We try to get data from the most definitive, authoritative, sources. Often the web is a good place to start in helping us identify those sources. But then we tend to go to them directly. Identifying the best sources is just the first step, though. Then we have to curate the data, organizing it, correlating it, validating it. It always ends up needing lots of automated work, with statistical analysis, visualization, etc. Together with input from actual human experts in each particular domain.