EA-Miner: Towards Automation in Aspect-Oriented Requirements Engineering - Transactions on Aspect-Oriented Software Development - page 28

time spent to process the document and produce the initial version (with all
viewpoints included) is not high even when one considers extremely large files. 4
However, the main problem is not the size of the files. The main issue is how long
it can take for the engineer to prune this large initial set of candidate viewpoints and
arrive at a final list of the relevant ones. This is where the filters and guidelines are
essential as explained in the previous sections. The tool offers the users the possibility
of shrinking the candidate list for the N most relevant concepts based on statistical
information of relative and absolute frequency. As shown in the examples above, it is
likely that the most important concepts are listed first (e.g., with a threshold of 20
recall was high for both systems). Therefore, in the case of large documents these
filters are essential for saving time.
On the other hand, one problem with using filters is that some important concepts
might be left out of the list. However, by using the tool, the engineer also has the
option to view the complete list if s/he wants. Moreover, other filters such as selecting
only the viewpoints that appear as subjects in the requirements can help to prune the
list of pertinent candidates. As mentioned before, the goal of the tool is to support and
not to replace the requirements engineer. In addition, doing a manual analysis does
not necessarily guarantee completeness (and is obviously not scalable in terms of
time) with respect to concepts identification as the user can skip requirements and
simply ignore important concepts due to lack of attention (see for example the next
case study in Sect. 5.2 where developers missed relevant concerns).
5.2 Toll System Industrial Case Study at Siemens
The case study described in this section is related to a family of toll systems. Part of
the real requirements is used as input to a prototypical implementation of the toll
system family using AO technologies developed by Siemens AG. The documentation
available for this system encompasses several requirements documents describing
different features of the system and written by several Siemens' developers.
In this study we analyzed a fraction of the original requirements documents
focused on communication. The size of all analyzed documents is around 4,000 words
and they represent an interesting exercise for our analysis as they vary in structure and
terminology and they were written by multiple authors. The requirements describe a
real system that is a product line of toll systems that are used in several European
countries containing several peculiarities each with its own country-specific
variations.
Our goal with this case study was not to undertake a comparison like the previous
one but to analyse how well does EA-Miner perform in an industrial setting. We
discuss our findings in the next section, however we are unable to give much details
about the requirements themselves as there are confidentiality issues involved.
5.2.1 Findings
Our task in this case study was to use the set of available documents as input to EA-
Miner to perform an aspect-oriented requirements analysis on them. The idea was to
4 In a recent test with a file of 56,031 words (237 pages), the time taken for identification was
around 20 min.