The question that data scientists at Quorum, a political analytics firm, sought to answer was this: Can computers use a similar process to come to the same conclusion? Could they teach a computer to predict political party from speech?
Mining the text of House and Senate floor speeches in the Congressional record, Quorum cofounder Jonathan Marks and his team wanted to see if they could accurately predict which congressional members belong to which party.
“We gave the computer a large amount of text, which had been fed by Republicans and Democrats,” Marks explains. “And then we asked it to identify patterns in the way that Democrats and Republicans talk that make them different.”
The program searched for the favorite words used by each party, but it also searched for the words that were uniquely favored by each party. Each party may say “America” often. But Republicans are much more likely to say “bureaucrats,” for example.