Overview - Probabilistic Databases - page 2

Figure 1.1: Facts extracted by NELL from the WWW. The thumbs-up and thumbs-down icons are
intended to solicit users' input for cleaning the data.

from ProducesProduct x, HeadquarteredIn y
where x.Company=y.Company and y.City='san_jose'

This is a join of the two tables in Figure 1.2 , and a fragment of the result is the following:

Product Company P

0.95

personal_computer ibm

0.83

adobe_indesign adobe

0.80

adobe_dreamweaver adobe

The first answer, ( personal_computer , ibm ), is obtained by joining the tuples marked with
X 1 and Y 1 in Figure 1.2 , and its probability is the product of the two probabilities: 0 . 96 · 0 . 99 ≈ 0 . 95.
Similarly, ( adobe_dreamweaver , adobe ) is obtained by joining the tuples X 2 , Y 2 . Distinct tuples in
NELL are considered independent probabilistic events; this is quite a reasonable assumption to make,
for example, the tuple X 1 was extracted from a different document than Y 1 , and, therefore, the two
tuples can be treated as independent probabilistic events. As a consequence, the probabilities of the
answers to our query are computed by multiplying the probabilities of the tuples that contributed to
these answers. This computation can be expressed directly in SQL in a standard relational database:

select distinct x.Product, x.Company, x.P * y.P as P
from ProducesProduct x, HeadquarteredIn y
where x.Company = y.Company and y.City = 'san_jose'
order by P desc