Robust Requirements Analysis in Complex Systems through Machine Learning - Trustworthy Eternal Systems via Evolving, Software Data and Knowledge - page 49

%

%

!
"

$!$

$!$

Sentences Containing
Requirements Requirements

Analyst

#
!$

#
!$

Presentation Application Learning

Fig. 1. Requisite Analysis System Architecture

- Domain specific Lexicon : it contains the specific domain dictionaries provid-
ing lexical information about the application domain, e.g. involved entities
and acronyms.
- Domain ontology : it provides an ontological model of the application do-
main, as well as an abstraction of the requirements (i.e. the template for the
Information Extraction activity). Moreover it provides the relations among
different requirements, e.g. dependency rules among pre-conditions and post-
conditions that enable the reasoning.
- Template Definition : it represents the repository of different templates in-
volved in the IE activity, that are domain specific (as the ontology), but
possibly more specific than the concepts or relations in the domain ontology.

According to our machine learning perspective, each module performs the corre-
sponding task according to a model of the domain that has been automatically
previously acquired from real data. These are requirement documents that have
been previously annotated by the analysts, with the same information the IE
system is expected to precisely detect in future texts. The general architecture is
thus divided in different main blocks to distinguish models directly employed in
the (on-line, i.e. interactive) Requirement Analysis Application workflow from
the ones employed in the ( off-line )Learningworkflow.Inthe Application Block
the following modules process requirement as follows:
- Requirement Identification Module : This module performs the analysis of
documents that are enriched with linguistic information in order to suit-
ably locate sentences containing concepts (and relations) of interest in the
requirements analysis domain.
- Information Extraction (IE) System : Once a specific requisite is found, the
extraction of its relevant information is carried out as a slot-filling process
over the existing templates. Once a template is filled, it is made available
(i.e. it populates) the Requirement Repository for the analysis.
In a machine learning perspective, each module performs the corresponding
task according to a model of the domain that must be automatically acquired
from annotated data. In this view, the second block in the architecture of Figure
1, i.e. the Learning block, is dedicated to the acquisition of the individual IE
models. Finally, the Presentation block is responsible for the interaction with