AIRE 2014 – Proceedings

Preface
Welcome to the First International Workshop on Artificial Intelligence for Requirements Engineering (AIRE 2014). AIRE is an interdisciplinary workshop in which the synergies between Artificial Intelligence and Requirements Engineering are explored and extended. Our objectives are to identify complex Requirements Engineering (RE) problems that could benefit from the application of AI techniques, explore initial solutions, and to promote a new and broad community for interdisciplinary discussions concerning novel research directions.

Pragmatic Ambiguity Detection in Natural Language Requirements
Alessio Ferrari, Giuseppe Lipari, Stefania Gnesi, and Giorgio O. Spagnolo
(ISTI-CNR, Italy)
This paper presents an approach for pragmatic ambiguity detection in natural language requirements. Pragmatic ambiguities depend on the context of a requirement, which includes the background knowledge of the reader: different backgrounds can lead to different interpretations. The presented approach employs a graph-based modelling of the background knowledge of different readers, and uses a shortest-path search algorithm to model the pragmatic interpretation of a requirement. The comparison of different pragmatic interpretations is used to decide if a requirement is ambiguous or not. The paper also provides a case study on real-world requirements, where we have assessed the effectiveness of the approach.

Weka Meets TraceLab: Toward Convenient Classification: Machine Learning for Requirements Engineering Problems: A Position Paper
Jane Huffman Hayes, Wenbin Li, and Mona Rahimi
(University of Kentucky, USA; DePaul University, USA)
Abstract—Requirements engineering encompasses many difficult, overarching problems inherent to its subareas of process, elicitation, specification, analysis, and validation. Requirements engineering researchers seek innovative, effective means of addressing these problems. One powerful tool that can be added to the researcher toolkit is that of machine learning. Some researchers have been experimenting with their own implementations of machine learning algorithms or with those available as part of the Weka machine learning software suite. There are some shortcomings to using “one off” solutions. It is the position of the authors that many problems exist in requirements engineering that can be supported by Weka’s machine learning algorithms, specifically by classification trees. Further, the authors posit that adoption will be boosted if machine learning is easy to use and is integrated into requirements research tools, such as TraceLab. Toward that end, an initial concept validation of a component in TraceLab is presented that applies the Weka classification trees. The component is demonstrated on two different requirements engineering problems. Finally, insights gained on using the TraceLab Weka component on these two problems are offered.

Transferring Research Into the Real World: How to Improve RE with AI in the Automotive Industry
Sven J. Körner, Mathias Landhäußer, and Walter F. Tichy
(KIT, Germany)
For specifications, people use natural language. We show that processing natural language and combining this with intelligent deduction and reasoning with ontologies can possibly replace some manual processes associated with requirements engineering (RE). Our prior research shows that the software tools we developed can indeed solve problems in the RE process. This paper shows this does not only work in the software engineering domain, but also for embedded software in the automotive industry. We use artificial intelligence in the sense of combining semantic knowledge from ontologies and natural language processing. This enables computer systems to “understand” requirement texts and process these with “common sense”. Our specification improver RESI detects flaws in texts such as ambiguous words, incomplete process words, and erroneous quantifiers and determiners.

Customizable Rule-Based Verification of Requirements Ontology
Dang Viet Dzung and Atsushi Ohnishi
(Ritsumeikan University, Japan)
In using ontology to support requirements engineering, quality of elicited requirements depends on quality of requirements ontology, so a rule-based verification method of the correctness of requirements ontology has been proposed. However, in recent evaluation experiments, users of the method (ontology verifiers) described only a few new rules based on rule grammars and rule examples. That led to the number of correctly detected errors were not so high (less than 50% of the total number of errors). To improve our method, in this paper, we propose a rules customization mechanism in which simple specific rules are generated using pre-defined and customizable meta-rules. We expect that by using the improvement, ontology verifiers can easily and effectively generate and customize rules for verification of requirements ontology. The customization mechanism is illustrated through examples and a case study.

Content-Based Recommendation Techniques for Requirements Engineering
Gerald Ninaus, Florian Reinfrank, Martin Stettinger, and Alexander Felfernig
(Graz University of Technology, Austria)
Assuring quality in software development processes is often a complex task. In many cases there are numerous needs which cannot be fulfilled with the limited resources given. Consequently it is crucial to identify the set of necessary requirements for a software project which needs to be complete and conflict-free. Additionally, the evolution of single requirements (artifacts) plays an important role because the quality of these artifacts has an impact on the overall quality of the project. To support stakeholders in mastering these tasks there is an increasing interest in AI techniques. In this paper we presents two content-based recommendation approaches that support the Requirements Engineering (RE) process. First, we propose a Keyword Recommender to increase requirements reuse. Second, we define a thesaurus enhanced Dependency Recommender to help stakeholders finding complete and conflict-free requirements. Finally, we present studies conducted at the Graz University of Technology to evaluate the applicability of the proposed recommendation technologies.

On Requirements Representation and Reasoning using Answer Set Programming
Julian Padget, Emad Eldeen Elakehal, Ken Satoh, and Fuyuki Ishikawa
(University of Bath, UK; National Institute of Informatics, Japan; Sokendai, Japan)
We describe an approach to the representation of requirements using answer set programming and how this leads to a vision for the role of artificial intelligence techniques in software engineering with a particular focus on adaptive business systems. We outline how the approach has developed over several years through a combination of commercial software development and artificial intelligence research, resulting in: (i) a metamodel that incorporates the notion of runtime requirements, (ii) a formal language for their representation and its supporting computational model (InstAL), and (iii) a software architecture that enables monitoring of distributed systems. The metamodel is the result of several years experience in the development of business systems for e-tailing, while InstAL and the runtime monitor is on-going research to support the specification, verification and application of normative frameworks in distributed intelligent systems. Our approach derives from the view that in order to build agile systems, the components need to be structured more like software that controls robots, in that it is designed to be relatively resilient in the face of a non-deterministic, dynamic, complex environment about which there is incomplete information. Thus, degrees of autonomy become a strength and an opportunity, but must somehow be constrained by informing these autonomous components what should be done in a certain situation or what system state ought to be achieved through norms as expressions of requirements. Because such a system made up of autonomous components is potentially behaviourally complex and not just complicated, it becomes essential to monitor both whether norms/requirements are being fulfilled and if not why not. Finally, because control over the system can be expressed through requirements in the form of data that can be changed, a route is opened to adjustment and dynamic re-direction of running systems.

A Case Study of Applying Data Mining to Sensor Data for Contextual Requirements Analysis
Angela Rook, Alessia Knauss, Daniela Damian, and Alex Thomo
(University of Victoria, Canada)
Determining the context situations specific to contextual requirements is challenging, particularly for environments that are largely unobservable by system designers (e.g., dangerous system contexts of use and mobile applications). In this paper, we describe the application of data mining techniques in a case study of identifying contextual requirements for a context-aware mobile application to be used by a team of four long-distance rowers. The context of use for this application was dangerous and isolated, making it unobservable by the developers. The context situations for five mobile application requirements were defined by using a data mining algorithm applied to historical sensor data passively collected by the users while they crossed the Atlantic Ocean in a rowboat. The performance of the resulting classifiers is analyzed over time with promising results demonstrating that the data mining approach is feasible with implications for requirements engineering, context-aware mobile applications, and group-context-aware mobile applications.

Using AI to Model Quality Attribute Tradeoffs
Neil A. Ernst and Ian Gorton
(SEI, USA)
Many AI techniques have been applied to goal-oriented requirements engineering. However, such techniques have focused mostly on the intellectual challenge and ignored the engineering challenge of RE at scale. We discuss some of these existing approaches. We then introduce some early work that aims to add contextual quality attribute information to leverage the power of AI techniques and tools with real-world engineering. We believe this will address some of the acquisition and context problems that have plagued AI in RE.

Applying Knowledge Representation and Reasoning to (Simple) Goal Models
Alexander Borgida, Jennifer Horkoff, and John Mylopoulos
(Rutgers University, USA; University of Trento, Italy)
We consider simple i*-style goal models with influence (contribution) links and AND/OR refinement (decomposition), and formalize them by translation into three standard logics that are actively studied in AI Knowledge Representation and Reasoning (KR&R): propositional logic, FOL and description logics (the first formalization is well known). In each case, this provides a semantics for the notation, on which we can base the definition of forward ("what if?") and backward ("how is this achievable?") reasoning, of interest to requirements engineers. We consider the manner in which AI KR&R research provides off-the-shelf algorithms that can be used to solve these tasks. We compare the representations by reporting known worst-case complexity results for the reasoning, as well as other criteria such as size/understandability of axiomatization, and ease of extension of modeling language.

AIRE 2014 – Proceedings

2014 IEEE 1st International Workshop on Artificial Intelligence for Requirements Engineering (AIRE)