I too thought it would be more specific. Nowhere does it mention sample selection bias, the scourge of most "big" datasets I've been offered. Quantity can't always compensate for lack of quality.