RiSD: A Methodology for Building i* Strategic Dependency Models Gemma Grau, Xavier Franch, Enric Mayol, Claudia Ayala, Carlos Cares, Mariela Haya, Fredy Navarrete, Pere Botella, Carme Quer Universitat Politècnica de Catalunya (UPC), LSI - Campus Nord, Barcelona (Catalunya, Spain) {ggrau, franch, mayol, cayala, ccares, mhaya, fjnavarrete, botella, cquer}@lsi.upc.edu Abstract Goal-oriented models have become a consolidated type of artefact in various software and knowledge engineering activities. Several languages exist for representing such type of models but there is a lack of associated methodologies for guiding their construction up to the necessary level of detail. In this paper we present RiSD, a methodology for building Strategic Dependency (SD) models in the i* notation. RiSD is defined in a prescriptive way to reduce uncertainness when constructing the model. RiSD also tackles two fundamental issues: on the one hand, it tends to reduce the average size of the resulting models and, on the other hand, it allows including some traceability relationships in the resulting models. As a result, we may say that RiSD increases the understandability of goal-oriented models whilst improving all construction. 1. Introduction In the last years, the construction of goal-oriented and agent-oriented models has become an extended practice in fields such as requirements engineering and organizational process modeling [1, 2]. One of the most widespread goal-oriented languages is the i* notation proposed by Eric Yu in the first half of the 90’s [3, 4]. i* allows for the clear and simple statement of goals that system actors have and dependencies among them. In despite of its utility, the intensive use of i* reveals some difficulties. In [5] we have tackled one of them, namely the diversity of i* dialects and variations that may be disturbing when learning the notation. In this paper we deal with another two drawbacks that we have experimented: the absence of detailed methodologies for building the models and the complexity of the resulting models. - Absence of methodology. Currently we can say that there is a lack of guidance for supporting the prescriptive construction of i* models. There exists a consolidated methodology such as Tropos [6] but it is aimed mainly to the guidance of the whole software development process. In this sense, it supports the conception of a global solution for the problem at hand, but gives a high degree of freedom for the construction of the models themselves (i.e., which intentional elements exist). One could argue that this is precisely a property inherent to agent-oriented methodologies [4], but the flexibility of the i* language means to have multiples choices when building a model (i.e. when to include an intentional element or not, which type of element is the most appropriate for a given situation, etc). - Complexity of the models. Models for non-trivial systems grow very quickly and are plenty of intentional elements of many types without obvious relationships among them. There are two main types of hidden relationships. On the one hand, two intentional elements may depend one on another (e.g., one may imply the other). On the other hand, two intentional elements may be at different levels of detail, being one a refinement of another. For some types of relationships we may find constructs in the language but not for all, especially when referring to one of the two types of models offered by i*, namely Strategic Dependency (SD) model. In this paper, we propose RiSD, a methodology for building R educed i * SD models for software systems. RiSD is defined as a set of activities structured in two phases, one for constructing the social system (without software) and the other for constructing the socio- technical system (with software). Both phases may involve the partial or total construction of the other type of i* models, namely Strategic Rationale (SR) models. RiSD includes precise questions and answers that guide the development process and provide cut criteria for choosing among different types of intentional elements when diverse options exist. The size of the resulting model is reduced due to these criteria. RiSD includes also some traceability constructs that show the relationships among intentional elements and enhances therefore understanding of the model. 2. The i* language The i* language defined by Eric Yu [3, 4] proposes the use of two models, each one corresponding to a different abstraction level: a Strategic Dependency (SD) model represents the intentional level and the Strategic