32
for (
i, i
0
)
∈ J →
combine the two terms as a search engine query:
i ,
i
0
;
collect the sentences in the search results containing both
i and
i
0
;
replace instances by placeholders for the corresponding classes ;
store the text fragments
S in a set
P ;
rof
initialize
c(
S) and
n(
S) to 0 for all
S ∈ P ;
for (
i, i
0
)
∈ J →
query
i ;
collect all sentences in the search results that contain both instances ;
for
S ∈ P →
c(
S) =
c(
S) + number of occurrences of
S with
i
0
;
n(
S) =
n(
S) + the total number of occurrences of
S ;
rof
rof
for
S ∈ P →
compute precision
f
pr
(
S) =
c(
S)
n(
S)
;
rof
select the most precise ones using the scores
f
pr
such that
c(
S)
≥ 5;
Table 3.1. Sketch of the pattern identification algorithm proposed by Ravichan-
dran and Hovy.
(
{c
Dostları ilə paylaş: