A Dependency Based Framework for the Evaluation of Agent Oriented Methodologies Reza Basseda, Tannaz Alinaghi, Camellia Ghoroghi Multi Agent System Research Group, School of Electrical and Computer Engineering, Faculty of Engineering, University of Tehran {R.Basseda, T.Alinaghi, C.Ghoroghi}@ece.ut.ac.ir Abstract Agent oriented methodologies have become one of the most important issues of software development. Several methodologies have been proposed to guide developers through various stages of agent based system development. Deciding which methodology to select in a specific case is a critical problem. This paper presents a framework to evaluate complexity of agent oriented methodologies with a quantitative approach. This framework focuses on the lifecycle of a methodology. The proposed models of the methodology and overall process of its lifecycle is discussed thoughtfully and the complexity of the entire methodology lifecycle is calculated. This method of evaluation will be applied to several agent oriented methodologies and finally the comparison among complexity of the proposed methodologies will be performed. 1. Introduction Great effort has been made to provide methodologies for the development of agent–based systems. These methodologies are successfully applied in developing complex, distributed and dynamic systems. The role of agent oriented methodologies is to help developers in all the phases of the lifecycle of an agent-based application. Some agent oriented methodologies are an extension of object oriented methodologies for instance MESSEGAE [9], PASSI [11], [10], MaSE [12] and Promtheus [13], [14]. On the other hand MAS-CommonKADS [14], [15] and CoMoMAS [4] are evolved from knowledge engineering concepts. Further more, some other methodologies such as Gaia [3] have neither knowledge engineering approach nor object oriented idea. They are absolutely based on agent concepts. Since each methodology offers different modeling concepts, analysis and design techniques, notations and supporting tools, selection of the appropriate methodology is crucial for developers. In general, the vast range of available methodologies may cause undesirable effects: (1) Confusion in selecting a methodology for a specific problem, in particular for industrial developers (2) Difficulty in reaching standards due to multiplicity of methodologies using different specifications; (3) Lack of sufficient samples and case studies in existing methodologies. To address these problems some frameworks are needed to evaluate agent oriented methodologies from different views. In the survey which is presented in [8] existing agent-oriented methodologies and the approaches which have been followed, are introduced and suitability of these approaches for agent modeling is discussed. Shehory and Sturm [2] developed a set of criteria for feature based analysis of methodologies consist of software Engineering criteria and agent based characteristics. The similar works have been done in [6] and [7]. Another effort has been performed in [5]. Cernuzzi and Rossi proposed a quantitative analysis employing quantitative methods. They construct attributes tree, allocate weight to each feature of model, grade each feature and finally calculate a weighted average grade to each methodology. So far several frameworks for evaluating agent oriented methodologies have been proposed but most of them use feature based analysis and evaluation and a small amount of work has been performed to compare agent oriented methodologies in a quantitative approach. Besides, it seems no effort has been performed to evaluate agent oriented methodologies based on overall development life cycle process as well as there is not any evaluation on complexity of models and techniques offered by