Information Technology Reference
In-Depth Information
Moreover, it is also important to mention work related to the issue of cataloguing
and reusing non-functional requirements (NFR) knowledge in case tools as we use
this in our lexicon-based approach for identifying NFRs. [35] proposes ways for
structuring and cataloguing NFR framework [27] knowledge with the intention of
reusing and querying this knowledge. [34] describes in detail an approach for eliciting
and building NFR models as well as integrating the non-functional perspective with
the functional one in scenarios as well as in class diagrams. This approach utilizes a
lexicon (language extended lexicon—LEL) that contains information about the
entities of the system as well as the non-functional requirements. We believe that our
NLP-based approach could benefit [34] by enabling to automatically identify the
entities and NFRs and suggest a list of candidates that could be added by the user in
the LEL lexicon. We also believe that the reusable NFR knowledge bases described in
[34, 35] can benefit our approach with regards to the automated lexicon update
process as explained in Sect. 4.1.3 (See Fig. 9).
8 Conclusions
Aspect-oriented requirements engineering (AORE) techniques do not offer adequate
tool support for costly tasks in early stages of requirements engineering such as
identifying model concepts from different requirements documents as well as
mapping these concepts to AORE models.
This paper described the EA-Miner tool and how it offers NLP-based support for
automating AORE activities. We have presented some mining analysis approaches
based on NLP techniques (e.g., identification of viewpoints, use cases, early aspects)
and showed how the tool implements them in the context of one example (toll
system). EA-Miner utilizes a precise NLP processor (Sect. 3) that offers several NLP
techniques such as part-of-speech and semantic tagging and frequency of occurrence
of words that are key to EA-Miner's mining features.
Our evaluation data shows encouraging results that suggest that EA-Miner is a
much more time-effective approach when compared to the usual manual analysis
conducted during AORE. Accuracy data on precision and recall also show that the
models created by the tool do not suffer from poor quality and can even be superior
(when combined with the experience of the engineer) than a pure manual analysis.
Moreover, an industrial case study conducted at Siemens AG investigated the
effectiveness of EA-Miner to analyse real-world requirements written by different
developers with various writing styles and vocabulary. Our analysis found relevant
concerns that were previously missed by Siemens' developers in their aspect-oriented
prototype implementation, but recognised as relevant by them. It also revealed some
challenges imposed by the structure of the documentation and the different use of
vocabulary terms hence providing new paths to explore and improve the tool in the
future which are better pre-processing support, “domain synonym” identification and
detection of poorly written requirements.
In addition to these paths other future work will focus on further investigating
mining analysis considering other AORE techniques such as Scenario Modelling [18]
and symmetric approaches such as [25]. Moreover, we plan to conduct evaluation
with diverse types of requirements-related documents such as interview transcripts
and legacy specifications.
Search WWH ::
Custom Search