
199Linked Data in Enterprise Integration
Results and Discussion
Table 5.5 shows the union of the rules learned by the batch version of CaRLA
in all ve runs. Note that the computation of a rule set lasted under 0.5 s
even for the largest data set, that is, Movies. The columns P
n
give the prob-
ability of nding a rule for a training set of size n in our experiments. R
2
is not reported because it remained empty in all setups. Our results show
that in all cases, CaRLA converges quickly and learns rules that are equiva-
lent to those utilized by the LATC experts with a sample set of 5 pairs. Note
that for each rule of the form <“@en” → y> with y