||Our input was the raw
database, and our goal was to see whether we could automatically reconstruct
the commonsense knowledge that the TB health care professional has. We discovered many useful patterns, mostly
ones that are known but also some new ones.
For example, gender is correlated with HIV status. But more surprising (and more novel), it is
also correlated with the type of tuberculosis. And there are also interactions between
objects. Contacts of foreign-born
patients are more likely to have TB, but are also more likely to obtain
treatment, because of the tightness of the social structure.