10 th International Research/Expert Conference “Trends in the Development of Machinery and Associated Technology” TMT 2006, Barcelona – Lloret de Mar, Spain, 11-15 September, 2006 CLUSTER ANALYSIS OF A SCALE-FREE NETWORK Nina Bijedic University of Dzemal Bijedic, Faculty of Information Technologies Mostar, Bosnia and Herzegovina Senad A. Burak University of Sarajevo, Faculty of Mechanical Engineering Sarajevo, Bosnia and Herzegovina ABSTRACT In this paper we present an analysis of a cluster based inference in a particular computer network. The faculty forum on a real community server, where students and stuff share their knowledge and experiences, is used for this purpose. In order to better understand the structure of the network, we represent it as a graph, where vertices are represented by the members of the forum and the edges act as the links between the forum posts. As in many similar systems, this forum is organized in threads that are divided into sections (subjects), and sections are divided into groups (academic years). It is shown that the resulting network exhibits a scale-free distribution with large clustering coefficients following the small-world properties. As the clusters hold some important information about the nature of the network, we developed a special software agent that explores the background SQL database and automatically acquires the relevant information. Based on this data, detailed information including the graphs degree distribution, clustering coefficient, Laplacian, and normalized Laplacian eigenvectors and average distance are calculated. The resulting analysis gives us a better understanding of the nature of this particular network, which can be valuable information for the administrators. Keywords: networking, programming, simulations 1. INTRODUCTION An important part of the education process at the Faculty of Information Technologies in Mostar, Bosnia Herzegovina, is its system for distance learning. Among many modules and technologies used for this purposes, the faculty forum is one of the most popular, because it improves the communication between the students and staff and offers a valuable source of knowledge and experience to the users. In order to better understand the knowledge sharing, we explore the topology of the forum’s communication as a directed graph. In this model the users that posted their messages act as vertices and the edges of the graph are established if the user n i posted an answer to a message posted by the user n j . The resulting graph is then represented by an adjacency matrix. The following topology exploration includes the three operations: the distribution of the in-degree and out-degree vertex, the calculation of the average path length and the determination of the graph’s clustering coefficient. 2. DATA ACQUISITION AND GRAPH MODELING We explored the forum’s data from the underlaying MS SQL Server 2000. During the observation period there were 602 registered users, but only 335 were active with the total number of 10,180 posts. In order to represent the graph by an adjacency matrix A, we developed a special software agent in C# language that aquired and automatically anlysed the data. The three most important fields 689