KELVIN: Extracting Knowledge from Large Text Collections