Expert knowledge is involved in every software development project since developers must face numerous decision tasks during requirements management, analysis, design, and implementation stages. Therefore, if expert knowledge could be properly modelled and incorporated in the di erent processes of software development as well as in the CASE tools that support these processes, that would mean a great advantage for any software development. In software development, requirements stage is considered a good application domain for Arti cial Intelligence (AI) techniques because of requirements nature. Software requirements express and establish the needs and constraints that contribute to the solution of a real world problem [7]. However, requirements tend to be imprecise, incomplete and ambiguous[3] and has a big impact in whole development stages [5, 15, 1, 2]. Therefore, the use of AI techniques in order to improve requirements stage will favorably a ect the whole software life cycle, but we need a seamless integration of Requirement Engineering (RE) and AI techniques to exploit the bene ts of collaboration between these two knowledge areas [10]. Besides, the biggest breakthrough in requirement management is when you stop thinking of documents and start thinking about information. Here, is where CARE (Computer-Aided Engineering Requirement) tools help us in order to be able to handle all of this information. InSCo Requisite is an academic web CARE tool, developed by DKSE group at the University of Almer a, which aids during the requirement development stage [11]. This work presents the architecture for the seamless integration of a CARE tool to manage requirements (i.e. InSCo Requisite) with some AI techniques (i.e. Bayesian networks [12, 6] and metaheuristics). Speci cally, a Bayesian network, called Requisites [13], is used in the requirement validation task in order to validate the Software Requirements Speci cation (SRS) of a software development project, they has been successfully applied in SE, [9, 4, 8, 13]. Metaheuristic techniques (Simulated Annealing, Genetic Algorithms and Ant Colony Systems) are used in the problem of selecting the subset of requirements among a whole ? This is a long abstract of the paper published in Lozano,J.A., Gamez, J.A., Moreno, J.A. (eds.) LNAI Series, Current Topics in Arti cial Intelligence. 14th Conference of the Spanish Association for Arti cial Intelligence, CAEPIA 2011, La Laguna, Spain, November 8-11, 2011, Selected Papers.
set of candidate requirements proposed by a group of stakeholders, that will be
included in the development of a nal software product [
The RE work ow depicted in Figure 1 shows an organization of the tasks that must be done in a software development project during RE stage. Requirements are elicited or gathered from users, next they are speci ed in a document or its electronic equivalent, known as Software Requirements Speci cation (SRS). CARE tools provide environments that make use of databases, allowing an effective management of the requirements of any software project. Requirements validation checks whether the elicited and speci ed requirements present inconsistencies; if the information is incomplete or if there are ambiguities in the system de nition. Requisite Bayesian network provide developers an aid, under the form of a probabilistic advice (i.e. an estimation of the degree of revision for the SRS), helping them at the time of making a decision about the stability of the current requirements speci cation. Finally, requirements selection task has as main objective to choose, from all the requirements de ned in the speci cation, the subset of requirements that will be implemented.
Bayesian networks and metaheuristic techniques have demonstrated to
obtain interesting results through di erent tests data [
The interface is a web environment accessed from a web browser. Data layer is in charge of storing and managing the electronic representation of SRS handled by InSCo Requisite tool and the knowledge base that contains the Bayesian network Requisites. Service layer is composed by the CARE tool (i.e. InSCo Requisite), the AI techniques used to address requirements validation (i.e. Bayesian network Requisites) and requirements selection (i.e. metaheuristics algorithms) tasks. Communication interface connect CARE and knowledge-based tools passing the required information needed for the execution of the appropriated processes. Thus, requirement validation receives metrics on the SRS and returns an estimation of the degree of revision for SRS; requirement selection receives resources e ort bound and speci c measures on individual requirements and set the set of requirements in order to be implemented. All of these communication processes are performed through XML les.
The purpose of this work is to de ne a three-layer architecture which: a) allows the seamless collaboration between RE tasks and some AI techniques (Bayesian networks, simulated annealing, genetic algorithms and ant colony sys
Fig. 1. Seamless synergic architecture. tems) in order to perform a software development project; b) facilitates their parallel and independent evolution.
Acknowledgments. This work was supported by the Spanish Ministry of Science and Innovation under project TIN2010-20900-C04-02 and by the Junta of Andaluc a under project TEP-06174.