Using CBR as Design Methodology for Developing Adaptable Decision Support Systems - Efficient Decision Support Systems: Practice and Challenges from Current to Future - page 140

identifies those e-mails that best represent the new incoming message (left upper quadrant
in Figure 10), only taking into account the set of messages with the highest number of terms
in common. Each one of the previous selected messages contributes with one vote to the
final class (left bottom quadrant in Figure 10). The revision stage is only carried out when
the system proposes an e-mail as spam. For this purpose, SpamHunting uses previous
encoded header meta-rules (right bottom quadrant in Figure 10). Every time the user checks
his mailbox and provides feedback to the system about a previous e-mail classification,
SpamHunting stores a new instance (or modifies an existing one) in the e-mail base for
future use (right upper quadrant in Figure 10).
The retrieval stage is carried out using our enhanced instance retrieval network (EIRN)
model. The EIRN network facilitates the indexation of instances and the selection of those
that are most similar to the instance-message. The reuse of similar e-mails is carried out by
means of the utilization of a unanimous voting mechanism, which generates an initial
solution by creating a model with the retrieved instances. In the revision stage the system
employs general knowledge in the form of meta-rules that are extracted from the e-mail
headers. Finally, the retain (learning) stage is carried out whenever the system classifies an
incoming e-mail, updating the knowledge structure of the whole system. The proposed
system also takes into account the feedback of the user when it receives an incorrectly
classified e-mail (dotted arrow from the user computer to the e-mail base in Figure 10).
Fig. 11. Screenshot of the EIRN viewer comprising the SpamHunting system
One of the most relevant features of our EIRN structure is the possibility of generating
different views of the underlying indexed knowledge. Figure 11 shows a screenshot of the
software developed for taking advantage of this capacity. As we can observe from Figure 11,