The Tradeoffs Between Open and Traditional Relation Extraction is a paper about extraction of relations between entities on a massive scale. The system is based on a self-training unsupervised approach derived by the Conditional Random Fields (CRF) theory. A set of training examples are extracted by a training corpus by means of hand-crafted defined parsing rules. The training examples are then used to train a linear chain of CRF.
Results are pretty impressive both in term of precision and recall.