Information Technology Reference
In-Depth Information
This higher-level architectural extension is objective of our future work. In the
dimension of process extension
of EA-Miner the goal is to catalogue other guidelines,
similar to those in Sect. 4.3, for other AORE approaches.
7 Related Work
The use of NLP techniques in automation of some tasks in RE has been discussed in
[6, 9, 10, 12, 13, 41]. They also focus on using NLP at the requirements level to
identify concepts and build models.
The Color-X approach [6] offers natural language processing to semi-automatically
parse the input (e.g., a document in natural language) into an intermediate formal
model based on the common representation language (CPL) [6]. The transformation is
not a fully automated step and requires manual structuring from the user (some
guidelines are offered) and some words need to be entered in a lexicon. After the CPL
representation is built two models can be generated: One static object model, called
CSOM, similar to a UML class diagram, and an event-based model, called CEM,
similar to a state machine.
The Circe [9, 41] environment provides a tool that process natural language text as
input and generates an intermediate model based on some rules called model, action
and substitution rules. After the intermediate model is created, different analysis
models can be generated such as entity relationship (E—R) models, data flow
diagrams (DFDs) or OO models. Similar to what happens with Color-X, the user has
to input some elements in the glossary along with some tags that refer to the
semantics of the rules. This characteristic is a bit different from our approach since
our NLP processor does not require any input from the user (apart from the natural
language documents). Nor does it require that the user has detailed knowledge on how
it works.
The Abstfinder [10] approach offers automation for the identification of
abstractions in requirements documents described in natural language. Abstractions
are considered to be relevant concepts that can be understood without having to know
their details such as “booking” and “flight” in a reservation system. The process of
concept identification is based on pattern matching between sentences in the text. The
output of the tool is a list of abstractions and the knowledge on what to do with this
information is left to the requirements engineer.
The NL-OOPS approach [12, 13] is based on a natural language processor, called
LOLITA, and utilizes semantic networks produced by LOLITA to automate the
production of object-oriented conceptual models (e.g., classes and their methods,
attributes and relationships). The semantic networks are updated every time a file is
provided as input to the NLP processor and represent concepts organized into
hierarchies (entity nodes) as well as their relationships (event nodes). The NL-OOPS
tool then applies some heuristics to filter out and classify the information contained in
the semantic networks in order to produce the object-oriented conceptual models.
Moreover, similar to what we have done in our evaluation this approach also utilized
precision and recall measures to compare the quality of models produced by engineers
using and not using the tool. LOLITA does not offer either POS or semantic tagging
what could also be very helpful for identifying class candidates (e.g., nouns) or
Search WWH ::
Custom Search