[Modeling] Comments on interaction diagram modeling doc

James Odell email@jamesodell.com
Wed, 19 Mar 2003 16:57:44 -0500


On 3/19/03 8:06 AM, "Lin Padgham" indited:

> What seems to be missing here, which is what we use most, is to give
> the agent type - not the instance name of an agent, but a type of
> agent. So we might have a "shop assistant" agent type, with roles that
> include "sales transaction", "customer welcoming", "product queries",
> etc. These roles are deliberately quite small and specific (and relate
> to particular goals and functionalities), whereas things like
> "employee" are more like what we would have as agent types. I don't
> think we have evenr found a need to use specific agent instance
> identifiers, though sometimes we do indicate that there *are* multiple
> instances and separate these out in some way in the diagrams.

Hmmm.  We might have a difference in our definitions of "role", then.  In
UML, the role would be "shop assistant" -- and in order to play that role,
the agents must be able to perform various services, such as  "sales
transaction", "customer welcoming", "product queries", etc.   Having said
that, there is no reason that you could not refer to  "customer welcoming"
as class with a set of interactive behavior that supports such a role --
even if it were a single-behavior role.  In this case, you would declaring
that the shop assistant is an aggregate of other "small and specific" roles
you mention.  Such aggregation is important in role formation.   Does that
work?

Regards,

Jim