TY - JOUR AU - AB - (1) London is the capital of the U.K. We present a Few-Shot Relation Classifica- (A) capital of (2) Washington is the capital of the U.S.A. tion Dataset (FewRel), consisting of 70; 000 (1) Newton served as the president of the sentences on 100 relations derived from (B) member of Royal Society. Wikipedia and annotated by crowdworkers. (2) Leibniz was a member of the Prussian Academy of Sciences. The relation of each sentence is first rec- (1) Samuel Langhorne Clemens, better ognized by distant supervision methods, and known by his pen name Mark Twain, was then filtered by crowdworkers. We adapt the (C) birth name an American writer. most recent state-of-the-art few-shot learning (2) Alexei Maximovich Peshkov, primarily known as Maxim Gorky, was a Russian and methods for relation classification and conduct Soviet writer. thorough evaluation of these methods. Empir- Test Instance ical results show that even the most competi- tive few-shot learning models struggle on this Euler was elected a foreign member of the (A) or (B) or (C) Royal Swedish Academy of Sciences. task, especially as compared with humans. We also show that a range of different reasoning Table 1: An example for a 3 way 2 shot TI - FewRel: A Large-Scale Supervised Few-Shot Relation Classification Dataset with State-of-the-Art Evaluation JF - Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing DO - 10.18653/v1/d18-1514 DA - 2018-01-01 UR - https://www.deepdyve.com/lp/unpaywall/fewrel-a-large-scale-supervised-few-shot-relation-classification-TOshQg7vZn DP - DeepDyve ER -