Information Technology Reference
In-Depth Information
•
POS Possible - The number of fills in the key which contribute to the final score,
•
ACT Actual - The number of fills in the response
the following values are calculated:
ACT = COR + INC + PAR + SPU .
(
1
)
POS = COR + INC + PAR + MIS .
(
2
)
COR
+
0
*
PAR
(
3
)
REC
=
.
POS
(
4
)
COR
+
0
*
PAR
PRE
=
.
ACT
van Rijsbergen's F-measure (5) is used to combine recall and precision measures into
one measure. The formula for F is
(
)
(
5
)
β
2
+
1
*
PRE
*
REC
F
=
(
)
.
2
β
*
PRE
+
REC
The definitions for precision and recall found in the literature may differ slightly from
those given above. Therefore it is recommended to be careful when applying these
scores for comparing IE systems.
The Problem when Applying IE to the Biological Domain.
The traditional IE tasks
as specified in the MUCs are concerned with newspaper articles. Compared to these
articles the structure of a sentence in the biological domain tends to be more
complicated. Also the NE task is quite more challenging because of the confusing
nomenclature of genetics. One gene often has more than one name or consists of a
compound noun phrase. Additionally these noun phrases do not follow strict
orthographical rules. For instance 'NFKB2' or 'nuclear factor of kappa light
polypeptide gene enhancer in B-cells 2 (p49/p100)'or 'LYT-10' are synonyms for the
same gene. And 'NF-kappa-b', 'nf-kappa-b' or 'NF Kappa B' presumably mean one
and the same thing. Moreover, researchers studying different organisms have created
quite different naming traditions. The Drosophila geneticists use such interesting
names like “hedgehog” or “lost in space”, while other communities name the genes
after the molecular function of the protein they encode. “Biologists would rather share
their toothbrush than share a genes name,” said Michael Ashburner, head of the
European Bioinformatics Institute in an interesting article in
Nature
about this subject
[24].
All these problems call for sophisticated methods for name identification
[25],
[26], [27].
