s, and d s is the sum of the degrees of the nodes in module s. In other words, Q is the fraction of links that fall within modules, minus the expected value of the same quantity if the links fall at random without regard for the modular structure. A good partition of a network into cluster must comprise many within-cluster links and as few as possible between-cluster links. The objective of a community identification algorithm is to find the partition with the largest modularity. The algorithm to optimize Q over all possible divisions is as follows. Starting with a state in which each node is the only member of one of n clusters, we repeatedly join clusters together in pairs, choosing the join which results in the greatest increase in Q at each step. Since a high value of Q represents a good cluster division, we stopped joining when Ǽ Q became minus. At the maximal value of Q, Q max , we obtain a cluster structure of a network with effective division. The clusters correspond to the informal communities in the email network. The cluster label can be assigned by examining characteristics of node attributes. A node in a cluster is characterized with its network centralities [21]. We calculate centralities as follows.  Degree centrality: the number of links of a node.  Betweenness centrality: the number of node pairs that pass through a node.  Closeness centrality: average shortest path to other nodes.  Pagerank centrality: the stationary distribution of the Markov chain corresponding to the stochastic transition matrix of a network. Assuming that leadership is influenced by communication and trust on one’s social network [5], leadership roles are characterized with these centralities. C. Email network visualization  To visualize the large-scale network, we employ the force-directed GEM layout [22]. GEM optimizes minimal 959 Proceedings of the 2010 IEEE ICMIT node distances and constant edge lengths and in turn visualizes a network as a circle. This layout helps give an overview of identified clusters in a network. Clustering process is visualized by a dendrogram which is a tree diagram frequently used to illustrate the arrangement of the clusters produced by hierarchical clustering. A dendrogram helps show hierarchical structure among clusters and therefore understand how identified communities are related each other. III. RESULTS We applied our method to actual email data from one firm. We collected two sets of one-month email log in September 2008 (data1) and in June 2009 (data2) in order to chronologically compare and analyze any changes. The data1 includes emails of 2,882 employees and the data2 includes emails of 2,459 employees. For reasons of privacy and complexity, we only used emails that had an internal origin and destination within the firm. Table I shows properties of a network from the data1. Each node has 51.77 links on average and the whole network showed power law in degree (see Fig. 1.). It also has “small-world” properties where clustering coefficients are much larger than the ones of random network (0.387 / 0.01) and the path length (2.67/ 2.74) is close to the one of random network (see TABLE I). Table II shows properties of a network from the data2. Each node has 36.151 links on average and the whole network showed power law in degree (see Fig. 2.). It also has “small-world” properties where clustering coefficients are much larger than the ones of random network (0.377 / 0.01) and the path length (2.72/ 2.76) is close to the one of random network (see TABLE II). We applied our algorithm as described above to identify the communities within the network. We obtained seven distinct clusters from the data1 as shown in Fig. 3. We also identified the hierarchical structure among communities from the data1 as shown in Fig. 5. From the data2, we obtained four distinct clusters as shown in Fig. 3. Consequently, we indentified the hierarchical structure among communities from the data1 as shown in Fig. 6. We manually checked division that each employee in a cluster belongs. We found that each cluster nearly corresponded to one or combination of some divisions in the firm. We showed the results to people from the firm and conducted interviews. They agreed with that both identified communities and hierarchical structure reflect actual status of organization structures of the firm. They pointed out that some identified clusters fit informal communities that play important roles in the organization management. We also showed people who have high network centralities in a community. They recognized most of people who have high network centralities as key persons in the firm. However, they also find some people who they did not expect have high network centralities. In fact, the further interviews reveal that such people have potential leadership for the organization management. In particular, we found that both betweenness and pagerank is a good indicator to detect such hidden leadership among the centralities. Fig. 1. Degree distribution of the email network (2008.09). Fig. 2. Degree distribution of the email network (2009.06). TABLE II PROPERTIES OF THE EMAIL NETWORK (2009.06) n average k C L 2,459 36.151 0.377 (0.010) 2.72 (2.76) n: number of nodes, k: number of links C: Clustering coefficient, L: Average path length TABLE I PROPERTIES OF THE EMAIL NETWORK (2008.09) n average k C L 2,882 51.77 0.387 (0.010) 2.67 (2.74) n: number of nodes, k: number of links C: Clustering coefficient, L: Average path length 960 Proceedings of the 2010 IEEE ICMIT Fig. 3. Clusters of the email network (2008.09). Fig. 4. Clusters of the email network (2009.06). Fig. 5. Dendrogram of Clusters of the email network (2008.09).  Fig. 6. Dendrogram of Clusters of the email network (2009.06). IV. DISCUSSION A. Small World The email communication network maintains the properties of scale-free and “small-world” network. The number of nodes, or senders and receivers of emails, has decreased drastically by 14.7% from 2,882 to 2,459, comparing with the previous period in September 2008. The number of edges, or email communication links between nodes, has decreased by 40.4% from 74,601 to 44,448. The average degree, or average number of people the nodes communicate with email, has also decreased by 30.2% from 52 to 36. Clustering coefficient, the tendency to group together, has decreased by 2.5% while average path length has increased by 2.1%. The results indicate the facts that there were drastic reduction of email users and changes in email behavioral pattern among employees. The number of email communication in the organization has been reduced. The scope of communication rather focused than the previous period. The interview with the managers revealed the company offered a voluntary early retirement program for highly paid seniors and managers to improve the company’s income statement. Consequently, the organization was slimed down and restructured. The concurrent reduction of both overtime and number of workers gave the employees time pressures to reduce issuances of emails. In the past the seniors and the managers retired early had to be included in the communication network. The early retirement of those people influenced the reduction of direct emails as well as carbon copies for red tapes. The top management realized its intention for higher productivities by reducing inputs of the management resources even though the sales have radically decreased during the global recession. The analysis shows the organization as a whole accommodate the challenge of time constraint with the radical reduction of email time along with preparation of attachments as one of the means. B. Formal  and  Informal  Organization The email network with clusters represents one aspect of the organizational reality. The number of clusters has decreased from 7 to 4. The previous cluster C was merged with A, forming a cluster of 770 nodes. The previous D remains as the smallest cluster D of 19 nodes, and the previous cluster B remains as the present B of 265 nodes. The previous clusters E and G were merged with the cluster F, forming the largest present cluster F of 1,405 nodes. According to the dendrogram analysis, the previous clusters A and B had stronger tie with each other. However, after the major organizational restructuring, the clusters B and F are closer now. As the cluster D supplies parts to the cluster A, they remains close relationship. 961 Proceedings of the 2010 IEEE ICMIT In the previous section, we observed the productivity increase of the new organization after the major change. On the other hand, we can deduce the decrease of the emails with lower priority, taken over the necessary work- related emails. The majority of the informal layer of communities was removed from the email communication network. With these assumptions, the current communication network with clusters represents rather job-related communication network. According to the interview, the top management aimed the integration of headquarters with business divisions. The largest cluster F demonstrates the integration of headquarters functions and one business division as well as its business branches. Physical locations and peer human relationships became less significant than work relationship in the dendrogram. That is evidenced by the merger of the cluster E and the cluster G with the cluster F. Although the physical locations of the clusters A and B are close, on business basis, the cluster B is now closer to the cluster F. However, the independence of the cluster A was emphasized as a self-sufficient organization. The phenomenal observation of the changes in the clusters is meaningful for the evaluation of the top management’s intention of the organizational change and reshuffle of managers. The email network analysis with communication network with clusters provides the top management with rather objective pictures of before and after the organizational changes as well as its environmental changes. This is a powerful feedback for the top management team to evaluate the organizational status and performance of their strategies. As changes become faster and stronger in magnitude of turbulence in business circumstances, quick feedback and chronological database surely assist the organizational leaders for effective management. C. Individual Centralities None of the top 30 employees in the previous pagerank or betweenness lists was ranked within the current top 30 this time. In other words, the people with high scores in the communication importance and bridge were replaced with the new groups. On the other hand, according to the interview, the pagerank and betweenness were still indicators of potential leaders. The communication structures were dynamically changed through the major restructuring. The managers told us in the interview that a year ago, the degree centrality of administrative assistants, office clerks, and people who had established their own informal networks over long periods of their career in the organization was higher. However, the analysis of data2 showed that new divisional managers’ degrees were higher. The returned overseas expatriates and young analytic engineers were with higher scores than before. Owing to the early retirements of seniors in their 50’s, the new organizational communication network was shifted toward the healthy directions as the top management intended. The innovative leaders have created the environment of the knowledge interactions through communication among their members and the ecosystem of knowledge creation. As the cores of the communication network clusters, they have managed the effective communication through their strong visions of the organizational success. The visualization of such leaders and their communication patterns as the managers of successful teams has helped the top management design and implement its strategy for the innovation management. D. Future Study In near future, we plan to narrow our focus organization down to engineering groups and their interactions with the entire organization. Recently, for innovative stimulus the top management engineered the system of an internal engineering community program within the engineering organization. An engineering community is engineers’ group of one technological element across the organization. The members of cross- divisional communities are from novices to experienced specialists. The top management needs to evaluate the activities of leaders and their communities. We are going to analyze the email communication networks with clusters and network centralities of the engineering communities. We also plan to compare the innovative activities before and after the engineering community program introductions. V. CONCLUSION We observed the chronological phenomena of managerial decisions of organizational changes with email log data by network analysis. Characteristics changes of communities are clearly curved in relief with cluster analysis. Leadership roles have not changed between before and after analysis while leaders reduced their influences as bridges among communities. Our method helps management systematically view its organization as a whole by using email network analysis. The email network analysis can be used to evaluate communication of interactions among the members. It also helps identify candidates of leaders acting as a hub of information channel of the communication network. Formal organization would be evaluated with informal communities before and after major organizational changes. There are traditional interviews and questionnaires to capture a state of organization. Email network analysis provides with one more significant, objective, and analytical tool in a manager’s tool box. 962 Proceedings of the 2010 IEEE ICMIT REFERENCES [1] J. Mori, H. Tashiro, K. Haraoka, and K. Matsushima, “Identifying Informal Communities and Leaders for Total Quality Management using Network Analysis of Email,” In Proc. of the International Conference on Industrial Engineering and Engineering Management (IEEM), 2009. [2] B. A. Huberman and T. Hogg, “Communities of Practice: Performance and Evolution”. Computational and Mathematical Organization Theory, Vol. 1, pp. 73-92, 1995. [3] D. Crane, Invisible Colleges: Diffusion of Knowledge in Scientific Communities. University of Chicago Press, Chicago, 1972 [4] J. Lave and E. Wenger, Situated Learning: Legitimate Peripheral Participation. Cambridge University Press, 1991. [5] Chen, J. Li, and H. Wang, “Structure and Dynamics of Distributed Leadership in the Perspective of Social Network Analysis,” In Proc. of the International Conference on Industrial Engineering and Engineering Management (IEEM), 2008. [6] T. Allen, Managing the Flow of Technology. MIT Press, 1984. [7] L. Garton, C. Haythornthwaite, and B. Wellman, “Studying online social networks,” Journal of Computer-Mediated Communication, Vol. 3, No. 1, 1997. [8] B. Wellman, “Computer Networks As Social Networks,” Science Vol. 293, No. 14, 2001. [9] L. A. Adamic, and E. Adar, “Friends and Neighbors on the Web,” Journal of Social Networks, Vol. 25, No. 3, 2002. [10] S. Whittaker and C. Sidner, “Email Overload: Exploring Personal Information Management of Email”, in Proc. of CHI ’96, pp. 276-283, 1996. [11] N. Ducheneaut and V. Bellotti, “A Study of Email Work Processes in Three Organizations,” Journal of CSCW, 2002. [12] M. F. Schwartz and D. C. M. Wood, “Discovering Shared Interests Among People Using Graph Analysis”, Communications of the ACM, volume 36, issue 8, pp. 78-89, 1992. [13] A. Culotta, R. Bekkerman and A. McCallum, “Extracting social networks and contact information from email and the Web,” First Conference on Email and Anti-Spam, 2004. [14] R. S. Burt, “Models of Network Structure”, Annual Review of Sociology, Vol. 6, pp. 79-141, 1980. [15] W. R. Scott, Organizations: Rational, Natural, and Open Systems. Prentice-Hall, Inc., Englewood Cliffs, New Jersey, 1992. [16] J. R. Tyler, D. M. Wilkinson, and B. A. Huberman, "Email as spectroscopy: Automated discovery of community structure within organizations," The Information Society, Vol. 21, No. 2, pp. 143-153, 2005. [17] J. Diesner and K. M. Carley, “Exploration of communication networks from the Enron email corpus,” Proceedings of Workshop on Link Analysis, Counterterrorism and Security, SIAM International Conference on Data Mining, pp. 3-14, 2005. [18] M. E. J. Newman, “Fast algorithm for detecting community structure in networks,” Physical Review E, vol. 69, art no. 066133, 2004. [19] M. E. J. Newman and M. Girvan, “Finding and evaluating community structure in networks,” Physical Review E, vol. 69, art no. 026113, 2004. [20] R. Guimerà, M. Sales–Pardo, and L. A. N. Amaral, “Modularity from fluctuations in random graphs and complex networks,” Physical Review E, vol. 70, art no. 025101, 2004. [21] L. C. Freeman, "Centrality in Social Networks: Conceptual Clarification," Social Networks, Vol.1, pp.215-239, 1978. [22] A. Frick, A. Ludwig, and H. Mehldau. A fast adaptive layout algorithm for undirected graphs. In R. Tamassia and I. G. Tollis, editors, Graph Drawing (Proc. GD ’94), volume 894 of Lecture Notes Comput. Sci., pages 388–403.Springer-Verlag, 1995 963 Proceedings of the 2010 IEEE ICMIT . tashiro}@ipr-ctr.t.u-tokyo.ac.jp nf_tomo_home@ybb.ne.jp matsushima@biz-model.t.u-tokyo.ac.jp 958 97 8-1 -4 24 4-6 56 7-5 /10/ $26.00 ©2 010 IEEE labor-intensive,. data sets for organizational analysis before and after the impact of global recession from a perspective of informal community by an email network analysis.