A Labeled Graph Approach to Analyze Organizational Performance Mark Hoogendoorn 1 , Jan Treur 1 , and Pınar Yolum 2 1 Vrije Universiteit Amsterdam, Department of Artificial Intelligence De Boelelaan 1081a, 1081 HV Amsterdam, The Netherlands {mhoogen, treur}@cs.vu.nl 2 Bogazici University, Department of Computer Engineering, TR-34342 Bebek, Istanbul, Turkey pinar.yolum@boun.edu.tr Abstract Determining the performance of an organization is a must for both human and multi-agent organizations. The performance analysis enables organizations to uncover unexpected properties of organizations and allow them to reconsider their internal workings. To perform such an analysis, this paper represents organizations as labeled graphs that capture, not only the interactions of the entities, but also the characteristics of those interactions, such as their content, frequency, and so on through labels in the graph. Algebraic representation and manipulations of the labels enable analysis of a given organization. Hence, well-known phenomena, such as overloading of participants or asymmetric distribution of workload among participants can easily be detected. Finally, a case study is performed within the domain of incident management. 1. Introduction Multi-agent organizations consist of agents that interact to carry out their tasks. Current models of multi-agent organizations usually represent organizations as consisting of roles that agents adopt. An organization model then specifies the structure and behavior of the organization in terms of the relations between the roles. An analysis of such an organization model could check if the model satisfies desired properties such as the possibility of completing a desired task given that all agents comply with the requirements of the organization. Whereas such an analysis is useful, it is not sufficient to analyze an executing organization. The main reason is that many design-time choices become concrete during execution. Agents choose who they want to interact with as well as how often they want to do so during run-time. For example, among two agents that enact a merchant role, one might be preferred over the other because the agent has better capabilities, more work capacity, and so on. These subtle interactions of agents at run-time can give rise to interesting situations that can only be detected during execution. That is, as a result of previous decision, one merchant agent will be more loaded than the second merchant will be. Further, the agents that participate in an organization might be designed and developed by independent parties, which requires them to interoperate and execute intelligently at run-time. In other words, such facts about the workings of a multi- agent organization cannot be discovered from a static representation of an organization during design time, but can only be analyzed during the execution time. Whereas there is a vast literature in the design of multi-agent organizations, there is little work on the analysis of executing multi-agent systems [9, 10]. For this reason, this paper provides a complementary treatment of multi-agent organizations, where in addition to existing design time dynamics of the organizations, a graph representation is used to analyze executing organizations. Executing multi-agent organizations are analyzed by logging the performance of the organization in traces. Graph representations are useful for analyzing organizations; for example for understanding the structure of an organization through theoretical concepts. This paper presents a formal specification language based on a graph representation. The directed graph captures the relationships between participants in the organization and the labels give semantics to the