We used the domain factors presented in the preceding section as basis for
requirements analysis and identification of alternative solutions. In general, all the
seven factors present their own important constraints. Here, however, we focus
only on the issues pertaining to expertise model since this factor constrains the
other operations like expertise indicator extraction, matching operations and
query mechanisms. Besides, as we will show later, we aim at an architecture that
is flexible enough to allow incorporation of various implementations of the other
domain factors.
As we saw before, expert models can be query-time generated, personal agent
based or associations to centralized knowledge model. Each of these approaches
have their strengths and weaknesses. As pointed out by Mattox, et al (1999),
using query-time generated expertise models have the advantage of avoiding
maintaining information internally by the expert finding system and permit
operation in real-time using the most recent information available to locate
experts. However, the authors reported that their system suffers from critical
speed inefficiency. We also have witnessed this shortcoming from the similar
expert finding system we developed for our research department. We feel that,
given the availability of search robots that can routinely gather sources of
expertise indicators, the advantage of real-time operation of query-time generated
models is less attractive. Besides, as we will demonstrate shortly in this section,
this approach also fails to meet other important requirements.
The tradeoffs between the other two approaches (i.e. association to central
knowledge model and distributed personalized agent based model) are more
difficult to weigh. The advantages of using distributed personal agents include the
following:
everyone’s expertise is locally determined hence possibility to extract
expertise indicators from personal information sources;
complete owner control;
the expertise model of the seeker can easily be used in finding an expert
(hence matching of expertise level possible).
However, this approach requires each agent to maintain its own domain model
and matching engine in addition to expertise models (eg. Vivacqua, 1999).
Furthermore, this approach, like the query-time generation method fail in meeting
the following important requirements:
(1) Visualization and browsing capability
In addition to basic search capabilities, presenting each expert as an element
within a broad and structured expertise framework is necessary to permit the
efficient identification and selection of experts. Moreover, a user may want to (1)
first overview the areas/types and structure/relationships of expertise available in
an organization, or (2) browse the list of experts at various hierarchies of
knowledge generality. Browsing can also be used in interactive refining of
queries (for eg. using a query formulation assistant agent).
(2) Support for Analysis (Expertise Selection)
As has been pointed out by McDonnald and Ackerman (1998) and discussed in
the preceding, expert finding systems should support expert selection in addition
to expert identification. Expert selection is mainly done by the user and the
system can augment this process by providing various analysis capabilities. For
example, the following features may considerably enhance selection of the right
expert: