Rule-Based Extraction of Tuple-Based Service Demand from Natural Language-Based Software Requirement for Automated Service Composition.

In recent years, automated service composition (ASC) has become popular and prevalent in software engineering and services computing. Nevertheless, although there are many existing research works on ASC, a number of problems and issues in this research area remain to be addressed. Based on our observations, one of the major problems and insufficiencies of current ASC approaches is that many of them accept and consider only tuple-based service requirements, whereas it is unrealistic to assume that all users of an ASC approach can express their service demands in this manner. To most service and software stakeholders, the most natural and best way to describe their requirements is to express them in natural language. Due to this difference in formations, most existing ASC approaches are useless or difficult to use for ordinary users. In this paper, we attempt to bridge this gap between the conventional descriptions of user requirements and the acceptable inputs of current ASC approaches by proposing and implementing a new approach. This approach uses natural language processing (NLP) technologies to obtain the required information from natural language-based requirement descriptions to generate tuple-based service demands for ASC approaches. Based on the part-of-speech and dependency information parsed and provided by an adopted NLP processor, we define a set of rules and preprocessing methods to extract the intended service requirement elements. Finally, with a real-world software description dataset, the developed system is fully tested and demonstrates accurate extraction performance.