Robust Requirements Analysis in Complex Systems through Machine Learning - Trustworthy Eternal Systems via Evolving, Software Data and Knowledge - page 44

Robust Requirements Analysis in Complex
Systems through Machine Learning

Francesco Garzoli 1 , Danilo Croce 1 , Manuela Nardini 2 ,
Francesco Ciambra 2 , and Roberto Basili 1

1 University of Roma, Tor Vergata, Italy
2 Finmeccanica SELEX Sistemi Integrati, Italy

Abstract. Requirement Analysis (RA) is a relevant application for Se-
mantic Technologies focused on the extraction and exploitation of knowl-
edge derived from technical documents. Language processing technologies
are useful for the automatic extraction of concepts as well as norms (e.g.
constraints on the use of devices) that play a key role in knowledge acquisi-
tion and design processes. A distributional method to train a kernel-based
learning algorithm is here proposed, as a cost-effective approach for the
validation stage in RA of Complex Systems, i.e. Naval Combat Systems.
The targeted application of Requirement Identification and Information
Extraction techniques is here discussed in the realm of robust search pro-
cesses that allows to suitably locate software functionalities within large
collections of requirements written in natural language.

1 Introduction

The objectives of Requirements Engineering (RE) include at least the identifica-
tion of the goals to be achieved by a target system, the operationalization of such
goals into services and constraints, and the assignment of responsibilities for the
resulting requirements to agents such as humans, devices and software. Different
processes are involved in RE, such as domain analysis, elicitation, specification,
assessment, negotiation, documentation and evolution. Sources of information
are mostly expressed in natural language and require manual analysis: getting
high quality requirements is dicult, critical and costly. During a novel system
design, all of these phases must be performed, and generally they are carried out
without any reuse of old analysis performed over previous systems.
In this scenario, search systems are usually required to help analysts to locate
and access the information stored in documents whereas key-word based search
may not be sucient. As an example, when searching for “ attack scenario ”doc-
uments containing an expression such as (the verb) “ assail ” may not be found as
for the mismatch between the query and the text. A more semantic aware pro-
cess is needed to increase the benefits of automatic search in RE. Furthermore
the validation of design choices could be automatized, e.g. checking the consis-
tency of requirement pre-conditions. However, translating user requirements and
problem domain described in natural language into the consistent modeling of
the target application is still challenging. According to [1], “ We are not really