An Automatic Requirements Modeling Method Based On Relation Extraction

Requirements are usually presented as NL-based (Natural Language based)documents. In the conceptual modeling phase, requirements are collected from differentstakeholders and analyzed by requirement engineers. However, the size of therequirments documents can become very large, and the modeling process is quite timeconsuming and resource consuming. In order to solve this problem, much has beenwritten on the processing of NL-based requirements documents to yield conceptualmodels. The main work of this paper is as follows:(1) In this paper, we proposed an approach for identifying and extracting relationsin a range of requirements documents with three steps: Text Analyzation, EntityExtraction and Relation Mapping. If the entities in the relation are quite close to eachother, for example, in the strategic dependency relationship, we will define a set oflinguistic patterns used for identifying relations and propose a matching algorithm ofsemantic automata to extract the relation. As for the two entities in the relationship isnon-local dependency, such as temporal relation between events, we will then extractthe feature vector and adopt SVM algorithm to classify the relation type.(2) Besides, we also provided a set of domain-oriented resources in this approch,including Domain Lexicom, Repair Rule Library and Background Knowledge Library.We adopted a transformation-based error-driven learning algorithm to build the repairrules in Repair Rule Library and computed the sematic relatedness of the concepts in thedomain and extract the interaction relations between every two concepts, which formthe Background Knowledge Library.(3) Based on this approach, we developed a system to automatically generate theSD model of i*framework and the temporal model from Chinese requirementsdocuments. A series of experiments were conducted to evaluate the performance of theautomated requirements analysis system. The results show that the system achieves highrecall with a consistent improvement in precision, which demonstrates the applicabilityof our approach.