Information Technology Reference
In-Depth Information
Fig. 7. The rewriting system in practice [11]
original CC 'Blood sugar increase' into the concept P:HYPERGLYCEMIA via the
recursive application of production rules in Stage 1. Extra white space is removed and
capitalization is carried out in stage 0. Stage 2 is category assignment. Further details
of the three-stage algorithm are presented in [9] and [11].
The current system version can increase its semantic analysis capability through an
on-line learning process supported by morphological analysis. This enables through-
use evolution of the system's ability to categorize chief complaints, i.e. to increase its
power of understanding. The server side architecture of the systems consists of a CC
interaction engine (IE), a CC analysis engine (AE), and a CC repository engine (RE).
[9] Functional details of these components and the entire system are described in [9].
Figure 8 shows the entry of a new CC with the interaction engine.
Figure 9 illustrates the collaborative learning process of the chief complaint agent.
The performance of the chief complaint agent in distilling categorical information
from free-text CC's was evaluated in [8-11]. Table 1 shows the results of the evalua-
tion in [9] where the system was trained with the CC's from one ED and then tested
with the CC's of this and four other ED's. Complete normalization means that all
free-text terms in the CC were matched with a category and correct normalization
refers to whether these normalizations were correct. Not surprisingly, emergency
department A, the same from which the training data originated, showed best nor-
malization performance. Overall, the results show that the approach worked reasona-
bly well after extensive initial training. Further evaluations of the system are
described in [8], [10] and [11].
Fig. 8. Assisted entry of chief complaints [9]
 
Search WWH ::




Custom Search