Page 138 -

AGENT-ORIENTED METHODS AND METHOD ENGINEERING 123
the internal view of a single agent, and applies those concepts to the external view in terms of
problem modeling as part of requirements engineering. It also relies heavily on the i* framework
of Yu (1995) for concepts and notation.
In summary, there is a tendency to reuse significant portions of object-oriented methodological
approaches, supplementing them with a new focus on organizations, social interactions, proactiv-
ity, and roles. There is still discussion about the extent to which UML can be useful. Several AO
methodologies use existing UML as a pragmatic option or, often, AUML diagrams but, at the same
time, find deficiencies for which they supply new diagrammatic representations.

COMPARING AGENT-ORIENTED METHODOLOGIES

Several authors have made direct comparisons of these (and other) AO methodologies. Cernuzzi
and Rossi (2002) proposed a framework containing a set of internal attributes (autonomy, reactiv-
ity, proactiveness, and mental notions), a set of interaction attributes (social ability, interaction
with the environment, multiple control, multiple interests, and subsystems interaction), and four
other requirements (modularity, abstraction, a system view, and communication support). They
used this framework in a case study to evaluate a BDI-focused methodology (Kinny, Georgeff,
and Rao, 1996; variously referred to as AAII or BDIM) and MAS-CommonKADS (Iglesias et
al., 1998) both qualitatively and, with an appropriate set of metrics, quantitatively. This study and
other comparative evaluations of both AO and OO methodologies were used as input to the frame-
work proposals of Dam and Winikoff (2004), who proposed four categories: concepts, modeling
language, process, and pragmatics. Their contribution is that the evaluation was done not only by
the authors but by surveying a set of students who had used the case study methodologies (MaSE,
Prometheus, and Tropos) on a design problem of a mobile travel planner. The same four categories
were used by Sturm and Shehory (2004) and used to evaluate Gaia (as a single example) using
a seven-point quantitative metric scale. The framework of Tran, Low, and Williams (2003) also
has four categories, but these are said to be process-related (fifteen criteria), technique-related
(five criteria), model-related (twenty-three criteria), and other supportive features (eight crite-
ria). The framework was applied by Tran, Low, and Williams (2004) to five well-referenced AO
methodologies—namely, MaSE, Gaia, BDIM, Prometheus, and MAS-CommonKADS. Different
ordinal scales are used for the several-criterion sets. A more extensive set of results (the evaluation
of ten AOSE methodologies) is shown in Table 8.1 (page 125), and a statistical evaluation of nine
(seven overlapping and two new) is presented by Elamy and Far (2006).

AN ALTERNATIVE TO A SINGLE AGENT-ORIENTED
METHODOLOGY: SITUATIONAL METHOD ENGINEERING

Using a single fixed AO methodology (e.g., any of those outlined in the second section of the
chapter) works well if that methodology and the project demands are in good alignment. This
is rarely the case. More likely is the situation when the user of the XYZ methodology finds he/
she needs something different or something additional. Improvisation can follow, but this runs
the risk of introducing incompatibilities and inconsistencies—as well as incurring high effort-
overhead costs.
An alternative is to use the well-founded tenets of “situational method engineering” or SME
(e.g., Brinkkemper, 1996; Kumar and Welke, 1992; Ter Hofstede and Verhoef, 1997). SME pro-
vides a flexible way of constructing a methodology from a set of method fragments in such a way
that the process requirements of the individual project are fully satisfied and the methodology is