Implementation-Centric Classification of Business Rules from Documents

Abstract

In large multi-site multi-vendor projects, studying requirement documents to understand the problem domain and inferring possible solution to the posed problem is an important activity in Requirements Engineering. The process of reading User require-ments Specification (URS) to create Software Requirement Speci-fication (SRS) is a knowledge intensive activity that precedes sev-eral other important Software Engineering (SE) activities such as design and test plans. Automated Interpretation of the URS in terms of implementation-specific knowledge elements for software engineers' consumption has been reported in the past. The aim of such an interpretation is to reduce the effort associated with a manual extraction of knowledge elements and subsequently, their "translation" into primitives understood by those who must build the intended software. In this paper, we present a deep learning model for an implementation-centric classification of one such knowledge element, namely, business rules. We discuss an approach based on a Bidirectional Long Short Term Memory Network (BiLSTM) to capture the context information for each word, followed by an attention model to aggregate useful infor-mation from these words to get to the final classification. Our model adopts an end-to-end architecture that does not rely on any handcrafted features.