Database Reference
In-Depth Information
Figure 1.3:
Google square
comedy movies
(left figure) as of November 2010. By clicking on the language
and director fields of
The Mask
, pop-ups are shown with possible languages and directors of this movie
(right figure).
“Release Date”, etc.). For each attribute value, Google Squared displays only the value with the
highest degree of confidence. However, if the user clicks on that value, then alternative choices are
shown. For example, the most likely director of “The Mask” is Chuck Russell, but Google Squared
has found other possible values (“John R. Dilworth,” etc.) with lower confidence, and the user can see
them by clicking on the director value (as shown in the figure). Similarly, for the language, English
is the most likely, but a few other possible values exists.
In attribute-level uncertainty, the value of an attribute is a random variable that can take one
of several possible outcomes. For example, the Director attribute of “The Mask” can be “Chuck
Russell”, “John R. Dilworth”, etc. Assuming each movie has only one director, these choices are
mutually exclusive
probabilistic events. On the other hand, the choices of different attribute values
are considered to be independent. For example, we assume that the Director attribute and the
Language attribute are independent, and, similarly, we assume that Director attributes of different
movies are also independent.
The power of external data sources such as NELL or Google Squared comes from merging
them and further integrating them with other offline data sources, using relational queries. For
instance, one can ask for birthplaces of directors of comedy movies with a budget of over $20M by
joining the square for
comedy movies
(where we can ask for the budget) with some other external
dataset like NELL (to obtain the directors' birthplaces). To do this, one needs a system that supports
complex SQL queries over databases with uncertain data. Such a system is, of course, a probabilistic
database system.
