Learning action models with quantified conditional effects for software requirement specification

Abstract

Software requirement specification (SRS) is an important step in software engineering. Extracting requirement specification from an application field is a difficult task. In this paper, we consider software requirement as a problem to be solved by intelligent planning. To do this, one of the difficult things is how to represent the domain, since the software requirement has a feature of changeability. Thus, we divide the work into two tasks: the first one is to describe an incomplete domain of software requirement using PDDL(Planning Domains Definition Language) [7]; the second one is to complete the domain by learning from plan samples which is extracted from business processes. We modify the tool of [9] to learn action models with quantified conditional effects, which is the second task. In this way, people only need to do the first task and extract plan samples, which means the efforts of human beings are saved. At the end of the paper, we give our experiment result to show the efficiency of our method. © 2008 Springer-Verlag Berlin Heidelberg.